diff --git "a/ir_dump/debug-phase0.py" "b/ir_dump/debug-phase0.py" new file mode 100644--- /dev/null +++ "b/ir_dump/debug-phase0.py" @@ -0,0 +1,4327 @@ +# from tvm.script import ir as I +# from tvm.script import tir as T +# from tvm.script import relax as R + +@I.ir_module +class Module: + I.module_attrs({"system_lib_prefix": "gpt_neox_q0f16_"}) + @T.prim_func + def apply_bitmask_inplace(var_logits: T.handle, var_seq_ids: T.handle, var_bitmask: T.handle): + T.func_attr({"target": T.target({"host": {"keys": ["arm_cpu", "cpu"], "kind": "llvm", "mtriple": "aarch64-linux-android", "tag": ""}, "keys": ["opencl", "gpu"], "kind": "opencl", "max_function_args": 128, "max_num_threads": 256, "max_shared_memory_per_block": 16384, "max_threads_per_block": 256, "tag": "", "texture_spatial_limit": 16384, "thread_warp_size": 1}), "tir.is_scheduled": T.bool(True), "tir.noalias": T.bool(True)}) + batch_size, vocab_size = T.int32(is_size_var=True), T.int32(is_size_var=True) + logits = T.match_buffer(var_logits, (batch_size, vocab_size)) + num_seq = T.int32(is_size_var=True) + seq_ids = T.match_buffer(var_seq_ids, (num_seq,), "int32") + bitmask = T.match_buffer(var_bitmask, (batch_size, (vocab_size + 31) // 32), "int32") + # with T.block("root"): + for fused_s_v_0 in T.thread_binding((num_seq * vocab_size + 255) // 256, thread="blockIdx.x"): + for fused_s_v_1 in T.thread_binding(256, thread="threadIdx.x"): + with T.block("block"): + vs = T.axis.spatial(num_seq, (fused_s_v_0 * 256 + fused_s_v_1) // vocab_size) + vv = T.axis.spatial(vocab_size, (fused_s_v_0 * 256 + fused_s_v_1) % vocab_size) + T.where(fused_s_v_0 * 256 + fused_s_v_1 < num_seq * vocab_size) + T.reads(bitmask[seq_ids[vs], vv // 32], seq_ids[vs], logits[seq_ids[vs], vv]) + T.writes(logits[seq_ids[vs], vv]) + logits[seq_ids[vs], vv] = T.if_then_else(T.bitwise_and(T.shift_right(bitmask[seq_ids[vs], vv // 32], vv % 32), 1) == 1, logits[seq_ids[vs], vv], T.float32(-340282346638528859811704183484516925440.0)) + + @T.prim_func + def apply_logit_bias_inplace(var_logits: T.handle, var_pos2seq_id: T.handle, var_token_ids: T.handle, var_logit_bias: T.handle): + T.func_attr({"target": T.target({"host": {"keys": ["arm_cpu", "cpu"], "kind": "llvm", "mtriple": "aarch64-linux-android", "tag": ""}, "keys": ["opencl", "gpu"], "kind": "opencl", "max_function_args": 128, "max_num_threads": 256, "max_shared_memory_per_block": 16384, "max_threads_per_block": 256, "tag": "", "texture_spatial_limit": 16384, "thread_warp_size": 1}), "tir.is_scheduled": T.bool(True), "tir.noalias": T.bool(True)}) + batch_size, vocab_size = T.int32(is_size_var=True), T.int32(is_size_var=True) + logits = T.match_buffer(var_logits, (batch_size, vocab_size)) + num_token = T.int32(is_size_var=True) + pos2seq_id = T.match_buffer(var_pos2seq_id, (num_token,), "int32") + token_ids = T.match_buffer(var_token_ids, (num_token,), "int32") + logit_bias = T.match_buffer(var_logit_bias, (num_token,)) + # with T.block("root"): + for p0 in T.thread_binding((num_token + 255) // 256, thread="blockIdx.x"): + for p1 in T.thread_binding(256, thread="threadIdx.x"): + with T.block("block"): + vp = T.axis.spatial(num_token, p0 * 256 + p1) + T.where(p0 * 256 + p1 < num_token) + T.reads(logits[pos2seq_id[vp], token_ids[vp]], pos2seq_id[vp], token_ids[vp], logit_bias[vp]) + T.writes(logits[pos2seq_id[vp], token_ids[vp]]) + logits[pos2seq_id[vp], token_ids[vp]] = logits[pos2seq_id[vp], token_ids[vp]] + logit_bias[vp] + + @T.prim_func + def apply_penalty_inplace(var_logits: T.handle, var_seq_ids: T.handle, var_pos2seq_id: T.handle, var_token_ids: T.handle, var_token_cnt: T.handle, var_penalties: T.handle): + T.func_attr({"target": T.target({"host": {"keys": ["arm_cpu", "cpu"], "kind": "llvm", "mtriple": "aarch64-linux-android", "tag": ""}, "keys": ["opencl", "gpu"], "kind": "opencl", "max_function_args": 128, "max_num_threads": 256, "max_shared_memory_per_block": 16384, "max_threads_per_block": 256, "tag": "", "texture_spatial_limit": 16384, "thread_warp_size": 1}), "tir.is_scheduled": T.bool(True), "tir.noalias": T.bool(True)}) + batch_size, vocab_size = T.int32(is_size_var=True), T.int32(is_size_var=True) + logits = T.match_buffer(var_logits, (batch_size, vocab_size)) + num_seq = T.int32(is_size_var=True) + seq_ids = T.match_buffer(var_seq_ids, (num_seq,), "int32") + num_token = T.int32(is_size_var=True) + pos2seq_id = T.match_buffer(var_pos2seq_id, (num_token,), "int32") + token_ids = T.match_buffer(var_token_ids, (num_token,), "int32") + token_cnt = T.match_buffer(var_token_cnt, (num_token,), "int32") + penalties = T.match_buffer(var_penalties, (num_seq, 3)) + # with T.block("root"): + for p0 in T.thread_binding((num_token + 255) // 256, thread="blockIdx.x"): + for p1 in T.thread_binding(256, thread="threadIdx.x"): + with T.block("block"): + vp = T.axis.spatial(num_token, p0 * 256 + p1) + T.where(p0 * 256 + p1 < num_token) + T.reads(logits[seq_ids[pos2seq_id[vp]], token_ids[vp]], seq_ids[pos2seq_id[vp]], pos2seq_id[vp], token_ids[vp], penalties[pos2seq_id[vp], 0:3], token_cnt[vp]) + T.writes(logits[seq_ids[pos2seq_id[vp]], token_ids[vp]]) + logits[seq_ids[pos2seq_id[vp]], token_ids[vp]] = logits[seq_ids[pos2seq_id[vp]], token_ids[vp]] - (penalties[pos2seq_id[vp], 0] + T.Cast("float32", token_cnt[vp]) * penalties[pos2seq_id[vp], 1]) + logits[seq_ids[pos2seq_id[vp]], token_ids[vp]] = T.if_then_else(logits[seq_ids[pos2seq_id[vp]], token_ids[vp]] < T.float32(0.0), logits[seq_ids[pos2seq_id[vp]], token_ids[vp]] * penalties[pos2seq_id[vp], 2], logits[seq_ids[pos2seq_id[vp]], token_ids[vp]] / penalties[pos2seq_id[vp], 2]) + + @T.prim_func + def batch_decode_paged_kv(_0: T.int32, Q_handle: T.handle, pages_handle: T.handle, page_table_indptr_handle: T.handle, page_table_values_handle: T.handle, var_length_info: T.handle, k_rope_pos_offset_handle: T.handle, q_rope_position_handle: T.handle, output_handle: T.handle, lse_handle: T.handle, rotary_mode: T.int32, rope_scale: T.float32, rope_theta: T.float32, attn_score_scaling_factor: T.float32): + T.func_attr({"target": T.target({"host": {"keys": ["arm_cpu", "cpu"], "kind": "llvm", "mtriple": "aarch64-linux-android", "tag": ""}, "keys": ["opencl", "gpu"], "kind": "opencl", "max_function_args": 128, "max_num_threads": 256, "max_shared_memory_per_block": 16384, "max_threads_per_block": 256, "tag": "", "texture_spatial_limit": 16384, "thread_warp_size": 1}), "tir.is_scheduled": 1}) + B = T.int32(is_size_var=True) + Q = T.match_buffer(Q_handle, (B, 12, 64), "float16") + max_num_pages = T.int32(is_size_var=True) + pages = T.match_buffer(pages_handle, (max_num_pages, 2, 12, 16, 64), "float16", offset_factor=1) + page_table_indptr = T.match_buffer(page_table_indptr_handle, (B + 1,), "int32", offset_factor=1) + nnz_pages = T.int32(is_size_var=True) + page_table_values = T.match_buffer(page_table_values_handle, (nnz_pages,), "int32", offset_factor=1) + length_info = T.match_buffer(var_length_info, (B,), "int32", offset_factor=1) + k_rope_pos_offset = T.match_buffer(k_rope_pos_offset_handle, (B,), "int32", offset_factor=1) + q_rope_position = T.match_buffer(q_rope_position_handle, (B,), "int32", offset_factor=1) + output = T.match_buffer(output_handle, (B, 12, 64), "float16") + lse = T.match_buffer(lse_handle, (B, 12)) + # with T.block("root"): + sm_scale: T.float32 = T.float32(0.18033688011112042) + for bx in T.thread_binding(B, thread="blockIdx.x"): + for fused_by_bz in T.thread_binding(12, thread="blockIdx.y"): + for ty in T.thread_binding(1, thread="threadIdx.y"): + for tx in T.thread_binding(16, thread="threadIdx.x"): + for tz in T.thread_binding(16, thread="threadIdx.z"): + with T.block("attn"): + T.reads(page_table_indptr[bx:bx + 2], length_info[bx], q_rope_position[bx], Q[bx, fused_by_bz // 12 + ty + fused_by_bz % 12, tx * 4 - 32:tx * 4 - 32 + 68]) + T.writes(output[bx, fused_by_bz % 12 + fused_by_bz // 12 + ty, tx * 4:tx * 4 + 4], lse[bx, fused_by_bz % 12 + fused_by_bz // 12 + ty]) + Q_local = T.alloc_buffer((4,), "float16", scope="local") + kv_chunk_len = T.alloc_buffer((1,), "int32", scope="local") + K_smem = T.alloc_buffer((16, 64), "float16", scope="shared") + V_smem = T.alloc_buffer((16, 64), "float16", scope="shared") + O_allreduce = T.alloc_buffer((16, 1, 64), scope="shared") + md_allreduce = T.alloc_buffer((16, 1, 2), scope="shared") + S_reduce_local = T.alloc_buffer((1,), scope="local") + t0 = T.alloc_buffer((1,), scope="local") + S_local = T.alloc_buffer((1,), scope="local") + QK_local = T.alloc_buffer((4,), scope="local") + V_local = T.alloc_buffer((4,), "float16", scope="local") + m_prev = T.alloc_buffer((1,), scope="local") + d_prev = T.alloc_buffer((1,), scope="local") + other_m = T.alloc_buffer((1,), scope="local") + other_d = T.alloc_buffer((1,), scope="local") + exp_mprev = T.alloc_buffer((1,), scope="local") + exp_otherm = T.alloc_buffer((1,), scope="local") + other_o = T.alloc_buffer((4,), scope="local") + st_m = T.alloc_buffer((1,), scope="local") + st_d = T.alloc_buffer((1,), scope="local") + O_local = T.alloc_buffer((4,), scope="local") + by: T.int32 = fused_by_bz % 12 + bz: T.int32 = fused_by_bz // 12 + batch_idx: T.int32 = bx + cur_page_indptr_begin: T.int32 = page_table_indptr[batch_idx] + cur_page_indptr_end: T.int32 = page_table_indptr[batch_idx + 1] + kv_chunk_len[0] = T.if_then_else(cur_page_indptr_begin != cur_page_indptr_end, (cur_page_indptr_end - cur_page_indptr_begin - 1) * 16 + length_info[batch_idx], 0) + st_m[0] = T.float32(-50000.0) + st_d[0] = T.float32(1.0) + for vec in T.vectorized(4): + O_local[vec] = T.float32(0.0) + for vec in T.vectorized(4): + freq = T.float32() + Q_local[vec] = T.if_then_else(rotary_mode == 1, T.Let(T.Cast("float16", T.cos(freq) * T.Cast("float32", Q[bx, by + bz + ty, tx * 4 + vec]) + T.sin(freq) * T.Cast("float32", T.if_then_else(tx * 4 + vec < 32, Q[bx, by + bz + ty, tx * 4 + vec + 32] * T.float16(-1.0), Q[bx, by + bz + ty, tx * 4 + vec - 32]))), where={freq: T.Cast("float32", q_rope_position[batch_idx]) * rope_scale / T.pow(rope_theta, T.Cast("float32", (tx * 4 + vec) * 2 % 64) / T.float32(64.0))}), Q[bx, by + bz + ty, tx * 4 + vec]) + for iterator in range((kv_chunk_len[0] + 15) // 16): + tile_start_s: T.int32 = tz + ty + tile_start_g: T.int32 = iterator * 16 + tz + ty + for j in range(1): + with T.block("KV_load"): + T.reads() + T.writes() + row_g: T.int32 = tile_start_g + j + if row_g < kv_chunk_len[0]: + seq_offset: T.int32 = row_g + page_no: T.int32 = page_table_values[cur_page_indptr_begin + seq_offset // 16] + page_offset: T.int32 = seq_offset % 16 + for vec in T.vectorized(4): + freq = T.float32() + K_smem[tile_start_s + j, tx * 4 + vec] = T.if_then_else(rotary_mode == 1, T.Let(T.Cast("float16", T.cos(freq) * T.Cast("float32", pages[page_no, 0, by, page_offset, tx * 4 + vec]) + T.sin(freq) * T.Cast("float32", T.if_then_else(tx * 4 + vec < 32, pages[page_no, 0, by, page_offset, tx * 4 + vec + 32] * T.float16(-1.0), pages[page_no, 0, by, page_offset, tx * 4 + vec - 32]))), where={freq: T.Cast("float32", k_rope_pos_offset[batch_idx] + row_g) * rope_scale / T.pow(rope_theta, T.Cast("float32", (tx * 4 + vec) * 2 % 64) / T.float32(64.0))}), pages[page_no, 0, by, page_offset, tx * 4 + vec]) + V_smem[tile_start_s + j, tx * 4 + vec] = pages[page_no, 1, by, page_offset, tx * 4 + vec] + else: + for vec in T.vectorized(4): + K_smem[tile_start_s + j, tx * 4 + vec] = T.float16(0.0) + V_smem[tile_start_s + j, tx * 4 + vec] = T.float16(0.0) + T.tvm_storage_sync("shared") + m_prev[0] = st_m[0] + for j in range(1): + for vec in T.vectorized(4): + QK_local[vec] = T.Cast("float32", Q_local[vec]) * T.Cast("float32", K_smem[tz + j, tx * 4 + vec]) * attn_score_scaling_factor * sm_scale + S_reduce_local[0] = T.float32(0.0) + for vec in T.unroll(4): + S_reduce_local[0] = S_reduce_local[0] + QK_local[vec] + with T.block("block_cross_thread"): + T.reads(S_reduce_local[0]) + T.writes(t0[0]) + T.attr(T.comm_reducer(lambda x0, y0: x0 + y0, [T.float32(0.0)]), "reduce_scope", T.reinterpret("handle", T.uint64(0))) + T.tvm_thread_allreduce(T.uint32(1), S_reduce_local[0], T.bool(True), t0[0], tx) + S_local[j] = T.float32(-50000.0) + if iterator * 16 + tz + j < kv_chunk_len[0]: + S_local[j] = t0[0] + st_m[0] = T.max(st_m[0], S_local[j]) + o_scale: T.float32 = T.exp2(m_prev[0] - st_m[0]) + st_d[0] = st_d[0] * o_scale + for j in range(1): + S_local[j] = T.exp2(S_local[j] - st_m[0]) + st_d[0] = st_d[0] + S_local[j] + for j in T.vectorized(4): + O_local[j] = O_local[j] * o_scale + for j in range(1): + for vec in T.vectorized(4): + V_local[vec] = V_smem[tz + j, tx * 4 + vec] + for vec in T.vectorized(4): + O_local[vec] = O_local[vec] + T.Cast("float32", V_local[vec]) * S_local[j] + for vec in T.vectorized(4): + O_allreduce[tz, ty, tx * 4 + vec] = O_local[vec] + md_allreduce[tz, ty, 0] = st_m[0] + md_allreduce[tz, ty, 1] = st_d[0] + T.tvm_storage_sync("shared") + st_m[0] = T.float32(-50000.0) + st_d[0] = T.float32(1.0) + for vec in T.vectorized(4): + O_local[vec] = T.float32(0.0) + for j in range(16): + m_prev[0] = st_m[0] + d_prev[0] = st_d[0] + other_m[0] = md_allreduce[j, ty, 0] + other_d[0] = md_allreduce[j, ty, 1] + for vec in T.vectorized(4): + other_o[vec] = O_allreduce[j, ty, tx * 4 + vec] + st_m[0] = T.max(st_m[0], other_m[0]) + st_d[0] = d_prev[0] * T.exp2(m_prev[0] - st_m[0]) + other_d[0] * T.exp2(other_m[0] - st_m[0]) + exp_mprev[0] = T.exp2(m_prev[0] - st_m[0]) + exp_otherm[0] = T.exp2(other_m[0] - st_m[0]) + for vec in T.vectorized(4): + O_local[vec] = O_local[vec] * exp_mprev[0] + other_o[vec] * exp_otherm[0] + for vec in T.vectorized(4): + O_local[vec] = O_local[vec] / st_d[0] + for vec in T.vectorized(4): + output[batch_idx, by + bz + ty, tx * 4 + vec] = T.Cast("float16", O_local[vec]) + lse[batch_idx, by + bz + ty] = st_m[0] + T.log2(st_d[0]) + + @T.prim_func + def batch_decode_paged_kv_sliding_window(_0: T.int32, Q_handle: T.handle, pages_handle: T.handle, page_table_indptr_handle: T.handle, page_table_values_handle: T.handle, var_length_info: T.handle, k_rope_pos_offset_handle: T.handle, q_rope_position_handle: T.handle, output_handle: T.handle, lse_handle: T.handle, rotary_mode: T.int32, rope_scale: T.float32, rope_theta: T.float32, attn_score_scaling_factor: T.float32): + T.func_attr({"target": T.target({"host": {"keys": ["arm_cpu", "cpu"], "kind": "llvm", "mtriple": "aarch64-linux-android", "tag": ""}, "keys": ["opencl", "gpu"], "kind": "opencl", "max_function_args": 128, "max_num_threads": 256, "max_shared_memory_per_block": 16384, "max_threads_per_block": 256, "tag": "", "texture_spatial_limit": 16384, "thread_warp_size": 1}), "tir.is_scheduled": 1}) + B = T.int32(is_size_var=True) + Q = T.match_buffer(Q_handle, (B, 12, 64), "float16") + max_num_pages = T.int32(is_size_var=True) + pages = T.match_buffer(pages_handle, (max_num_pages, 2, 12, 16, 64), "float16", offset_factor=1) + page_table_indptr = T.match_buffer(page_table_indptr_handle, (B + 1,), "int32", offset_factor=1) + nnz_pages = T.int32(is_size_var=True) + page_table_values = T.match_buffer(page_table_values_handle, (nnz_pages,), "int32", offset_factor=1) + length_info = T.match_buffer(var_length_info, (3, B), "int32", offset_factor=1) + k_rope_pos_offset = T.match_buffer(k_rope_pos_offset_handle, (B,), "int32", offset_factor=1) + q_rope_position = T.match_buffer(q_rope_position_handle, (B,), "int32", offset_factor=1) + output = T.match_buffer(output_handle, (B, 12, 64), "float16") + lse = T.match_buffer(lse_handle, (B, 12)) + # with T.block("root"): + sm_scale: T.float32 = T.float32(0.18033688011112042) + for bx in T.thread_binding(B, thread="blockIdx.x"): + for fused_by_bz in T.thread_binding(12, thread="blockIdx.y"): + for ty in T.thread_binding(1, thread="threadIdx.y"): + for tx in T.thread_binding(16, thread="threadIdx.x"): + for tz in T.thread_binding(16, thread="threadIdx.z"): + with T.block("attn"): + T.reads(page_table_indptr[bx:bx + 2], length_info[0:3, bx], q_rope_position[bx], Q[bx, fused_by_bz // 12 + ty + fused_by_bz % 12, tx * 4 - 32:tx * 4 - 32 + 68]) + T.writes(output[bx, fused_by_bz % 12 + fused_by_bz // 12 + ty, tx * 4:tx * 4 + 4], lse[bx, fused_by_bz % 12 + fused_by_bz // 12 + ty]) + Q_local = T.alloc_buffer((4,), "float16", scope="local") + kv_chunk_len = T.alloc_buffer((1,), "int32", scope="local") + K_smem = T.alloc_buffer((16, 64), "float16", scope="shared") + V_smem = T.alloc_buffer((16, 64), "float16", scope="shared") + O_allreduce = T.alloc_buffer((16, 1, 64), scope="shared") + md_allreduce = T.alloc_buffer((16, 1, 2), scope="shared") + S_reduce_local = T.alloc_buffer((1,), scope="local") + t0 = T.alloc_buffer((1,), scope="local") + S_local = T.alloc_buffer((1,), scope="local") + QK_local = T.alloc_buffer((4,), scope="local") + V_local = T.alloc_buffer((4,), "float16", scope="local") + m_prev = T.alloc_buffer((1,), scope="local") + d_prev = T.alloc_buffer((1,), scope="local") + other_m = T.alloc_buffer((1,), scope="local") + other_d = T.alloc_buffer((1,), scope="local") + exp_mprev = T.alloc_buffer((1,), scope="local") + exp_otherm = T.alloc_buffer((1,), scope="local") + other_o = T.alloc_buffer((4,), scope="local") + st_m = T.alloc_buffer((1,), scope="local") + st_d = T.alloc_buffer((1,), scope="local") + O_local = T.alloc_buffer((4,), scope="local") + by: T.int32 = fused_by_bz % 12 + bz: T.int32 = fused_by_bz // 12 + batch_idx: T.int32 = bx + cur_page_indptr_begin: T.int32 = page_table_indptr[batch_idx] + cur_page_indptr_end: T.int32 = page_table_indptr[batch_idx + 1] + kv_chunk_len[0] = T.if_then_else(cur_page_indptr_begin != cur_page_indptr_end, (cur_page_indptr_end - cur_page_indptr_begin - 1) * 16 + length_info[0, batch_idx] - length_info[1, batch_idx] + length_info[2, batch_idx], 0) + st_m[0] = T.float32(-50000.0) + st_d[0] = T.float32(1.0) + for vec in T.vectorized(4): + O_local[vec] = T.float32(0.0) + for vec in T.vectorized(4): + freq = T.float32() + Q_local[vec] = T.if_then_else(rotary_mode == 1, T.Let(T.Cast("float16", T.cos(freq) * T.Cast("float32", Q[bx, by + bz + ty, tx * 4 + vec]) + T.sin(freq) * T.Cast("float32", T.if_then_else(tx * 4 + vec < 32, Q[bx, by + bz + ty, tx * 4 + vec + 32] * T.float16(-1.0), Q[bx, by + bz + ty, tx * 4 + vec - 32]))), where={freq: T.Cast("float32", q_rope_position[batch_idx]) * rope_scale / T.pow(rope_theta, T.Cast("float32", (tx * 4 + vec) * 2 % 64) / T.float32(64.0))}), Q[bx, by + bz + ty, tx * 4 + vec]) + for iterator in range((kv_chunk_len[0] + 15) // 16): + tile_start_s: T.int32 = tz + ty + tile_start_g: T.int32 = iterator * 16 + tz + ty + for j in range(1): + with T.block("KV_load"): + T.reads() + T.writes() + row_g: T.int32 = tile_start_g + j + if row_g < kv_chunk_len[0]: + seq_offset: T.int32 = T.if_then_else(row_g < length_info[2, batch_idx], row_g, row_g - length_info[2, batch_idx] + length_info[1, batch_idx]) + page_no: T.int32 = page_table_values[cur_page_indptr_begin + seq_offset // 16] + page_offset: T.int32 = seq_offset % 16 + for vec in T.vectorized(4): + freq = T.float32() + K_smem[tile_start_s + j, tx * 4 + vec] = T.if_then_else(rotary_mode == 1, T.Let(T.Cast("float16", T.cos(freq) * T.Cast("float32", pages[page_no, 0, by, page_offset, tx * 4 + vec]) + T.sin(freq) * T.Cast("float32", T.if_then_else(tx * 4 + vec < 32, pages[page_no, 0, by, page_offset, tx * 4 + vec + 32] * T.float16(-1.0), pages[page_no, 0, by, page_offset, tx * 4 + vec - 32]))), where={freq: T.Cast("float32", k_rope_pos_offset[batch_idx] + row_g) * rope_scale / T.pow(rope_theta, T.Cast("float32", (tx * 4 + vec) * 2 % 64) / T.float32(64.0))}), pages[page_no, 0, by, page_offset, tx * 4 + vec]) + V_smem[tile_start_s + j, tx * 4 + vec] = pages[page_no, 1, by, page_offset, tx * 4 + vec] + else: + for vec in T.vectorized(4): + K_smem[tile_start_s + j, tx * 4 + vec] = T.float16(0.0) + V_smem[tile_start_s + j, tx * 4 + vec] = T.float16(0.0) + T.tvm_storage_sync("shared") + m_prev[0] = st_m[0] + for j in range(1): + for vec in T.vectorized(4): + QK_local[vec] = T.Cast("float32", Q_local[vec]) * T.Cast("float32", K_smem[tz + j, tx * 4 + vec]) * attn_score_scaling_factor * sm_scale + S_reduce_local[0] = T.float32(0.0) + for vec in T.unroll(4): + S_reduce_local[0] = S_reduce_local[0] + QK_local[vec] + with T.block("block_cross_thread"): + T.reads(S_reduce_local[0]) + T.writes(t0[0]) + T.attr(T.comm_reducer(lambda x0, y0: x0 + y0, [T.float32(0.0)]), "reduce_scope", T.reinterpret("handle", T.uint64(0))) + T.tvm_thread_allreduce(T.uint32(1), S_reduce_local[0], T.bool(True), t0[0], tx) + S_local[j] = T.float32(-50000.0) + if iterator * 16 + tz + j < kv_chunk_len[0]: + S_local[j] = t0[0] + st_m[0] = T.max(st_m[0], S_local[j]) + o_scale: T.float32 = T.exp2(m_prev[0] - st_m[0]) + st_d[0] = st_d[0] * o_scale + for j in range(1): + S_local[j] = T.exp2(S_local[j] - st_m[0]) + st_d[0] = st_d[0] + S_local[j] + for j in T.vectorized(4): + O_local[j] = O_local[j] * o_scale + for j in range(1): + for vec in T.vectorized(4): + V_local[vec] = V_smem[tz + j, tx * 4 + vec] + for vec in T.vectorized(4): + O_local[vec] = O_local[vec] + T.Cast("float32", V_local[vec]) * S_local[j] + for vec in T.vectorized(4): + O_allreduce[tz, ty, tx * 4 + vec] = O_local[vec] + md_allreduce[tz, ty, 0] = st_m[0] + md_allreduce[tz, ty, 1] = st_d[0] + T.tvm_storage_sync("shared") + st_m[0] = T.float32(-50000.0) + st_d[0] = T.float32(1.0) + for vec in T.vectorized(4): + O_local[vec] = T.float32(0.0) + for j in range(16): + m_prev[0] = st_m[0] + d_prev[0] = st_d[0] + other_m[0] = md_allreduce[j, ty, 0] + other_d[0] = md_allreduce[j, ty, 1] + for vec in T.vectorized(4): + other_o[vec] = O_allreduce[j, ty, tx * 4 + vec] + st_m[0] = T.max(st_m[0], other_m[0]) + st_d[0] = d_prev[0] * T.exp2(m_prev[0] - st_m[0]) + other_d[0] * T.exp2(other_m[0] - st_m[0]) + exp_mprev[0] = T.exp2(m_prev[0] - st_m[0]) + exp_otherm[0] = T.exp2(other_m[0] - st_m[0]) + for vec in T.vectorized(4): + O_local[vec] = O_local[vec] * exp_mprev[0] + other_o[vec] * exp_otherm[0] + for vec in T.vectorized(4): + O_local[vec] = O_local[vec] / st_d[0] + for vec in T.vectorized(4): + output[batch_idx, by + bz + ty, tx * 4 + vec] = T.Cast("float16", O_local[vec]) + lse[batch_idx, by + bz + ty] = st_m[0] + T.log2(st_d[0]) + + @T.prim_func + def batch_prefill_paged_kv(_0: T.int32, var_q: T.handle, var_q_indptr: T.handle, var_pages: T.handle, var_page_indptr: T.handle, var_page_values: T.handle, var_length_info: T.handle, var_k_rope_pos_offset: T.handle, var_q_rope_position: T.handle, var_output: T.handle, var_lse: T.handle, causal: T.int32, rotary_mode: T.int32, rope_scale: T.float32, rope_theta: T.float32, attn_score_scaling_factor: T.float32): + T.func_attr({"target": T.target({"host": {"keys": ["arm_cpu", "cpu"], "kind": "llvm", "mtriple": "aarch64-linux-android", "tag": ""}, "keys": ["opencl", "gpu"], "kind": "opencl", "max_function_args": 128, "max_num_threads": 256, "max_shared_memory_per_block": 16384, "max_threads_per_block": 256, "tag": "", "texture_spatial_limit": 16384, "thread_warp_size": 1}), "tir.is_scheduled": 1}) + total_len = T.int32(is_size_var=True) + q = T.match_buffer(var_q, (total_len, 12, 64), "float16") + batch_size = T.int32(is_size_var=True) + q_indptr = T.match_buffer(var_q_indptr, (batch_size + 1,), "int32", offset_factor=1) + max_num_pages = T.int32(is_size_var=True) + pages = T.match_buffer(var_pages, (max_num_pages, 2, 12, 16, 64), "float16", offset_factor=1) + page_indptr = T.match_buffer(var_page_indptr, (batch_size + 1,), "int32", offset_factor=1) + nnz_pages = T.int32(is_size_var=True) + page_values = T.match_buffer(var_page_values, (nnz_pages,), "int32", offset_factor=1) + length_info = T.match_buffer(var_length_info, (batch_size,), "int32", offset_factor=1) + k_rope_pos_offset = T.match_buffer(var_k_rope_pos_offset, (batch_size,), "int32", offset_factor=1) + q_rope_position = T.match_buffer(var_q_rope_position, (total_len,), "int32", offset_factor=1) + output = T.match_buffer(var_output, (total_len, 12, 64), "float16") + lse = T.match_buffer(var_lse, (total_len, 12)) + # with T.block("root"): + for lbx in T.thread_binding(16, thread="blockIdx.x"): + for lby in T.thread_binding(12, thread="blockIdx.y"): + for lty in T.thread_binding(4, thread="threadIdx.y"): + for ltx in T.thread_binding(32, thread="threadIdx.x"): + with T.block("attn"): + bx, by, ty, tx = T.axis.remap("SSSS", [lbx, lby, lty, ltx]) + T.reads() + T.writes() + tile_id = T.alloc_buffer((1,), "int32", scope="local") + batch_idx = T.alloc_buffer((1,), "int32", scope="local") + batch_tiles = T.alloc_buffer((1,), "int32", scope="local") + batch_rows = T.alloc_buffer((1,), "int32", scope="local") + iterator = T.alloc_buffer((1,), "int32", scope="local") + kv_chunk_len = T.alloc_buffer((1,), "int32", scope="local") + Q_smem = T.alloc_buffer((32, 64), "float16", scope="shared") + K_smem = T.alloc_buffer((16, 64), "float16", scope="shared") + V_smem = T.alloc_buffer((16, 64), "float16", scope="shared") + S_smem = T.alloc_buffer((32, 16), scope="shared") + S_local = T.alloc_buffer((32, 16), scope="local") + O_local = T.alloc_buffer((32, 64), scope="local") + m_smem = T.alloc_buffer((32,), scope="shared") + m_prev_smem = T.alloc_buffer((32,), scope="shared") + d_smem = T.alloc_buffer((32,), scope="shared") + m_new = T.alloc_buffer((1,), scope="local") + m_prev = T.alloc_buffer((1,), scope="local") + d_new = T.alloc_buffer((1,), scope="local") + tile_id[0] = bx + batch_idx[0] = 0 + batch_rows[0] = q_indptr[1] - q_indptr[0] + batch_tiles[0] = (batch_rows[0] + 32 - 1) // 32 + while T.tvm_thread_invariant(batch_idx[0] < batch_size): + while tile_id[0] >= batch_tiles[0] and batch_idx[0] < batch_size: + tile_id[0] = tile_id[0] - batch_tiles[0] + batch_idx[0] = batch_idx[0] + 1 + if batch_idx[0] < batch_size: + b_idx: T.int32 = batch_idx[0] + batch_rows[0] = q_indptr[b_idx + 1] - q_indptr[b_idx] + batch_tiles[0] = (batch_rows[0] + 32 - 1) // 32 + if T.tvm_thread_invariant(batch_idx[0] < batch_size): + b_idx: T.int32 = batch_idx[0] + LH_start: T.int32 = tile_id[0] * 32 + q_indptr_val: T.int32 = q_indptr[b_idx] + cur_page_indptr_begin: T.int32 = page_indptr[b_idx] + cur_page_indptr_end: T.int32 = page_indptr[b_idx + 1] + kv_chunk_len[0] = T.if_then_else(cur_page_indptr_begin != cur_page_indptr_end, (cur_page_indptr_end - cur_page_indptr_begin - 1) * 16 + length_info[b_idx], 0) + T.tvm_storage_sync("shared") + for i in range(1): + row: T.int32 = i * 32 * 4 + ty * 32 + tx + if row < 32: + m_smem[row] = T.float32(-50000.0) + d_smem[row] = T.float32(1.0) + for li_0_lj_0_fused_0 in T.thread_binding(4, thread="threadIdx.y"): + for li_0_lj_0_fused_1 in T.thread_binding(32, thread="threadIdx.x"): + for li_1, lj_1 in T.grid(4, 4): + with T.block("O_init"): + i = T.axis.spatial(32, (li_0_lj_0_fused_0 * 32 + li_0_lj_0_fused_1) // 16 * 4 + li_1) + j = T.axis.spatial(64, (li_0_lj_0_fused_0 * 32 + li_0_lj_0_fused_1) % 16 * 4 + lj_1) + T.reads() + T.writes(O_local[i, j]) + O_local[i, j] = T.float32(0.0) + T.tvm_storage_sync("shared") + for li_lj_fused_0 in range(4): + for li_lj_fused_1 in T.thread_binding(4, thread="threadIdx.y"): + for li_lj_fused_2 in T.thread_binding(32, thread="threadIdx.x"): + for li_lj_fused_3 in T.vectorized(4): + with T.block("Q_load"): + i = T.axis.spatial(32, (li_lj_fused_0 * 512 + li_lj_fused_1 * 128 + li_lj_fused_2 * 4 + li_lj_fused_3) // 64) + j = T.axis.spatial(64, (li_lj_fused_0 * 512 + li_lj_fused_1 * 128 + li_lj_fused_2 * 4 + li_lj_fused_3) % 64) + T.reads() + T.writes() + cur_L: T.int32 = q_indptr_val + (LH_start + i) + cur_H_qo: T.int32 = by + if cur_L < q_indptr[b_idx + 1]: + freq = T.float32() + Q_smem[i, j] = T.if_then_else(rotary_mode == 1, T.Let(T.Cast("float16", T.cos(freq) * T.Cast("float32", q[cur_L, cur_H_qo, j]) + T.sin(freq) * T.Cast("float32", T.if_then_else(j < 32, q[cur_L, cur_H_qo, j + 32] * T.float16(-1.0), q[cur_L, cur_H_qo, j - 32]))), where={freq: T.Cast("float32", q_rope_position[cur_L]) * rope_scale / T.pow(rope_theta, T.Cast("float32", j * 2 % 64) / T.float32(64.0))}), q[cur_L, cur_H_qo, j]) + else: + Q_smem[i, j] = T.float16(0.0) + T.tvm_storage_sync("shared") + for iterator_1 in range((kv_chunk_len[0] + 15) // 16): + L_kv_start: T.int32 = iterator_1 * 16 + for lz_ly_fused_0 in range(2): + for lz_ly_fused_1 in T.thread_binding(4, thread="threadIdx.y"): + for lz_ly_fused_2 in T.thread_binding(32, thread="threadIdx.x"): + for lz_ly_fused_3 in T.vectorized(4): + with T.block("K_load"): + i = T.axis.spatial(16, (lz_ly_fused_0 * 512 + lz_ly_fused_1 * 128 + lz_ly_fused_2 * 4 + lz_ly_fused_3) // 64) + j = T.axis.spatial(64, (lz_ly_fused_0 * 512 + lz_ly_fused_1 * 128 + lz_ly_fused_2 * 4 + lz_ly_fused_3) % 64) + T.reads() + T.writes() + cur_L: T.int32 = L_kv_start + i + if cur_L < kv_chunk_len[0]: + seq_offset: T.int32 = cur_L + page_no: T.int32 = page_values[cur_page_indptr_begin + seq_offset // 16] + page_offset: T.int32 = seq_offset % 16 + freq = T.float32() + K_smem[i, j] = T.if_then_else(rotary_mode == 1, T.Let(T.Cast("float16", T.cos(freq) * T.Cast("float32", pages[page_no, 0, by, page_offset, j]) + T.sin(freq) * T.Cast("float32", T.if_then_else(j < 32, pages[page_no, 0, by, page_offset, j + 32] * T.float16(-1.0), pages[page_no, 0, by, page_offset, j - 32]))), where={freq: T.Cast("float32", k_rope_pos_offset[b_idx] + cur_L) * rope_scale / T.pow(rope_theta, T.Cast("float32", j * 2 % 64) / T.float32(64.0))}), pages[page_no, 0, by, page_offset, j]) + else: + K_smem[i, j] = T.float16(0.0) + T.tvm_storage_sync("shared") + for lz_ly_fused_0 in range(2): + for lz_ly_fused_1 in T.thread_binding(4, thread="threadIdx.y"): + for lz_ly_fused_2 in T.thread_binding(32, thread="threadIdx.x"): + for lz_ly_fused_3 in T.vectorized(4): + with T.block("V_load"): + i = T.axis.spatial(16, (lz_ly_fused_0 * 512 + lz_ly_fused_1 * 128 + lz_ly_fused_2 * 4 + lz_ly_fused_3) // 64) + j = T.axis.spatial(64, (lz_ly_fused_0 * 512 + lz_ly_fused_1 * 128 + lz_ly_fused_2 * 4 + lz_ly_fused_3) % 64) + T.reads() + T.writes() + cur_L: T.int32 = L_kv_start + i + if cur_L < kv_chunk_len[0]: + seq_offset: T.int32 = cur_L + page_no: T.int32 = page_values[cur_page_indptr_begin + seq_offset // 16] + page_offset: T.int32 = seq_offset % 16 + V_smem[i, j] = pages[page_no, 1, by, page_offset, j] + else: + V_smem[i, j] = T.float16(0.0) + T.tvm_storage_sync("shared") + with T.block(""): + T.reads(Q_smem[0:32, 0:64], K_smem[0:16, 0:64]) + T.writes(S_local[0:32, 0:16]) + for li_0_lj_0_fused_0_init in T.thread_binding(4, thread="threadIdx.y"): + for li_0_lj_0_fused_1_init in T.thread_binding(32, thread="threadIdx.x"): + for li_1_init, lj_1_init in T.grid(2, 2): + with T.block("S_gemm_init"): + i = T.axis.spatial(32, (li_0_lj_0_fused_0_init * 32 + li_0_lj_0_fused_1_init) // 8 * 2 + li_1_init) + j = T.axis.spatial(16, (li_0_lj_0_fused_0_init * 32 + li_0_lj_0_fused_1_init) % 8 * 2 + lj_1_init) + T.reads() + T.writes(S_local[i, j]) + S_local[i, j] = T.float32(0.0) + for li_0_lj_0_fused_0 in T.thread_binding(4, thread="threadIdx.y"): + for li_0_lj_0_fused_1 in T.thread_binding(32, thread="threadIdx.x"): + for lk_0, li_1, lj_1, lk_1 in T.grid(8, 2, 2, 8): + with T.block("S_gemm_update"): + i = T.axis.spatial(32, (li_0_lj_0_fused_0 * 32 + li_0_lj_0_fused_1) // 8 * 2 + li_1) + j = T.axis.spatial(16, (li_0_lj_0_fused_0 * 32 + li_0_lj_0_fused_1) % 8 * 2 + lj_1) + k = T.axis.reduce(64, lk_0 * 8 + lk_1) + T.reads(S_local[i, j], Q_smem[i, k], K_smem[j, k]) + T.writes(S_local[i, j]) + S_local[i, j] = S_local[i, j] + T.Cast("float32", Q_smem[i, k]) * T.Cast("float32", K_smem[j, k]) * attn_score_scaling_factor * T.float32(0.18033688011112042) + T.tvm_storage_sync("shared") + for li_0_lj_0_fused_0 in T.thread_binding(4, thread="threadIdx.y"): + for li_0_lj_0_fused_1 in T.thread_binding(32, thread="threadIdx.x"): + for li_1, lj_1 in T.grid(2, 2): + with T.block("S_store"): + i = T.axis.spatial(32, (li_0_lj_0_fused_0 * 32 + li_0_lj_0_fused_1) // 8 * 2 + li_1) + j = T.axis.spatial(16, (li_0_lj_0_fused_0 * 32 + li_0_lj_0_fused_1) % 8 * 2 + lj_1) + T.reads(S_local[i, j]) + T.writes(S_smem[i, j]) + S_smem[i, j] = S_local[i, j] + T.tvm_storage_sync("shared") + for i in range(1): + row: T.int32 = i * 32 * 4 + ty * 32 + tx + if row < 32: + with T.block("update1"): + T.reads(m_smem[row], kv_chunk_len[0], q_indptr[b_idx:b_idx + 2], m_new[i], S_smem[row, 0:16], d_smem[row], m_prev[i]) + T.writes(m_prev[i], m_new[i], d_new[i]) + m_prev[i] = m_smem[row] + m_new[i] = m_smem[row] + row_: T.int32 = LH_start + row + for j in range(16): + if T.if_then_else(causal > 0, L_kv_start + j < kv_chunk_len[0] - (q_indptr[b_idx + 1] - q_indptr[b_idx]) + row_ + 1, L_kv_start + j < kv_chunk_len[0]): + m_new[i] = T.max(m_new[i], S_smem[row, j]) + d_new[i] = d_smem[row] * T.exp2(m_prev[i] - m_new[i]) + for i in range(1): + row: T.int32 = i * 32 * 4 + ty * 32 + tx + with T.block("update"): + T.reads(kv_chunk_len[0], q_indptr[b_idx:b_idx + 2], S_smem[row, 0:16], m_new[i]) + T.writes(S_smem[row, 0:16]) + for j in range(16): + if row < 32: + row_: T.int32 = LH_start + row + if T.if_then_else(causal > 0, L_kv_start + j < kv_chunk_len[0] - (q_indptr[b_idx + 1] - q_indptr[b_idx]) + row_ + 1, L_kv_start + j < kv_chunk_len[0]): + S_smem[row, j] = T.exp2(S_smem[row, j] - m_new[i]) + else: + S_smem[row, j] = T.exp2(T.float32(-50000.0) - m_new[i]) + for i in range(1): + row: T.int32 = i * 32 * 4 + ty * 32 + tx + if row < 32: + with T.block("update"): + T.reads(d_new[i], S_smem[row, 0:16], m_new[i], m_prev[i]) + T.writes(d_new[i], m_smem[row], d_smem[row], m_prev_smem[row]) + for j in range(16): + d_new[i] = d_new[i] + S_smem[row, j] + m_smem[row] = m_new[i] + d_smem[row] = d_new[i] + m_prev_smem[row] = m_prev[i] + T.tvm_storage_sync("shared") + with T.block(""): + T.reads(m_prev_smem[0:32], m_smem[0:32], S_smem[0:32, 0:16], V_smem[0:16, 0:64]) + T.writes(O_local[0:32, 0:64]) + for li_0_lj_0_fused_0_init in T.thread_binding(4, thread="threadIdx.y"): + for li_0_lj_0_fused_1_init in T.thread_binding(32, thread="threadIdx.x"): + for li_1_init, lj_1_init in T.grid(4, 4): + with T.block("O_gemm_init"): + i = T.axis.spatial(32, (li_0_lj_0_fused_0_init * 32 + li_0_lj_0_fused_1_init) // 16 * 4 + li_1_init) + j = T.axis.spatial(64, (li_0_lj_0_fused_0_init * 32 + li_0_lj_0_fused_1_init) % 16 * 4 + lj_1_init) + T.reads() + T.writes(O_local[i, j]) + O_local[i, j] = O_local[i, j] * T.exp2(m_prev_smem[i] - m_smem[i]) + for li_0_lj_0_fused_0 in T.thread_binding(4, thread="threadIdx.y"): + for li_0_lj_0_fused_1 in T.thread_binding(32, thread="threadIdx.x"): + for lk_0, lk_1, li_1, lj_1 in T.grid(2, 8, 4, 4): + with T.block("O_gemm_update"): + i = T.axis.spatial(32, (li_0_lj_0_fused_0 * 32 + li_0_lj_0_fused_1) // 16 * 4 + li_1) + j = T.axis.spatial(64, (li_0_lj_0_fused_0 * 32 + li_0_lj_0_fused_1) % 16 * 4 + lj_1) + k = T.axis.reduce(16, lk_0 * 8 + lk_1) + T.reads(O_local[i, j], m_prev_smem[i], m_smem[i], S_smem[i, k], V_smem[k, j]) + T.writes(O_local[i, j]) + O_local[i, j] = O_local[i, j] + S_smem[i, k] * T.Cast("float32", V_smem[k, j]) + for li_0_lj_0_fused_0 in T.thread_binding(4, thread="threadIdx.y"): + for li_0_lj_0_fused_1 in T.thread_binding(32, thread="threadIdx.x"): + for li_1, lj_1 in T.grid(4, 4): + with T.block("O_store"): + i = T.axis.spatial(32, (li_0_lj_0_fused_0 * 32 + li_0_lj_0_fused_1) // 16 * 4 + li_1) + j = T.axis.spatial(64, (li_0_lj_0_fused_0 * 32 + li_0_lj_0_fused_1) % 16 * 4 + lj_1) + T.reads(q_indptr[b_idx:b_idx + 2], O_local[i, j], d_smem[i]) + T.writes(output[q_indptr[b_idx] + (LH_start + i), by, j]) + cur_L: T.int32 = q_indptr[b_idx] + (LH_start + i) + cur_H_qo: T.int32 = by + if cur_L < q_indptr[b_idx + 1]: + output[cur_L, cur_H_qo, j] = T.Cast("float16", O_local[i, j] / d_smem[i]) + for li_0 in range(1): + for li_1 in T.thread_binding(4, thread="threadIdx.y"): + for li_2 in T.thread_binding(32, thread="threadIdx.x"): + with T.block("lse_store"): + i = T.axis.spatial(32, li_0 * 128 + li_1 * 32 + li_2) + T.where((li_0 * 4 + li_1) * 32 + li_2 < 32) + T.reads(q_indptr[b_idx:b_idx + 2], m_smem[i], d_smem[i]) + T.writes(lse[q_indptr[b_idx] + (LH_start + i), by]) + cur_L: T.int32 = q_indptr[b_idx] + (LH_start + i) + cur_H_qo: T.int32 = by + if cur_L < q_indptr[b_idx + 1]: + lse[cur_L, cur_H_qo] = m_smem[i] + T.log2(d_smem[i]) + tile_id[0] = tile_id[0] + 16 + + @T.prim_func + def batch_prefill_paged_kv_sliding_window(_0: T.int32, var_q: T.handle, var_q_indptr: T.handle, var_pages: T.handle, var_page_indptr: T.handle, var_page_values: T.handle, var_length_info: T.handle, var_k_rope_pos_offset: T.handle, var_q_rope_position: T.handle, var_output: T.handle, var_lse: T.handle, causal: T.int32, rotary_mode: T.int32, rope_scale: T.float32, rope_theta: T.float32, attn_score_scaling_factor: T.float32): + T.func_attr({"target": T.target({"host": {"keys": ["arm_cpu", "cpu"], "kind": "llvm", "mtriple": "aarch64-linux-android", "tag": ""}, "keys": ["opencl", "gpu"], "kind": "opencl", "max_function_args": 128, "max_num_threads": 256, "max_shared_memory_per_block": 16384, "max_threads_per_block": 256, "tag": "", "texture_spatial_limit": 16384, "thread_warp_size": 1}), "tir.is_scheduled": 1}) + total_len = T.int32(is_size_var=True) + q = T.match_buffer(var_q, (total_len, 12, 64), "float16") + batch_size = T.int32(is_size_var=True) + q_indptr = T.match_buffer(var_q_indptr, (batch_size + 1,), "int32", offset_factor=1) + max_num_pages = T.int32(is_size_var=True) + pages = T.match_buffer(var_pages, (max_num_pages, 2, 12, 16, 64), "float16", offset_factor=1) + page_indptr = T.match_buffer(var_page_indptr, (batch_size + 1,), "int32", offset_factor=1) + nnz_pages = T.int32(is_size_var=True) + page_values = T.match_buffer(var_page_values, (nnz_pages,), "int32", offset_factor=1) + length_info = T.match_buffer(var_length_info, (3, batch_size), "int32", offset_factor=1) + k_rope_pos_offset = T.match_buffer(var_k_rope_pos_offset, (batch_size,), "int32", offset_factor=1) + q_rope_position = T.match_buffer(var_q_rope_position, (total_len,), "int32", offset_factor=1) + output = T.match_buffer(var_output, (total_len, 12, 64), "float16") + lse = T.match_buffer(var_lse, (total_len, 12)) + # with T.block("root"): + for lbx in T.thread_binding(16, thread="blockIdx.x"): + for lby in T.thread_binding(12, thread="blockIdx.y"): + for lty in T.thread_binding(4, thread="threadIdx.y"): + for ltx in T.thread_binding(32, thread="threadIdx.x"): + with T.block("attn"): + bx, by, ty, tx = T.axis.remap("SSSS", [lbx, lby, lty, ltx]) + T.reads() + T.writes() + tile_id = T.alloc_buffer((1,), "int32", scope="local") + batch_idx = T.alloc_buffer((1,), "int32", scope="local") + batch_tiles = T.alloc_buffer((1,), "int32", scope="local") + batch_rows = T.alloc_buffer((1,), "int32", scope="local") + iterator = T.alloc_buffer((1,), "int32", scope="local") + kv_chunk_len = T.alloc_buffer((1,), "int32", scope="local") + Q_smem = T.alloc_buffer((32, 64), "float16", scope="shared") + K_smem = T.alloc_buffer((16, 64), "float16", scope="shared") + V_smem = T.alloc_buffer((16, 64), "float16", scope="shared") + S_smem = T.alloc_buffer((32, 16), scope="shared") + S_local = T.alloc_buffer((32, 16), scope="local") + O_local = T.alloc_buffer((32, 64), scope="local") + m_smem = T.alloc_buffer((32,), scope="shared") + m_prev_smem = T.alloc_buffer((32,), scope="shared") + d_smem = T.alloc_buffer((32,), scope="shared") + m_new = T.alloc_buffer((1,), scope="local") + m_prev = T.alloc_buffer((1,), scope="local") + d_new = T.alloc_buffer((1,), scope="local") + tile_id[0] = bx + batch_idx[0] = 0 + batch_rows[0] = q_indptr[1] - q_indptr[0] + batch_tiles[0] = (batch_rows[0] + 32 - 1) // 32 + while T.tvm_thread_invariant(batch_idx[0] < batch_size): + while tile_id[0] >= batch_tiles[0] and batch_idx[0] < batch_size: + tile_id[0] = tile_id[0] - batch_tiles[0] + batch_idx[0] = batch_idx[0] + 1 + if batch_idx[0] < batch_size: + b_idx: T.int32 = batch_idx[0] + batch_rows[0] = q_indptr[b_idx + 1] - q_indptr[b_idx] + batch_tiles[0] = (batch_rows[0] + 32 - 1) // 32 + if T.tvm_thread_invariant(batch_idx[0] < batch_size): + b_idx: T.int32 = batch_idx[0] + LH_start: T.int32 = tile_id[0] * 32 + q_indptr_val: T.int32 = q_indptr[b_idx] + cur_page_indptr_begin: T.int32 = page_indptr[b_idx] + cur_page_indptr_end: T.int32 = page_indptr[b_idx + 1] + kv_chunk_len[0] = T.if_then_else(cur_page_indptr_begin != cur_page_indptr_end, (cur_page_indptr_end - cur_page_indptr_begin - 1) * 16 + length_info[0, b_idx] - length_info[1, b_idx] + length_info[2, b_idx], 0) + T.tvm_storage_sync("shared") + for i in range(1): + row: T.int32 = i * 32 * 4 + ty * 32 + tx + if row < 32: + m_smem[row] = T.float32(-50000.0) + d_smem[row] = T.float32(1.0) + for li_0_lj_0_fused_0 in T.thread_binding(4, thread="threadIdx.y"): + for li_0_lj_0_fused_1 in T.thread_binding(32, thread="threadIdx.x"): + for li_1, lj_1 in T.grid(4, 4): + with T.block("O_init"): + i = T.axis.spatial(32, (li_0_lj_0_fused_0 * 32 + li_0_lj_0_fused_1) // 16 * 4 + li_1) + j = T.axis.spatial(64, (li_0_lj_0_fused_0 * 32 + li_0_lj_0_fused_1) % 16 * 4 + lj_1) + T.reads() + T.writes(O_local[i, j]) + O_local[i, j] = T.float32(0.0) + T.tvm_storage_sync("shared") + for li_lj_fused_0 in range(4): + for li_lj_fused_1 in T.thread_binding(4, thread="threadIdx.y"): + for li_lj_fused_2 in T.thread_binding(32, thread="threadIdx.x"): + for li_lj_fused_3 in T.vectorized(4): + with T.block("Q_load"): + i = T.axis.spatial(32, (li_lj_fused_0 * 512 + li_lj_fused_1 * 128 + li_lj_fused_2 * 4 + li_lj_fused_3) // 64) + j = T.axis.spatial(64, (li_lj_fused_0 * 512 + li_lj_fused_1 * 128 + li_lj_fused_2 * 4 + li_lj_fused_3) % 64) + T.reads() + T.writes() + cur_L: T.int32 = q_indptr_val + (LH_start + i) + cur_H_qo: T.int32 = by + if cur_L < q_indptr[b_idx + 1]: + freq = T.float32() + Q_smem[i, j] = T.if_then_else(rotary_mode == 1, T.Let(T.Cast("float16", T.cos(freq) * T.Cast("float32", q[cur_L, cur_H_qo, j]) + T.sin(freq) * T.Cast("float32", T.if_then_else(j < 32, q[cur_L, cur_H_qo, j + 32] * T.float16(-1.0), q[cur_L, cur_H_qo, j - 32]))), where={freq: T.Cast("float32", q_rope_position[cur_L]) * rope_scale / T.pow(rope_theta, T.Cast("float32", j * 2 % 64) / T.float32(64.0))}), q[cur_L, cur_H_qo, j]) + else: + Q_smem[i, j] = T.float16(0.0) + T.tvm_storage_sync("shared") + for iterator_1 in range((kv_chunk_len[0] + 15) // 16): + L_kv_start: T.int32 = iterator_1 * 16 + for lz_ly_fused_0 in range(2): + for lz_ly_fused_1 in T.thread_binding(4, thread="threadIdx.y"): + for lz_ly_fused_2 in T.thread_binding(32, thread="threadIdx.x"): + for lz_ly_fused_3 in T.vectorized(4): + with T.block("K_load"): + i = T.axis.spatial(16, (lz_ly_fused_0 * 512 + lz_ly_fused_1 * 128 + lz_ly_fused_2 * 4 + lz_ly_fused_3) // 64) + j = T.axis.spatial(64, (lz_ly_fused_0 * 512 + lz_ly_fused_1 * 128 + lz_ly_fused_2 * 4 + lz_ly_fused_3) % 64) + T.reads() + T.writes() + cur_L: T.int32 = L_kv_start + i + if cur_L < kv_chunk_len[0]: + seq_offset: T.int32 = T.if_then_else(cur_L < length_info[2, b_idx], cur_L, cur_L - length_info[2, b_idx] + length_info[1, b_idx]) + page_no: T.int32 = page_values[cur_page_indptr_begin + seq_offset // 16] + page_offset: T.int32 = seq_offset % 16 + freq = T.float32() + K_smem[i, j] = T.if_then_else(rotary_mode == 1, T.Let(T.Cast("float16", T.cos(freq) * T.Cast("float32", pages[page_no, 0, by, page_offset, j]) + T.sin(freq) * T.Cast("float32", T.if_then_else(j < 32, pages[page_no, 0, by, page_offset, j + 32] * T.float16(-1.0), pages[page_no, 0, by, page_offset, j - 32]))), where={freq: T.Cast("float32", k_rope_pos_offset[b_idx] + cur_L) * rope_scale / T.pow(rope_theta, T.Cast("float32", j * 2 % 64) / T.float32(64.0))}), pages[page_no, 0, by, page_offset, j]) + else: + K_smem[i, j] = T.float16(0.0) + T.tvm_storage_sync("shared") + for lz_ly_fused_0 in range(2): + for lz_ly_fused_1 in T.thread_binding(4, thread="threadIdx.y"): + for lz_ly_fused_2 in T.thread_binding(32, thread="threadIdx.x"): + for lz_ly_fused_3 in T.vectorized(4): + with T.block("V_load"): + i = T.axis.spatial(16, (lz_ly_fused_0 * 512 + lz_ly_fused_1 * 128 + lz_ly_fused_2 * 4 + lz_ly_fused_3) // 64) + j = T.axis.spatial(64, (lz_ly_fused_0 * 512 + lz_ly_fused_1 * 128 + lz_ly_fused_2 * 4 + lz_ly_fused_3) % 64) + T.reads() + T.writes() + cur_L: T.int32 = L_kv_start + i + if cur_L < kv_chunk_len[0]: + seq_offset: T.int32 = T.if_then_else(cur_L < length_info[2, b_idx], cur_L, cur_L - length_info[2, b_idx] + length_info[1, b_idx]) + page_no: T.int32 = page_values[cur_page_indptr_begin + seq_offset // 16] + page_offset: T.int32 = seq_offset % 16 + V_smem[i, j] = pages[page_no, 1, by, page_offset, j] + else: + V_smem[i, j] = T.float16(0.0) + T.tvm_storage_sync("shared") + with T.block(""): + T.reads(Q_smem[0:32, 0:64], K_smem[0:16, 0:64]) + T.writes(S_local[0:32, 0:16]) + for li_0_lj_0_fused_0_init in T.thread_binding(4, thread="threadIdx.y"): + for li_0_lj_0_fused_1_init in T.thread_binding(32, thread="threadIdx.x"): + for li_1_init, lj_1_init in T.grid(2, 2): + with T.block("S_gemm_init"): + i = T.axis.spatial(32, (li_0_lj_0_fused_0_init * 32 + li_0_lj_0_fused_1_init) // 8 * 2 + li_1_init) + j = T.axis.spatial(16, (li_0_lj_0_fused_0_init * 32 + li_0_lj_0_fused_1_init) % 8 * 2 + lj_1_init) + T.reads() + T.writes(S_local[i, j]) + S_local[i, j] = T.float32(0.0) + for li_0_lj_0_fused_0 in T.thread_binding(4, thread="threadIdx.y"): + for li_0_lj_0_fused_1 in T.thread_binding(32, thread="threadIdx.x"): + for lk_0, li_1, lj_1, lk_1 in T.grid(8, 2, 2, 8): + with T.block("S_gemm_update"): + i = T.axis.spatial(32, (li_0_lj_0_fused_0 * 32 + li_0_lj_0_fused_1) // 8 * 2 + li_1) + j = T.axis.spatial(16, (li_0_lj_0_fused_0 * 32 + li_0_lj_0_fused_1) % 8 * 2 + lj_1) + k = T.axis.reduce(64, lk_0 * 8 + lk_1) + T.reads(S_local[i, j], Q_smem[i, k], K_smem[j, k]) + T.writes(S_local[i, j]) + S_local[i, j] = S_local[i, j] + T.Cast("float32", Q_smem[i, k]) * T.Cast("float32", K_smem[j, k]) * attn_score_scaling_factor * T.float32(0.18033688011112042) + T.tvm_storage_sync("shared") + for li_0_lj_0_fused_0 in T.thread_binding(4, thread="threadIdx.y"): + for li_0_lj_0_fused_1 in T.thread_binding(32, thread="threadIdx.x"): + for li_1, lj_1 in T.grid(2, 2): + with T.block("S_store"): + i = T.axis.spatial(32, (li_0_lj_0_fused_0 * 32 + li_0_lj_0_fused_1) // 8 * 2 + li_1) + j = T.axis.spatial(16, (li_0_lj_0_fused_0 * 32 + li_0_lj_0_fused_1) % 8 * 2 + lj_1) + T.reads(S_local[i, j]) + T.writes(S_smem[i, j]) + S_smem[i, j] = S_local[i, j] + T.tvm_storage_sync("shared") + for i in range(1): + row: T.int32 = i * 32 * 4 + ty * 32 + tx + if row < 32: + with T.block("update1"): + T.reads(m_smem[row], kv_chunk_len[0], q_indptr[b_idx:b_idx + 2], m_new[i], S_smem[row, 0:16], d_smem[row], m_prev[i]) + T.writes(m_prev[i], m_new[i], d_new[i]) + m_prev[i] = m_smem[row] + m_new[i] = m_smem[row] + row_: T.int32 = LH_start + row + for j in range(16): + if T.if_then_else(causal > 0, L_kv_start + j < kv_chunk_len[0] - (q_indptr[b_idx + 1] - q_indptr[b_idx]) + row_ + 1, L_kv_start + j < kv_chunk_len[0]): + m_new[i] = T.max(m_new[i], S_smem[row, j]) + d_new[i] = d_smem[row] * T.exp2(m_prev[i] - m_new[i]) + for i in range(1): + row: T.int32 = i * 32 * 4 + ty * 32 + tx + with T.block("update"): + T.reads(kv_chunk_len[0], q_indptr[b_idx:b_idx + 2], S_smem[row, 0:16], m_new[i]) + T.writes(S_smem[row, 0:16]) + for j in range(16): + if row < 32: + row_: T.int32 = LH_start + row + if T.if_then_else(causal > 0, L_kv_start + j < kv_chunk_len[0] - (q_indptr[b_idx + 1] - q_indptr[b_idx]) + row_ + 1, L_kv_start + j < kv_chunk_len[0]): + S_smem[row, j] = T.exp2(S_smem[row, j] - m_new[i]) + else: + S_smem[row, j] = T.exp2(T.float32(-50000.0) - m_new[i]) + for i in range(1): + row: T.int32 = i * 32 * 4 + ty * 32 + tx + if row < 32: + with T.block("update"): + T.reads(d_new[i], S_smem[row, 0:16], m_new[i], m_prev[i]) + T.writes(d_new[i], m_smem[row], d_smem[row], m_prev_smem[row]) + for j in range(16): + d_new[i] = d_new[i] + S_smem[row, j] + m_smem[row] = m_new[i] + d_smem[row] = d_new[i] + m_prev_smem[row] = m_prev[i] + T.tvm_storage_sync("shared") + with T.block(""): + T.reads(m_prev_smem[0:32], m_smem[0:32], S_smem[0:32, 0:16], V_smem[0:16, 0:64]) + T.writes(O_local[0:32, 0:64]) + for li_0_lj_0_fused_0_init in T.thread_binding(4, thread="threadIdx.y"): + for li_0_lj_0_fused_1_init in T.thread_binding(32, thread="threadIdx.x"): + for li_1_init, lj_1_init in T.grid(4, 4): + with T.block("O_gemm_init"): + i = T.axis.spatial(32, (li_0_lj_0_fused_0_init * 32 + li_0_lj_0_fused_1_init) // 16 * 4 + li_1_init) + j = T.axis.spatial(64, (li_0_lj_0_fused_0_init * 32 + li_0_lj_0_fused_1_init) % 16 * 4 + lj_1_init) + T.reads() + T.writes(O_local[i, j]) + O_local[i, j] = O_local[i, j] * T.exp2(m_prev_smem[i] - m_smem[i]) + for li_0_lj_0_fused_0 in T.thread_binding(4, thread="threadIdx.y"): + for li_0_lj_0_fused_1 in T.thread_binding(32, thread="threadIdx.x"): + for lk_0, lk_1, li_1, lj_1 in T.grid(2, 8, 4, 4): + with T.block("O_gemm_update"): + i = T.axis.spatial(32, (li_0_lj_0_fused_0 * 32 + li_0_lj_0_fused_1) // 16 * 4 + li_1) + j = T.axis.spatial(64, (li_0_lj_0_fused_0 * 32 + li_0_lj_0_fused_1) % 16 * 4 + lj_1) + k = T.axis.reduce(16, lk_0 * 8 + lk_1) + T.reads(O_local[i, j], m_prev_smem[i], m_smem[i], S_smem[i, k], V_smem[k, j]) + T.writes(O_local[i, j]) + O_local[i, j] = O_local[i, j] + S_smem[i, k] * T.Cast("float32", V_smem[k, j]) + for li_0_lj_0_fused_0 in T.thread_binding(4, thread="threadIdx.y"): + for li_0_lj_0_fused_1 in T.thread_binding(32, thread="threadIdx.x"): + for li_1, lj_1 in T.grid(4, 4): + with T.block("O_store"): + i = T.axis.spatial(32, (li_0_lj_0_fused_0 * 32 + li_0_lj_0_fused_1) // 16 * 4 + li_1) + j = T.axis.spatial(64, (li_0_lj_0_fused_0 * 32 + li_0_lj_0_fused_1) % 16 * 4 + lj_1) + T.reads(q_indptr[b_idx:b_idx + 2], O_local[i, j], d_smem[i]) + T.writes(output[q_indptr[b_idx] + (LH_start + i), by, j]) + cur_L: T.int32 = q_indptr[b_idx] + (LH_start + i) + cur_H_qo: T.int32 = by + if cur_L < q_indptr[b_idx + 1]: + output[cur_L, cur_H_qo, j] = T.Cast("float16", O_local[i, j] / d_smem[i]) + for li_0 in range(1): + for li_1 in T.thread_binding(4, thread="threadIdx.y"): + for li_2 in T.thread_binding(32, thread="threadIdx.x"): + with T.block("lse_store"): + i = T.axis.spatial(32, li_0 * 128 + li_1 * 32 + li_2) + T.where((li_0 * 4 + li_1) * 32 + li_2 < 32) + T.reads(q_indptr[b_idx:b_idx + 2], m_smem[i], d_smem[i]) + T.writes(lse[q_indptr[b_idx] + (LH_start + i), by]) + cur_L: T.int32 = q_indptr[b_idx] + (LH_start + i) + cur_H_qo: T.int32 = by + if cur_L < q_indptr[b_idx + 1]: + lse[cur_L, cur_H_qo] = m_smem[i] + T.log2(d_smem[i]) + tile_id[0] = tile_id[0] + 16 + + @T.prim_func + def batch_prefill_ragged_kv(var_q: T.handle, var_q_indptr: T.handle, var_k: T.handle, var_v: T.handle, var_kv_indptr: T.handle, var_q_rope_position: T.handle, var_k_rope_pos_offset: T.handle, var_output: T.handle, var_lse: T.handle, causal: T.int32, rotary_mode: T.int32, rope_scale: T.float32, rope_theta: T.float32, attn_score_scaling_factor: T.float32): + T.func_attr({"target": T.target({"host": {"keys": ["arm_cpu", "cpu"], "kind": "llvm", "mtriple": "aarch64-linux-android", "tag": ""}, "keys": ["opencl", "gpu"], "kind": "opencl", "max_function_args": 128, "max_num_threads": 256, "max_shared_memory_per_block": 16384, "max_threads_per_block": 256, "tag": "", "texture_spatial_limit": 16384, "thread_warp_size": 1}), "tir.is_scheduled": 1}) + qo_len = T.int32(is_size_var=True) + q = T.match_buffer(var_q, (qo_len, 12, 64), "float16") + batch_size = T.int32(is_size_var=True) + q_indptr = T.match_buffer(var_q_indptr, (batch_size + 1,), "int32", offset_factor=1) + kv_len = T.int32(is_size_var=True) + k = T.match_buffer(var_k, (kv_len, 12, 64), "float16") + v = T.match_buffer(var_v, (kv_len, 12, 64), "float16") + kv_indptr = T.match_buffer(var_kv_indptr, (batch_size + 1,), "int32", offset_factor=1) + q_rope_position = T.match_buffer(var_q_rope_position, (qo_len,), "int32", offset_factor=1) + k_rope_pos_offset = T.match_buffer(var_k_rope_pos_offset, (batch_size,), "int32", offset_factor=1) + output = T.match_buffer(var_output, (qo_len, 12, 64), "float16") + lse = T.match_buffer(var_lse, (qo_len, 12)) + # with T.block("root"): + for lbx in T.thread_binding(8, thread="blockIdx.x"): + for lby in T.thread_binding(12, thread="blockIdx.y"): + for lty in T.thread_binding(4, thread="threadIdx.y"): + for ltx in T.thread_binding(32, thread="threadIdx.x"): + with T.block("attn"): + bx, by, ty, tx = T.axis.remap("SSSS", [lbx, lby, lty, ltx]) + T.reads() + T.writes() + tile_id = T.alloc_buffer((1,), "int32", scope="local") + batch_idx = T.alloc_buffer((1,), "int32", scope="local") + batch_tiles = T.alloc_buffer((1,), "int32", scope="local") + batch_rows = T.alloc_buffer((1,), "int32", scope="local") + iterator = T.alloc_buffer((1,), "int32", scope="local") + kv_chunk_len = T.alloc_buffer((1,), "int32", scope="local") + Q_smem = T.alloc_buffer((32, 64), "float16", scope="shared") + K_smem = T.alloc_buffer((64, 32), "float16", scope="shared") + V_smem = T.alloc_buffer((32, 64), "float16", scope="shared") + S_smem = T.alloc_buffer((32, 32), scope="shared") + S_local = T.alloc_buffer((32, 32), scope="local") + O_local = T.alloc_buffer((32, 64), scope="local") + m_smem = T.alloc_buffer((32,), scope="shared") + m_prev_smem = T.alloc_buffer((32,), scope="shared") + d_smem = T.alloc_buffer((32,), scope="shared") + m_new = T.alloc_buffer((1,), scope="local") + m_prev = T.alloc_buffer((1,), scope="local") + d_new = T.alloc_buffer((1,), scope="local") + tile_id[0] = bx + batch_idx[0] = 0 + batch_rows[0] = q_indptr[1] - q_indptr[0] + batch_tiles[0] = (batch_rows[0] + 32 - 1) // 32 + while T.tvm_thread_invariant(batch_idx[0] < batch_size): + while tile_id[0] >= batch_tiles[0] and batch_idx[0] < batch_size: + tile_id[0] = tile_id[0] - batch_tiles[0] + batch_idx[0] = batch_idx[0] + 1 + if batch_idx[0] < batch_size: + b_idx: T.int32 = batch_idx[0] + batch_rows[0] = q_indptr[b_idx + 1] - q_indptr[b_idx] + batch_tiles[0] = (batch_rows[0] + 32 - 1) // 32 + if T.tvm_thread_invariant(batch_idx[0] < batch_size): + b_idx: T.int32 = batch_idx[0] + q_indptr_val: T.int32 = q_indptr[b_idx] + LH_start: T.int32 = tile_id[0] * 32 + kv_chunk_len[0] = kv_indptr[b_idx + 1] - kv_indptr[b_idx] + T.tvm_storage_sync("shared") + for i in range(1): + row: T.int32 = i * 32 * 4 + ty * 32 + tx + if row < 32: + m_smem[row] = T.float32(-50000.0) + d_smem[row] = T.float32(1.0) + for li_0_lj_0_fused_0 in T.thread_binding(4, thread="threadIdx.y"): + for li_0_lj_0_fused_1 in T.thread_binding(32, thread="threadIdx.x"): + for li_1 in range(4): + for lj_1_0 in T.unroll(1): + for lj_1_1 in T.vectorized(4): + with T.block("O_init"): + i = T.axis.spatial(32, (li_0_lj_0_fused_0 * 32 + li_0_lj_0_fused_1) // 16 * 4 + li_1) + j = T.axis.spatial(64, (li_0_lj_0_fused_0 * 32 + li_0_lj_0_fused_1) % 16 * 4 + lj_1_0 * 4 + lj_1_1) + T.reads() + T.writes(O_local[i, j]) + O_local[i, j] = T.float32(0.0) + T.tvm_storage_sync("shared") + for li_1_lj_0_1_fused_0 in T.thread_binding(4, thread="threadIdx.y"): + for li_1_lj_0_1_fused_1 in T.thread_binding(32, thread="threadIdx.x"): + for li_0, lj_0_0 in T.grid(1, 2): + for lj_1 in T.vectorized(8): + with T.block("Q_load"): + i = T.axis.spatial(32, li_0 * 32 + (li_1_lj_0_1_fused_0 * 32 + li_1_lj_0_1_fused_1) // 4) + j = T.axis.spatial(64, lj_0_0 * 32 + (li_1_lj_0_1_fused_0 * 32 + li_1_lj_0_1_fused_1) % 4 * 8 + lj_1) + T.reads() + T.writes() + cur_L: T.int32 = q_indptr_val + (LH_start + i) + cur_H_qo: T.int32 = by + if cur_L < q_indptr[b_idx + 1]: + freq = T.float32() + Q_smem[i, j] = T.if_then_else(rotary_mode == 1, T.Let(T.Cast("float16", T.cos(freq) * T.Cast("float32", q[cur_L, cur_H_qo, j]) + T.sin(freq) * T.Cast("float32", T.if_then_else(j < 32, q[cur_L, cur_H_qo, j + 32] * T.float16(-1.0), q[cur_L, cur_H_qo, j - 32]))), where={freq: T.Cast("float32", q_rope_position[cur_L]) * rope_scale / T.pow(rope_theta, T.Cast("float32", j * 2 % 64) / T.float32(64.0))}), q[cur_L, cur_H_qo, j]) + else: + Q_smem[i, j] = T.float16(0.0) + T.tvm_storage_sync("shared") + for iterator_1 in range((kv_chunk_len[0] + 31) // 32): + L_kv_start: T.int32 = iterator_1 * 32 + L_kv_base: T.int32 = kv_indptr[b_idx] + for lz_1_ly_0_1_fused_0 in T.thread_binding(4, thread="threadIdx.y"): + for lz_1_ly_0_1_fused_1 in T.thread_binding(32, thread="threadIdx.x"): + for lz_0, ly_0_0 in T.grid(1, 2): + for ly_1 in T.vectorized(8): + with T.block("K_load"): + i = T.axis.spatial(32, lz_0 * 32 + (lz_1_ly_0_1_fused_0 * 32 + lz_1_ly_0_1_fused_1) // 4) + j = T.axis.spatial(64, ly_0_0 * 32 + (lz_1_ly_0_1_fused_0 * 32 + lz_1_ly_0_1_fused_1) % 4 * 8 + ly_1) + T.reads(kv_chunk_len[0], k_rope_pos_offset[b_idx], k[L_kv_base + L_kv_start + i, by, j - 32:j - 32 + 65]) + T.writes(K_smem[j, i]) + cur_L: T.int32 = L_kv_start + i + if cur_L < kv_chunk_len[0]: + freq = T.float32() + K_smem[j, i] = T.if_then_else(rotary_mode == 1, T.Let(T.Cast("float16", T.cos(freq) * T.Cast("float32", k[L_kv_base + cur_L, by, j]) + T.sin(freq) * T.Cast("float32", T.if_then_else(j < 32, k[L_kv_base + cur_L, by, j + 32] * T.float16(-1.0), k[L_kv_base + cur_L, by, j - 32]))), where={freq: T.Cast("float32", k_rope_pos_offset[b_idx] + cur_L) * rope_scale / T.pow(rope_theta, T.Cast("float32", j * 2 % 64) / T.float32(64.0))}), k[L_kv_base + cur_L, by, j]) + else: + K_smem[j, i] = T.float16(0.0) + T.tvm_storage_sync("shared") + for lz_1_ly_0_1_fused_0 in T.thread_binding(4, thread="threadIdx.y"): + for lz_1_ly_0_1_fused_1 in T.thread_binding(32, thread="threadIdx.x"): + for lz_0, ly_0_0 in T.grid(1, 2): + for ly_1 in T.vectorized(8): + with T.block("V_load"): + i = T.axis.spatial(32, lz_0 * 32 + (lz_1_ly_0_1_fused_0 * 32 + lz_1_ly_0_1_fused_1) // 4) + j = T.axis.spatial(64, ly_0_0 * 32 + (lz_1_ly_0_1_fused_0 * 32 + lz_1_ly_0_1_fused_1) % 4 * 8 + ly_1) + T.reads() + T.writes() + cur_L: T.int32 = L_kv_start + i + if cur_L < kv_chunk_len[0]: + V_smem[i, j] = v[L_kv_base + cur_L, by, j] + else: + V_smem[i, j] = T.float16(0.0) + T.tvm_storage_sync("shared") + with T.block(""): + T.reads(Q_smem[0:32, 0:64], K_smem[0:64, 0:32]) + T.writes(S_local[0:32, 0:32]) + for li_0_lj_0_fused_0_init in T.thread_binding(4, thread="threadIdx.y"): + for li_0_lj_0_fused_1_init in T.thread_binding(32, thread="threadIdx.x"): + for li_1_init in T.unroll(2): + for lj_1_0_init in T.unroll(1): + for lj_1_1_init in T.vectorized(4): + with T.block("S_gemm_init"): + i = T.axis.spatial(32, (li_0_lj_0_fused_0_init * 32 + li_0_lj_0_fused_1_init) // 8 * 2 + li_1_init) + j = T.axis.spatial(32, (li_0_lj_0_fused_0_init * 32 + li_0_lj_0_fused_1_init) % 8 * 4 + lj_1_0_init * 4 + lj_1_1_init) + T.reads() + T.writes(S_local[i, j]) + S_local[i, j] = T.float32(0.0) + for li_0_lj_0_fused_0 in T.thread_binding(4, thread="threadIdx.y"): + for li_0_lj_0_fused_1 in T.thread_binding(32, thread="threadIdx.x"): + for lk_0 in range(4): + for li_1 in T.unroll(2): + for lj_1_0 in T.unroll(1): + for lj_1_1 in T.vectorized(4): + for lk_1 in range(16): + with T.block("S_gemm_update"): + i = T.axis.spatial(32, (li_0_lj_0_fused_0 * 32 + li_0_lj_0_fused_1) // 8 * 2 + li_1) + j = T.axis.spatial(32, (li_0_lj_0_fused_0 * 32 + li_0_lj_0_fused_1) % 8 * 4 + lj_1_0 * 4 + lj_1_1) + k_1 = T.axis.reduce(64, lk_0 * 16 + lk_1) + T.reads(S_local[i, j], Q_smem[i, k_1], K_smem[k_1, j]) + T.writes(S_local[i, j]) + S_local[i, j] = S_local[i, j] + T.Cast("float32", Q_smem[i, k_1]) * T.Cast("float32", K_smem[k_1, j]) * attn_score_scaling_factor * T.float32(0.18033688011112042) + T.tvm_storage_sync("shared") + for li_0_lj_0_fused_0 in T.thread_binding(4, thread="threadIdx.y"): + for li_0_lj_0_fused_1 in T.thread_binding(32, thread="threadIdx.x"): + for li_1 in range(2): + for lj_1_0 in T.unroll(1): + for lj_1_1 in T.vectorized(4): + with T.block("S_store"): + i = T.axis.spatial(32, (li_0_lj_0_fused_0 * 32 + li_0_lj_0_fused_1) // 8 * 2 + li_1) + j = T.axis.spatial(32, (li_0_lj_0_fused_0 * 32 + li_0_lj_0_fused_1) % 8 * 4 + lj_1_0 * 4 + lj_1_1) + T.reads(S_local[i, j]) + T.writes(S_smem[i, j]) + S_smem[i, j] = S_local[i, j] + T.tvm_storage_sync("shared") + for i in range(1): + row: T.int32 = i * 32 * 4 + ty * 32 + tx + if row < 32: + with T.block("update1"): + T.reads(m_smem[row], kv_chunk_len[0], q_indptr[b_idx:b_idx + 2], m_new[i], S_smem[row, 0:32], d_smem[row], m_prev[i]) + T.writes(m_prev[i], m_new[i], d_new[i]) + m_prev[i] = m_smem[row] + m_new[i] = m_smem[row] + row_: T.int32 = LH_start + row + for j in range(32): + if T.if_then_else(causal > 0, L_kv_start + j < kv_chunk_len[0] - (q_indptr[b_idx + 1] - q_indptr[b_idx]) + row_ + 1, L_kv_start + j < kv_chunk_len[0]): + m_new[i] = T.max(m_new[i], S_smem[row, j]) + d_new[i] = d_smem[row] * T.exp2(m_prev[i] - m_new[i]) + for i in range(1): + row: T.int32 = i * 32 * 4 + ty * 32 + tx + with T.block("update"): + T.reads(kv_chunk_len[0], q_indptr[b_idx:b_idx + 2], S_smem[row, 0:32], m_new[i]) + T.writes(S_smem[row, 0:32]) + for j in range(32): + if row < 32: + row_: T.int32 = LH_start + row + if T.if_then_else(causal > 0, L_kv_start + j < kv_chunk_len[0] - (q_indptr[b_idx + 1] - q_indptr[b_idx]) + row_ + 1, L_kv_start + j < kv_chunk_len[0]): + S_smem[row, j] = T.exp2(S_smem[row, j] - m_new[i]) + else: + S_smem[row, j] = T.exp2(T.float32(-50000.0) - m_new[i]) + for i in range(1): + row: T.int32 = i * 32 * 4 + ty * 32 + tx + if row < 32: + with T.block("update"): + T.reads(d_new[i], S_smem[row, 0:32], m_new[i], m_prev[i]) + T.writes(d_new[i], m_smem[row], d_smem[row], m_prev_smem[row]) + for j in range(32): + d_new[i] = d_new[i] + S_smem[row, j] + m_smem[row] = m_new[i] + d_smem[row] = d_new[i] + m_prev_smem[row] = m_prev[i] + T.tvm_storage_sync("shared") + with T.block(""): + T.reads(m_prev_smem[0:32], m_smem[0:32], S_smem[0:32, 0:32], V_smem[0:32, 0:64]) + T.writes(O_local[0:32, 0:64]) + for li_0_lj_0_fused_0_init in T.thread_binding(4, thread="threadIdx.y"): + for li_0_lj_0_fused_1_init in T.thread_binding(32, thread="threadIdx.x"): + for li_1_init in T.unroll(4): + for lj_1_0_init in T.unroll(1): + for lj_1_1_init in T.vectorized(4): + with T.block("O_gemm_init"): + i = T.axis.spatial(32, (li_0_lj_0_fused_0_init * 32 + li_0_lj_0_fused_1_init) // 16 * 4 + li_1_init) + j = T.axis.spatial(64, (li_0_lj_0_fused_0_init * 32 + li_0_lj_0_fused_1_init) % 16 * 4 + lj_1_0_init * 4 + lj_1_1_init) + T.reads() + T.writes(O_local[i, j]) + O_local[i, j] = O_local[i, j] * T.exp2(m_prev_smem[i] - m_smem[i]) + for li_0_lj_0_fused_0 in T.thread_binding(4, thread="threadIdx.y"): + for li_0_lj_0_fused_1 in T.thread_binding(32, thread="threadIdx.x"): + for lk_0, lk_1 in T.grid(2, 16): + for li_1 in T.unroll(4): + for lj_1_0 in T.unroll(1): + for lj_1_1 in T.vectorized(4): + with T.block("O_gemm_update"): + i = T.axis.spatial(32, (li_0_lj_0_fused_0 * 32 + li_0_lj_0_fused_1) // 16 * 4 + li_1) + j = T.axis.spatial(64, (li_0_lj_0_fused_0 * 32 + li_0_lj_0_fused_1) % 16 * 4 + lj_1_0 * 4 + lj_1_1) + k_1 = T.axis.reduce(32, lk_0 * 16 + lk_1) + T.reads(O_local[i, j], m_prev_smem[i], m_smem[i], S_smem[i, k_1], V_smem[k_1, j]) + T.writes(O_local[i, j]) + O_local[i, j] = O_local[i, j] + S_smem[i, k_1] * T.Cast("float32", V_smem[k_1, j]) + for li_0_lj_0_fused_0 in T.thread_binding(4, thread="threadIdx.y"): + for li_0_lj_0_fused_1 in T.thread_binding(32, thread="threadIdx.x"): + for li_1 in range(4): + for lj_1_0 in T.unroll(1): + for lj_1_1 in T.vectorized(4): + with T.block("O_store"): + i = T.axis.spatial(32, (li_0_lj_0_fused_0 * 32 + li_0_lj_0_fused_1) // 16 * 4 + li_1) + j = T.axis.spatial(64, (li_0_lj_0_fused_0 * 32 + li_0_lj_0_fused_1) % 16 * 4 + lj_1_0 * 4 + lj_1_1) + T.reads(q_indptr[b_idx:b_idx + 2], O_local[i, j], d_smem[i]) + T.writes(output[q_indptr[b_idx] + (LH_start + i), by, j]) + cur_L: T.int32 = q_indptr[b_idx] + (LH_start + i) + cur_H_qo: T.int32 = by + if cur_L < q_indptr[b_idx + 1]: + output[cur_L, cur_H_qo, j] = T.Cast("float16", O_local[i, j] / d_smem[i]) + for li_0 in range(1): + for li_1 in T.thread_binding(4, thread="threadIdx.y"): + for li_2 in T.thread_binding(32, thread="threadIdx.x"): + with T.block("lse_store"): + i = T.axis.spatial(32, li_0 * 128 + li_1 * 32 + li_2) + T.where((li_0 * 4 + li_1) * 32 + li_2 < 32) + T.reads(q_indptr[b_idx:b_idx + 2], m_smem[i], d_smem[i]) + T.writes(lse[q_indptr[b_idx] + (LH_start + i), by]) + cur_L: T.int32 = q_indptr[b_idx] + (LH_start + i) + cur_H_qo: T.int32 = by + if cur_L < q_indptr[b_idx + 1]: + lse[cur_L, cur_H_qo] = m_smem[i] + T.log2(d_smem[i]) + tile_id[0] = tile_id[0] + 8 + + @T.prim_func + def batch_tree_attn(var_q: T.handle, var_q_indptr: T.handle, var_k: T.handle, var_v: T.handle, var_kv_indptr: T.handle, var_q_rope_position: T.handle, var_mn_indptr: T.handle, var_mask: T.handle, var_output: T.handle, var_lse: T.handle, rotary_mode: T.int32, rope_scale: T.float32, rope_theta: T.float32, attn_score_scaling_factor: T.float32, batch_size: T.int32): + T.func_attr({"target": T.target({"host": {"keys": ["arm_cpu", "cpu"], "kind": "llvm", "mtriple": "aarch64-linux-android", "tag": ""}, "keys": ["opencl", "gpu"], "kind": "opencl", "max_function_args": 128, "max_num_threads": 256, "max_shared_memory_per_block": 16384, "max_threads_per_block": 256, "tag": "", "texture_spatial_limit": 16384, "thread_warp_size": 1}), "tir.is_scheduled": 1}) + qo_len = T.int32(is_size_var=True) + q = T.match_buffer(var_q, (qo_len, 12, 64), "float16") + q_indptr = T.match_buffer(var_q_indptr, (batch_size + 1,), "int32", offset_factor=1) + kv_len = T.int32(is_size_var=True) + k = T.match_buffer(var_k, (kv_len, 12, 64), "float16") + v = T.match_buffer(var_v, (kv_len, 12, 64), "float16") + kv_indptr = T.match_buffer(var_kv_indptr, (batch_size + 1,), "int32", offset_factor=1) + q_rope_position = T.match_buffer(var_q_rope_position, (qo_len,), "int32", offset_factor=1) + mn_indptr = T.match_buffer(var_mn_indptr, (batch_size + 1,), "int32", offset_factor=1) + tree_size = T.int32(is_size_var=True) + mask = T.match_buffer(var_mask, (tree_size, 2), "int32", offset_factor=1) + output = T.match_buffer(var_output, (qo_len, 12, 64), "float16") + lse = T.match_buffer(var_lse, (qo_len, 12)) + # with T.block("root"): + for lbx in T.thread_binding(16, thread="blockIdx.x"): + for lby in T.thread_binding(12, thread="blockIdx.y"): + for lty in T.thread_binding(4, thread="threadIdx.y"): + for ltx in T.thread_binding(32, thread="threadIdx.x"): + with T.block("attn"): + bx, by, ty, tx = T.axis.remap("SSSS", [lbx, lby, lty, ltx]) + T.reads() + T.writes() + tile_id = T.alloc_buffer((1,), "int32", scope="local") + batch_idx = T.alloc_buffer((1,), "int32", scope="local") + batch_tiles = T.alloc_buffer((1,), "int32", scope="local") + batch_rows = T.alloc_buffer((1,), "int32", scope="local") + iterator = T.alloc_buffer((1,), "int32", scope="local") + kv_chunk_len = T.alloc_buffer((1,), "int32", scope="local") + Q_smem = T.alloc_buffer((32, 64), "float16", scope="shared") + K_smem = T.alloc_buffer((32, 64), "float16", scope="shared") + V_smem = T.alloc_buffer((32, 64), "float16", scope="shared") + S_smem = T.alloc_buffer((32, 32), scope="shared") + S_local = T.alloc_buffer((32, 32), scope="local") + O_local = T.alloc_buffer((32, 64), scope="local") + m_smem = T.alloc_buffer((32,), scope="shared") + m_prev_smem = T.alloc_buffer((32,), scope="shared") + d_smem = T.alloc_buffer((32,), scope="shared") + m_new = T.alloc_buffer((1,), scope="local") + m_prev = T.alloc_buffer((1,), scope="local") + d_new = T.alloc_buffer((1,), scope="local") + tile_id[0] = bx + batch_idx[0] = 0 + batch_rows[0] = q_indptr[1] - q_indptr[0] + batch_tiles[0] = (batch_rows[0] + 32 - 1) // 32 + while T.tvm_thread_invariant(batch_idx[0] < batch_size): + while tile_id[0] >= batch_tiles[0] and batch_idx[0] < batch_size: + tile_id[0] = tile_id[0] - batch_tiles[0] + batch_idx[0] = batch_idx[0] + 1 + if batch_idx[0] < batch_size: + b_idx: T.int32 = batch_idx[0] + batch_rows[0] = q_indptr[b_idx + 1] - q_indptr[b_idx] + batch_tiles[0] = (batch_rows[0] + 32 - 1) // 32 + if T.tvm_thread_invariant(batch_idx[0] < batch_size): + b_idx: T.int32 = batch_idx[0] + LH_start: T.int32 = tile_id[0] * 32 + q_indptr_val: T.int32 = q_indptr[b_idx] + kv_chunk_len[0] = kv_indptr[b_idx + 1] - kv_indptr[b_idx] + T.tvm_storage_sync("shared") + for i in range(1): + row: T.int32 = i * 32 * 4 + ty * 32 + tx + if row < 32: + m_smem[row] = T.float32(-50000.0) + d_smem[row] = T.float32(1.0) + for li_0_lj_0_fused_0 in T.thread_binding(4, thread="threadIdx.y"): + for li_0_lj_0_fused_1 in T.thread_binding(32, thread="threadIdx.x"): + for li_1, lj_1 in T.grid(4, 4): + with T.block("O_init"): + i = T.axis.spatial(32, (li_0_lj_0_fused_0 * 32 + li_0_lj_0_fused_1) // 16 * 4 + li_1) + j = T.axis.spatial(64, (li_0_lj_0_fused_0 * 32 + li_0_lj_0_fused_1) % 16 * 4 + lj_1) + T.reads() + T.writes(O_local[i, j]) + O_local[i, j] = T.float32(0.0) + T.tvm_storage_sync("shared") + for li_lj_fused_0 in range(4): + for li_lj_fused_1 in T.thread_binding(4, thread="threadIdx.y"): + for li_lj_fused_2 in T.thread_binding(32, thread="threadIdx.x"): + for li_lj_fused_3 in T.vectorized(4): + with T.block("Q_load"): + i = T.axis.spatial(32, (li_lj_fused_0 * 512 + li_lj_fused_1 * 128 + li_lj_fused_2 * 4 + li_lj_fused_3) // 64) + j = T.axis.spatial(64, (li_lj_fused_0 * 512 + li_lj_fused_1 * 128 + li_lj_fused_2 * 4 + li_lj_fused_3) % 64) + T.reads() + T.writes() + cur_L: T.int32 = q_indptr_val + (LH_start + i) + cur_H_qo: T.int32 = by + if cur_L < q_indptr[b_idx + 1]: + freq = T.float32() + Q_smem[i, j] = T.if_then_else(rotary_mode == 1, T.Let(T.Cast("float16", T.cos(freq) * T.Cast("float32", q[cur_L, cur_H_qo, j]) + T.sin(freq) * T.Cast("float32", T.if_then_else(j < 32, q[cur_L, cur_H_qo, j + 32] * T.float16(-1.0), q[cur_L, cur_H_qo, j - 32]))), where={freq: T.Cast("float32", q_rope_position[cur_L]) * rope_scale / T.pow(rope_theta, T.Cast("float32", j * 2 % 64) / T.float32(64.0))}), q[cur_L, cur_H_qo, j]) + else: + Q_smem[i, j] = T.float16(0.0) + T.tvm_storage_sync("shared") + for iterator_1 in range((kv_chunk_len[0] + 31) // 32): + L_kv_start: T.int32 = iterator_1 * 32 + L_kv_base: T.int32 = kv_indptr[b_idx] + for lz_ly_fused_0 in range(4): + for lz_ly_fused_1 in T.thread_binding(4, thread="threadIdx.y"): + for lz_ly_fused_2 in T.thread_binding(32, thread="threadIdx.x"): + for lz_ly_fused_3 in T.vectorized(4): + with T.block("KV_load"): + i = T.axis.spatial(32, (lz_ly_fused_0 * 512 + lz_ly_fused_1 * 128 + lz_ly_fused_2 * 4 + lz_ly_fused_3) // 64) + j = T.axis.spatial(64, (lz_ly_fused_0 * 512 + lz_ly_fused_1 * 128 + lz_ly_fused_2 * 4 + lz_ly_fused_3) % 64) + T.reads() + T.writes() + cur_L: T.int32 = L_kv_base + L_kv_start + i + if L_kv_start + i < kv_chunk_len[0]: + freq = T.float32() + K_smem[i, j] = T.if_then_else(rotary_mode == 1, T.Let(T.Cast("float16", T.cos(freq) * T.Cast("float32", k[cur_L, by, j]) + T.sin(freq) * T.Cast("float32", T.if_then_else(j < 32, k[cur_L, by, j + 32] * T.float16(-1.0), k[cur_L, by, j - 32]))), where={freq: T.Cast("float32", q_rope_position[cur_L]) * rope_scale / T.pow(rope_theta, T.Cast("float32", j * 2 % 64) / T.float32(64.0))}), k[cur_L, by, j]) + V_smem[i, j] = v[cur_L, by, j] + else: + K_smem[i, j] = T.float16(0.0) + V_smem[i, j] = T.float16(0.0) + T.tvm_storage_sync("shared") + with T.block(""): + T.reads(Q_smem[0:32, 0:64], K_smem[0:32, 0:64]) + T.writes(S_local[0:32, 0:32]) + for li_0_lj_0_fused_0_init in T.thread_binding(4, thread="threadIdx.y"): + for li_0_lj_0_fused_1_init in T.thread_binding(32, thread="threadIdx.x"): + for li_1_init, lj_1_init in T.grid(2, 4): + with T.block("S_gemm_init"): + i = T.axis.spatial(32, (li_0_lj_0_fused_0_init * 32 + li_0_lj_0_fused_1_init) // 8 * 2 + li_1_init) + j = T.axis.spatial(32, (li_0_lj_0_fused_0_init * 32 + li_0_lj_0_fused_1_init) % 8 * 4 + lj_1_init) + T.reads() + T.writes(S_local[i, j]) + S_local[i, j] = T.float32(0.0) + for li_0_lj_0_fused_0 in T.thread_binding(4, thread="threadIdx.y"): + for li_0_lj_0_fused_1 in T.thread_binding(32, thread="threadIdx.x"): + for lk_0, li_1, lj_1, lk_1 in T.grid(8, 2, 4, 8): + with T.block("S_gemm_update"): + i = T.axis.spatial(32, (li_0_lj_0_fused_0 * 32 + li_0_lj_0_fused_1) // 8 * 2 + li_1) + j = T.axis.spatial(32, (li_0_lj_0_fused_0 * 32 + li_0_lj_0_fused_1) % 8 * 4 + lj_1) + k_1 = T.axis.reduce(64, lk_0 * 8 + lk_1) + T.reads(S_local[i, j], Q_smem[i, k_1], K_smem[j, k_1]) + T.writes(S_local[i, j]) + S_local[i, j] = S_local[i, j] + T.Cast("float32", Q_smem[i, k_1]) * T.Cast("float32", K_smem[j, k_1]) * attn_score_scaling_factor * T.float32(0.18033688011112042) + T.tvm_storage_sync("shared") + for li_0_lj_0_fused_0 in T.thread_binding(4, thread="threadIdx.y"): + for li_0_lj_0_fused_1 in T.thread_binding(32, thread="threadIdx.x"): + for li_1, lj_1 in T.grid(2, 4): + with T.block("S_store"): + i = T.axis.spatial(32, (li_0_lj_0_fused_0 * 32 + li_0_lj_0_fused_1) // 8 * 2 + li_1) + j = T.axis.spatial(32, (li_0_lj_0_fused_0 * 32 + li_0_lj_0_fused_1) % 8 * 4 + lj_1) + T.reads(S_local[i, j]) + T.writes(S_smem[i, j]) + S_smem[i, j] = S_local[i, j] + T.tvm_storage_sync("shared") + for i in range(1): + row: T.int32 = i * 32 * 4 + ty * 32 + tx + if row < 32: + with T.block("update1"): + T.reads(m_smem[row], kv_chunk_len[0], mn_indptr[b_idx:b_idx + 2], mask[T.min(LH_start + row + mn_indptr[b_idx + 1] + q_indptr[b_idx] - q_indptr[b_idx + 1], L_kv_start + mn_indptr[b_idx + 1] - kv_chunk_len[0]):T.min(LH_start + row + mn_indptr[b_idx + 1] + q_indptr[b_idx] - q_indptr[b_idx + 1], L_kv_start + mn_indptr[b_idx + 1] - kv_chunk_len[0]) + (T.max(LH_start + row + mn_indptr[b_idx + 1] + q_indptr[b_idx] - q_indptr[b_idx + 1], L_kv_start + mn_indptr[b_idx + 1] + 31 - kv_chunk_len[0]) + 1 - T.min(LH_start + row + mn_indptr[b_idx + 1] + q_indptr[b_idx] - q_indptr[b_idx + 1], L_kv_start + mn_indptr[b_idx + 1] - kv_chunk_len[0])), 0:2], q_indptr[b_idx:b_idx + 2], m_new[i], S_smem[row, 0:32], d_smem[row], m_prev[i]) + T.writes(m_prev[i], m_new[i], d_new[i]) + m_prev[i] = m_smem[row] + m_new[i] = m_smem[row] + row_: T.int32 = LH_start + row + for j in range(32): + if L_kv_start + j < kv_chunk_len[0] and (L_kv_start + j < kv_chunk_len[0] - (mn_indptr[b_idx + 1] - mn_indptr[b_idx]) or mask[mn_indptr[b_idx] + (row_ + (mn_indptr[b_idx + 1] - mn_indptr[b_idx]) - (q_indptr[b_idx + 1] - q_indptr[b_idx])), 0] >= mask[mn_indptr[b_idx] + (L_kv_start + j - (kv_chunk_len[0] - (mn_indptr[b_idx + 1] - mn_indptr[b_idx]))), 0] and mask[mn_indptr[b_idx] + (row_ + (mn_indptr[b_idx + 1] - mn_indptr[b_idx]) - (q_indptr[b_idx + 1] - q_indptr[b_idx])), 0] < mask[mn_indptr[b_idx] + (L_kv_start + j - (kv_chunk_len[0] - (mn_indptr[b_idx + 1] - mn_indptr[b_idx]))), 1]): + m_new[i] = T.max(m_new[i], S_smem[row, j]) + d_new[i] = d_smem[row] * T.exp2(m_prev[i] - m_new[i]) + for i in range(1): + row: T.int32 = i * 32 * 4 + ty * 32 + tx + with T.block("update"): + T.reads(kv_chunk_len[0], mn_indptr[b_idx:b_idx + 2], mask[T.min(LH_start + row + mn_indptr[b_idx + 1] + q_indptr[b_idx] - q_indptr[b_idx + 1], L_kv_start + mn_indptr[b_idx + 1] - kv_chunk_len[0]):T.min(LH_start + row + mn_indptr[b_idx + 1] + q_indptr[b_idx] - q_indptr[b_idx + 1], L_kv_start + mn_indptr[b_idx + 1] - kv_chunk_len[0]) + (T.max(LH_start + row + mn_indptr[b_idx + 1] + q_indptr[b_idx] - q_indptr[b_idx + 1], L_kv_start + mn_indptr[b_idx + 1] + 31 - kv_chunk_len[0]) + 1 - T.min(LH_start + row + mn_indptr[b_idx + 1] + q_indptr[b_idx] - q_indptr[b_idx + 1], L_kv_start + mn_indptr[b_idx + 1] - kv_chunk_len[0])), 0:2], q_indptr[b_idx:b_idx + 2], S_smem[row, 0:32], m_new[i]) + T.writes(S_smem[row, 0:32]) + for j in range(32): + if row < 32: + row_: T.int32 = LH_start + row + if L_kv_start + j < kv_chunk_len[0] and (L_kv_start + j < kv_chunk_len[0] - (mn_indptr[b_idx + 1] - mn_indptr[b_idx]) or mask[mn_indptr[b_idx] + (row_ + (mn_indptr[b_idx + 1] - mn_indptr[b_idx]) - (q_indptr[b_idx + 1] - q_indptr[b_idx])), 0] >= mask[mn_indptr[b_idx] + (L_kv_start + j - (kv_chunk_len[0] - (mn_indptr[b_idx + 1] - mn_indptr[b_idx]))), 0] and mask[mn_indptr[b_idx] + (row_ + (mn_indptr[b_idx + 1] - mn_indptr[b_idx]) - (q_indptr[b_idx + 1] - q_indptr[b_idx])), 0] < mask[mn_indptr[b_idx] + (L_kv_start + j - (kv_chunk_len[0] - (mn_indptr[b_idx + 1] - mn_indptr[b_idx]))), 1]): + S_smem[row, j] = T.exp2(S_smem[row, j] - m_new[i]) + else: + S_smem[row, j] = T.exp2(T.float32(-50000.0) - m_new[i]) + for i in range(1): + row: T.int32 = i * 32 * 4 + ty * 32 + tx + if row < 32: + with T.block("update"): + T.reads(d_new[i], S_smem[row, 0:32], m_new[i], m_prev[i]) + T.writes(d_new[i], m_smem[row], d_smem[row], m_prev_smem[row]) + for j in range(32): + d_new[i] = d_new[i] + S_smem[row, j] + m_smem[row] = m_new[i] + d_smem[row] = d_new[i] + m_prev_smem[row] = m_prev[i] + T.tvm_storage_sync("shared") + with T.block(""): + T.reads(m_prev_smem[0:32], m_smem[0:32], S_smem[0:32, 0:32], V_smem[0:32, 0:64]) + T.writes(O_local[0:32, 0:64]) + for li_0_lj_0_fused_0_init in T.thread_binding(4, thread="threadIdx.y"): + for li_0_lj_0_fused_1_init in T.thread_binding(32, thread="threadIdx.x"): + for li_1_init, lj_1_init in T.grid(4, 4): + with T.block("O_gemm_init"): + i = T.axis.spatial(32, (li_0_lj_0_fused_0_init * 32 + li_0_lj_0_fused_1_init) // 16 * 4 + li_1_init) + j = T.axis.spatial(64, (li_0_lj_0_fused_0_init * 32 + li_0_lj_0_fused_1_init) % 16 * 4 + lj_1_init) + T.reads() + T.writes(O_local[i, j]) + O_local[i, j] = O_local[i, j] * T.exp2(m_prev_smem[i] - m_smem[i]) + for li_0_lj_0_fused_0 in T.thread_binding(4, thread="threadIdx.y"): + for li_0_lj_0_fused_1 in T.thread_binding(32, thread="threadIdx.x"): + for lk_0, lk_1, li_1, lj_1 in T.grid(4, 8, 4, 4): + with T.block("O_gemm_update"): + i = T.axis.spatial(32, (li_0_lj_0_fused_0 * 32 + li_0_lj_0_fused_1) // 16 * 4 + li_1) + j = T.axis.spatial(64, (li_0_lj_0_fused_0 * 32 + li_0_lj_0_fused_1) % 16 * 4 + lj_1) + k_1 = T.axis.reduce(32, lk_0 * 8 + lk_1) + T.reads(O_local[i, j], m_prev_smem[i], m_smem[i], S_smem[i, k_1], V_smem[k_1, j]) + T.writes(O_local[i, j]) + O_local[i, j] = O_local[i, j] + S_smem[i, k_1] * T.Cast("float32", V_smem[k_1, j]) + for li_0_lj_0_fused_0 in T.thread_binding(4, thread="threadIdx.y"): + for li_0_lj_0_fused_1 in T.thread_binding(32, thread="threadIdx.x"): + for li_1, lj_1 in T.grid(4, 4): + with T.block("O_store"): + i = T.axis.spatial(32, (li_0_lj_0_fused_0 * 32 + li_0_lj_0_fused_1) // 16 * 4 + li_1) + j = T.axis.spatial(64, (li_0_lj_0_fused_0 * 32 + li_0_lj_0_fused_1) % 16 * 4 + lj_1) + T.reads(q_indptr[b_idx:b_idx + 2], O_local[i, j], d_smem[i]) + T.writes(output[q_indptr[b_idx] + (LH_start + i), by, j]) + cur_L: T.int32 = q_indptr[b_idx] + (LH_start + i) + cur_H_qo: T.int32 = by + if cur_L < q_indptr[b_idx + 1]: + output[cur_L, cur_H_qo, j] = T.Cast("float16", O_local[i, j] / d_smem[i]) + for li_0 in range(1): + for li_1 in T.thread_binding(4, thread="threadIdx.y"): + for li_2 in T.thread_binding(32, thread="threadIdx.x"): + with T.block("lse_store"): + i = T.axis.spatial(32, li_0 * 128 + li_1 * 32 + li_2) + T.where((li_0 * 4 + li_1) * 32 + li_2 < 32) + T.reads(q_indptr[b_idx:b_idx + 2], m_smem[i], d_smem[i]) + T.writes(lse[q_indptr[b_idx] + (LH_start + i), by]) + cur_L: T.int32 = q_indptr[b_idx] + (LH_start + i) + cur_H_qo: T.int32 = by + if cur_L < q_indptr[b_idx + 1]: + lse[cur_L, cur_H_qo] = m_smem[i] + T.log2(d_smem[i]) + tile_id[0] = tile_id[0] + 16 + + @T.prim_func + def chunk_lse(var_A: T.handle, var_temperature: T.handle, var_chunked_sum: T.handle, var_chunked_max: T.handle): + T.func_attr({"target": T.target({"host": {"keys": ["arm_cpu", "cpu"], "kind": "llvm", "mtriple": "aarch64-linux-android", "tag": ""}, "keys": ["opencl", "gpu"], "kind": "opencl", "max_function_args": 128, "max_num_threads": 256, "max_shared_memory_per_block": 16384, "max_threads_per_block": 256, "tag": "", "texture_spatial_limit": 16384, "thread_warp_size": 1}), "tir.noalias": T.bool(True)}) + batch_size, vocab_size = T.int64(is_size_var=True), T.int64(is_size_var=True) + A = T.match_buffer(var_A, (batch_size, vocab_size)) + temperature = T.match_buffer(var_temperature, (batch_size,)) + num_chunks = T.int64(is_size_var=True) + chunked_sum = T.match_buffer(var_chunked_sum, (batch_size, num_chunks)) + chunked_max = T.match_buffer(var_chunked_max, (batch_size, num_chunks)) + # with T.block("root"): + A_pad = T.alloc_buffer((batch_size, num_chunks, T.int64(4096))) + temp_max = T.alloc_buffer((batch_size, num_chunks)) + temp_sum = T.alloc_buffer((batch_size, num_chunks)) + for l0, l1, l2 in T.grid(batch_size, num_chunks, T.int64(4096)): + with T.block("pad"): + v0, v1, v2 = T.axis.remap("SSS", [l0, l1, l2]) + T.reads(temperature[v0], A[v0, v1 * T.int64(4096) + v2]) + T.writes(A_pad[v0, v1, v2]) + A_pad[v0, v1, v2] = T.if_then_else(v1 * T.int64(4096) + v2 < vocab_size, T.if_then_else(temperature[v0] > T.float32(1.0000000000000001e-05), A[v0, v1 * T.int64(4096) + v2] / temperature[v0], A[v0, v1 * T.int64(4096) + v2]), T.float32(-340282346638528859811704183484516925440.0)) + for l0, l1, l2 in T.grid(batch_size, num_chunks, T.int64(4096)): + with T.block("max"): + v0, v1, v2 = T.axis.remap("SSR", [l0, l1, l2]) + T.reads(A_pad[v0, v1, v2]) + T.writes(temp_max[v0, v1]) + with T.init(): + temp_max[v0, v1] = T.float32(-340282346638528859811704183484516925440.0) + temp_max[v0, v1] = T.max(temp_max[v0, v1], A_pad[v0, v1, v2]) + for l0, l1, l2 in T.grid(batch_size, num_chunks, T.int64(4096)): + with T.block("sum_exp"): + v0, v1, v2 = T.axis.remap("SSR", [l0, l1, l2]) + T.reads(temperature[v0], A_pad[v0, v1, v2], temp_max[v0, v1]) + T.writes(temp_sum[v0, v1]) + with T.init(): + temp_sum[v0, v1] = T.float32(0.0) + temp_sum[v0, v1] = temp_sum[v0, v1] + T.if_then_else(v1 * T.int64(4096) + v2 < vocab_size, T.Select(temperature[v0] > T.float32(1.0000000000000001e-05), T.exp(A_pad[v0, v1, v2] - temp_max[v0, v1]), T.Cast("float32", A_pad[v0, v1, v2] == temp_max[v0, v1])), T.float32(0.0)) + for l0, l1, l2 in T.grid(batch_size, num_chunks, T.int64(1)): + with T.block("log"): + v0, v1, v2 = T.axis.remap("SSS", [l0, l1, l2]) + T.reads(temperature[v0], temp_sum[v0, v1], temp_max[v0, v1]) + T.writes(chunked_sum[v0, v1], chunked_max[v0, v1]) + chunked_sum[v0, v1] = T.Select(temperature[v0] > T.float32(1.0000000000000001e-05), T.log(temp_sum[v0, v1]), temp_sum[v0, v1]) + chunked_max[v0, v1] = temp_max[v0, v1] + + @T.prim_func + def compact_kv_copy(var_pages: T.handle, var_copy_length_indptr: T.handle, var_copy_src_dst_pos: T.handle, batch_size: T.int32): + T.func_attr({"target": T.target({"host": {"keys": ["arm_cpu", "cpu"], "kind": "llvm", "mtriple": "aarch64-linux-android", "tag": ""}, "keys": ["opencl", "gpu"], "kind": "opencl", "max_function_args": 128, "max_num_threads": 256, "max_shared_memory_per_block": 16384, "max_threads_per_block": 256, "tag": "", "texture_spatial_limit": 16384, "thread_warp_size": 1}), "tir.is_scheduled": 1}) + num_pages = T.int32() + pages = T.match_buffer(var_pages, (num_pages, 2, 12, 16, 64), "float16", offset_factor=1) + copy_length_indptr = T.match_buffer(var_copy_length_indptr, (batch_size + 1,), "int32", offset_factor=1) + total_copy_length = T.int32() + copy_src_dst_pos = T.match_buffer(var_copy_src_dst_pos, (2, total_copy_length), "int32", offset_factor=1) + with T.block("root"): + T.reads() + T.writes() + for bhd_o in T.thread_binding(batch_size * 3, thread="blockIdx.x"): + for bhd_i in T.thread_binding(256, thread="threadIdx.x"): + b: T.int32 = (bhd_o * 256 + bhd_i) // 768 + h: T.int32 = (bhd_o * 256 + bhd_i) // 64 % 12 + d: T.int32 = (bhd_o * 256 + bhd_i) % 64 + if bhd_o * 256 + bhd_i < batch_size * 12 * 64: + for i in range(copy_length_indptr[b + 1] - copy_length_indptr[b]): + src_pos: T.int32 = copy_src_dst_pos[0, copy_length_indptr[b] + i] + dst_pos: T.int32 = copy_src_dst_pos[1, copy_length_indptr[b] + i] + pages[dst_pos // 16, 0, h, dst_pos % 16, d] = pages[src_pos // 16, 0, h, src_pos % 16, d] + pages[dst_pos // 16, 1, h, dst_pos % 16, d] = pages[src_pos // 16, 1, h, src_pos % 16, d] + + @T.prim_func + def copy_single_page(var_pages: T.handle, src_page_id: T.int64, tgt_page_id: T.int64, copy_length: T.int64): + T.func_attr({"target": T.target({"host": {"keys": ["arm_cpu", "cpu"], "kind": "llvm", "mtriple": "aarch64-linux-android", "tag": ""}, "keys": ["opencl", "gpu"], "kind": "opencl", "max_function_args": 128, "max_num_threads": 256, "max_shared_memory_per_block": 16384, "max_threads_per_block": 256, "tag": "", "texture_spatial_limit": 16384, "thread_warp_size": 1}), "tir.is_scheduled": 1}) + num_pages, page_size = T.int32(), T.int64() + pages = T.match_buffer(var_pages, (num_pages, 2, 12, page_size, 64), "float16", offset_factor=1) + # with T.block("root"): + for b in T.thread_binding(copy_length * T.int64(3), thread="blockIdx.x"): + for t in T.thread_binding(256, thread="threadIdx.x"): + with T.block("copy"): + vh = T.axis.spatial(12, T.Cast("int32", (b * T.int64(256) + T.Cast("int64", t)) // (copy_length * T.int64(64)))) + vp = T.axis.spatial(copy_length, (b * T.int64(256) + T.Cast("int64", t)) % (copy_length * T.int64(64)) // T.int64(64)) + vd = T.axis.spatial(64, T.Cast("int32", (b * T.int64(256) + T.Cast("int64", t)) % T.int64(64))) + T.where(b * T.int64(256) + T.Cast("int64", t) < copy_length * T.int64(12) * T.int64(64)) + T.reads(pages[src_page_id, 0:2, vh, vp, vd]) + T.writes(pages[tgt_page_id, 0:2, vh, vp, vd]) + pages[tgt_page_id, 0, vh, vp, vd] = pages[src_page_id, 0, vh, vp, vd] + pages[tgt_page_id, 1, vh, vp, vd] = pages[src_page_id, 1, vh, vp, vd] + + @T.prim_func + def fused_rope(var_qkv: T.handle, var_position_map: T.handle, var_q: T.handle, var_k: T.handle, var_v: T.handle, apply_rope: T.int32): + T.func_attr({"op_pattern": 8, "target": T.target({"host": {"keys": ["arm_cpu", "cpu"], "kind": "llvm", "mtriple": "aarch64-linux-android", "tag": ""}, "keys": ["opencl", "gpu"], "kind": "opencl", "max_function_args": 128, "max_num_threads": 256, "max_shared_memory_per_block": 16384, "max_threads_per_block": 256, "tag": "", "texture_spatial_limit": 16384, "thread_warp_size": 1}), "tir.noalias": T.bool(True)}) + seq_len = T.int32() + qkv = T.match_buffer(var_qkv, (seq_len, 36, 64), "float16") + position_map = T.match_buffer(var_position_map, (seq_len,), "int32", offset_factor=1) + q = T.match_buffer(var_q, (seq_len, 12, 64), "float16") + k = T.match_buffer(var_k, (seq_len, 12, 64), "float16") + v = T.match_buffer(var_v, (seq_len, 12, 64), "float16") + # with T.block("root"): + for iters_0, iters_1, iters_2 in T.grid(seq_len, 36, 64): + with T.block("llama_fused_rope"): + s, h, d = T.axis.remap("SSS", [iters_0, iters_1, iters_2]) + T.reads(position_map[s], qkv[s, h, d - 8:d - 8 + 17]) + T.writes(q[s, h, d], k[s, h - 12, d], v[s, h - 24, d]) + if h < 12: + freq = T.float32() + q[s, h, d] = T.if_then_else(apply_rope > 0 and d < 16, T.Let(T.Cast("float16", T.cos(freq) * T.Cast("float32", qkv[s, h, d]) + T.sin(freq) * T.Cast("float32", T.if_then_else(d < 8, qkv[s, h, d + 8] * T.float16(-1.0), qkv[s, h, d - 8]))), where={freq: T.Cast("float32", position_map[s]) / T.pow(T.float32(10000.0), T.Cast("float32", d * 2 % 16) / T.float32(16.0))}), qkv[s, h, d]) + else: + if h < 24: + freq = T.float32() + k[s, h - 12, d] = T.if_then_else(apply_rope > 0 and d < 16, T.Let(T.Cast("float16", T.cos(freq) * T.Cast("float32", qkv[s, h, d]) + T.sin(freq) * T.Cast("float32", T.if_then_else(d < 8, qkv[s, h, d + 8] * T.float16(-1.0), qkv[s, h, d - 8]))), where={freq: T.Cast("float32", position_map[s]) / T.pow(T.float32(10000.0), T.Cast("float32", d * 2 % 16) / T.float32(16.0))}), qkv[s, h, d]) + else: + v[s, h - 24, d] = qkv[s, h, d] + + @T.prim_func + def gather_probs(var_src: T.handle, var_indices: T.handle, var_dst: T.handle): + T.func_attr({"target": T.target({"host": {"keys": ["arm_cpu", "cpu"], "kind": "llvm", "mtriple": "aarch64-linux-android", "tag": ""}, "keys": ["opencl", "gpu"], "kind": "opencl", "max_function_args": 128, "max_num_threads": 256, "max_shared_memory_per_block": 16384, "max_threads_per_block": 256, "tag": "", "texture_spatial_limit": 16384, "thread_warp_size": 1}), "tir.noalias": T.bool(True)}) + m, n = T.int32(is_size_var=True), T.int32(is_size_var=True) + src = T.match_buffer(var_src, (m, n)) + batch_size = T.int32(is_size_var=True) + indices = T.match_buffer(var_indices, (batch_size,), "int32") + dst = T.match_buffer(var_dst, (batch_size, n)) + # with T.block("root"): + for b, j in T.grid(batch_size, n): + with T.block("gather_2d"): + vb, vj = T.axis.remap("SS", [b, j]) + T.reads(src[indices[vb], vj], indices[vb]) + T.writes(dst[vb, vj]) + dst[vb, vj] = src[indices[vb], vj] + + @T.prim_func(private=True) + def index(var_layer_norm24: T.handle, index: T.Buffer((T.int64(1), T.int64(1), T.int64(768)), "float16")): + T.func_attr({"target": T.target({"keys": ["opencl", "gpu"], "kind": "opencl", "max_function_args": 128, "max_num_threads": 256, "max_shared_memory_per_block": 16384, "max_threads_per_block": 256, "tag": "", "texture_spatial_limit": 16384, "thread_warp_size": 1}), "tir.noalias": T.bool(True)}) + seq_len = T.int64() + layer_norm24 = T.match_buffer(var_layer_norm24, (T.int64(1), seq_len, T.int64(768)), "float16") + # with T.block("root"): + for i, _, k in T.grid(T.int64(1), T.int64(1), T.int64(768)): + with T.block("index"): + v_i, v__, v_k = T.axis.remap("SSS", [i, _, k]) + T.reads(layer_norm24[v_i, seq_len - T.int64(1), v_k]) + T.writes(index[v_i, v__, v_k]) + index[v_i, v__, v_k] = layer_norm24[v_i, seq_len - T.int64(1), v_k] + + @T.prim_func + def merge_state_inplace(v: T.handle, s: T.handle, v_other: T.handle, s_other: T.handle): + T.func_attr({"target": T.target({"host": {"keys": ["arm_cpu", "cpu"], "kind": "llvm", "mtriple": "aarch64-linux-android", "tag": ""}, "keys": ["opencl", "gpu"], "kind": "opencl", "max_function_args": 128, "max_num_threads": 256, "max_shared_memory_per_block": 16384, "max_threads_per_block": 256, "tag": "", "texture_spatial_limit": 16384, "thread_warp_size": 1}), "tir.is_scheduled": 1}) + N, H, D = T.int32(is_size_var=True), T.int32(is_size_var=True), T.int32(is_size_var=True) + V = T.match_buffer(v, (N, H, D), "float16") + S = T.match_buffer(s, (N, H)) + V_other = T.match_buffer(v_other, (N, H, D), "float16") + S_other = T.match_buffer(s_other, (N, H)) + # with T.block("root"): + for bx in T.thread_binding(N, thread="blockIdx.x"): + for by in T.thread_binding(1, thread="blockIdx.y"): + for ty in T.thread_binding(12, thread="threadIdx.y"): + for tx in T.thread_binding(16, thread="threadIdx.x"): + with T.block("merge"): + T.reads(S[bx, ty + by * 12], S_other[bx, ty + by * 12], V[bx, ty + by * 12, tx * 4:tx * 4 + 4], V_other[bx, ty + by * 12, tx * 4:tx * 4 + 4]) + T.writes(V[bx, ty + by * 12, tx * 4:tx * 4 + 4], S[bx, ty + by * 12]) + s_val = T.alloc_buffer((1,), scope="local") + s_other_val = T.alloc_buffer((1,), scope="local") + s_max = T.alloc_buffer((1,), scope="local") + scale = T.alloc_buffer((1,), scope="local") + other_scale = T.alloc_buffer((1,), scope="local") + v_vec = T.alloc_buffer((4,), "float16", scope="local") + v_other_vec = T.alloc_buffer((4,), "float16", scope="local") + s_val[0] = S[bx, ty + by * 12] + s_other_val[0] = S_other[bx, ty + by * 12] + s_max[0] = T.max(s_val[0], s_other_val[0]) + s_val[0] = T.exp2(s_val[0] - s_max[0]) + s_other_val[0] = T.exp2(s_other_val[0] - s_max[0]) + scale[0] = s_val[0] / (s_val[0] + s_other_val[0]) + other_scale[0] = s_other_val[0] / (s_val[0] + s_other_val[0]) + for vec in T.vectorized(4): + v_vec[vec] = V[bx, ty + by * 12, tx * 4 + vec] + for vec in T.vectorized(4): + v_other_vec[vec] = V_other[bx, ty + by * 12, tx * 4 + vec] + for vec in range(4): + v_vec[vec] = T.Cast("float16", T.Cast("float32", v_vec[vec]) * scale[0] + T.Cast("float32", v_other_vec[vec]) * other_scale[0]) + for vec in T.vectorized(4): + V[bx, ty + by * 12, tx * 4 + vec] = v_vec[vec] + S[bx, ty + by * 12] = T.log2(s_val[0] + s_other_val[0]) + s_max[0] + + @T.prim_func + def scatter_probs(var_src: T.handle, var_indices: T.handle, var_dst: T.handle): + T.func_attr({"target": T.target({"host": {"keys": ["arm_cpu", "cpu"], "kind": "llvm", "mtriple": "aarch64-linux-android", "tag": ""}, "keys": ["opencl", "gpu"], "kind": "opencl", "max_function_args": 128, "max_num_threads": 256, "max_shared_memory_per_block": 16384, "max_threads_per_block": 256, "tag": "", "texture_spatial_limit": 16384, "thread_warp_size": 1}), "tir.noalias": T.bool(True)}) + batch_size, n = T.int32(is_size_var=True), T.int32(is_size_var=True) + src = T.match_buffer(var_src, (batch_size, n)) + indices = T.match_buffer(var_indices, (batch_size,), "int32") + m = T.int32(is_size_var=True) + dst = T.match_buffer(var_dst, (m, n)) + # with T.block("root"): + for b, j in T.grid(batch_size, n): + with T.block("scatter_2d"): + vb, vj = T.axis.remap("SS", [b, j]) + T.reads(src[vb, vj], indices[vb]) + T.writes(dst[indices[vb], vj]) + dst[indices[vb], vj] = src[vb, vj] + + @T.prim_func + def softmax_with_chunked_sum(var_A: T.handle, var_temperature: T.handle, var_chunked_sum: T.handle, var_chunked_max: T.handle, var_softmax: T.handle): + T.func_attr({"target": T.target({"host": {"keys": ["arm_cpu", "cpu"], "kind": "llvm", "mtriple": "aarch64-linux-android", "tag": ""}, "keys": ["opencl", "gpu"], "kind": "opencl", "max_function_args": 128, "max_num_threads": 256, "max_shared_memory_per_block": 16384, "max_threads_per_block": 256, "tag": "", "texture_spatial_limit": 16384, "thread_warp_size": 1}), "tir.is_scheduled": 1, "tir.noalias": T.bool(True)}) + batch_size, vocab_size = T.int64(is_size_var=True), T.int64(is_size_var=True) + A = T.match_buffer(var_A, (batch_size, vocab_size)) + temperature = T.match_buffer(var_temperature, (batch_size,)) + num_chunks = T.int64(is_size_var=True) + chunked_sum = T.match_buffer(var_chunked_sum, (batch_size, num_chunks)) + chunked_max = T.match_buffer(var_chunked_max, (batch_size, num_chunks)) + softmax = T.match_buffer(var_softmax, (batch_size, vocab_size)) + # with T.block("root"): + temp_max_shared = T.alloc_buffer((batch_size,), scope="shared") + temp_sum_shared = T.alloc_buffer((batch_size,), scope="shared") + for l0_l1_fused in T.thread_binding(batch_size * num_chunks, thread="blockIdx.x"): + for ax0_1 in T.thread_binding(T.int64(32), thread="threadIdx.x"): + for ax0_0 in T.serial((num_chunks + T.int64(31)) // T.int64(32), annotations={"pragma_auto_unroll_max_step": 64, "pragma_unroll_explicit": 1}): + with T.block("max"): + v0 = T.axis.spatial(batch_size, l0_l1_fused % (num_chunks * batch_size) // num_chunks) + v1 = T.axis.reduce(num_chunks, ax0_0 * T.int64(32) + ax0_1) + T.where(ax0_0 * T.int64(32) + ax0_1 < num_chunks) + T.reads(chunked_max[v0, v1]) + T.writes(temp_max_shared[v0]) + with T.init(): + temp_max_shared[v0] = T.float32(-340282346638528859811704183484516925440.0) + temp_max_shared[v0] = T.max(temp_max_shared[v0], chunked_max[v0, v1]) + for ax0_1 in T.thread_binding(T.int64(32), thread="threadIdx.x"): + for ax0_0 in T.serial((num_chunks + T.int64(31)) // T.int64(32), annotations={"pragma_auto_unroll_max_step": 64, "pragma_unroll_explicit": 1}): + with T.block("sum_exp"): + v0 = T.axis.spatial(batch_size, l0_l1_fused % (num_chunks * batch_size) // num_chunks) + v1 = T.axis.reduce(num_chunks, ax0_0 * T.int64(32) + ax0_1) + T.where(ax0_0 * T.int64(32) + ax0_1 < num_chunks) + T.reads(temperature[v0], chunked_sum[v0, v1], chunked_max[v0, v1], temp_max_shared[v0]) + T.writes(temp_sum_shared[v0]) + with T.init(): + temp_sum_shared[v0] = T.float32(0.0) + temp_sum_shared[v0] = temp_sum_shared[v0] + T.Select(temperature[v0] > T.float32(1.0000000000000001e-05), T.exp(chunked_sum[v0, v1] + chunked_max[v0, v1] - temp_max_shared[v0]), T.Cast("float32", chunked_max[v0, v1] == temp_max_shared[v0]) * chunked_sum[v0, v1]) + for l2_0 in T.serial(T.int64(16), annotations={"pragma_auto_unroll_max_step": 64, "pragma_unroll_explicit": 1}): + for l2_1 in T.thread_binding(T.int64(8), thread="threadIdx.y"): + for l2_2 in T.thread_binding(T.int64(32), thread="threadIdx.x"): + with T.block("log_pad"): + v0 = T.axis.spatial(batch_size, l0_l1_fused % (num_chunks * batch_size) // num_chunks) + v1 = T.axis.spatial(num_chunks, l0_l1_fused % num_chunks) + v2 = T.axis.spatial(T.int64(4096), l2_0 * T.int64(256) + l2_1 * T.int64(32) + l2_2) + T.reads(temperature[v0], A[v0, v1 * T.int64(4096) + v2], temp_sum_shared[v0], temp_max_shared[v0]) + T.writes(softmax[v0, v1 * T.int64(4096) + v2]) + if v1 * T.int64(4096) + v2 < vocab_size: + softmax[v0, v1 * T.int64(4096) + v2] = T.if_then_else(temperature[v0] > T.float32(1.0000000000000001e-05), T.exp(A[v0, v1 * T.int64(4096) + v2] / temperature[v0] - (T.log(temp_sum_shared[v0]) + temp_max_shared[v0])), T.Cast("float32", A[v0, v1 * T.int64(4096) + v2] == temp_max_shared[v0]) / temp_sum_shared[v0]) + + @T.prim_func + def tir_kv_cache_debug_get_kv(var_pages: T.handle, var_position_map: T.handle, var_k_data: T.handle, var_v_data: T.handle, layer_id: T.int64): + T.func_attr({"target": T.target({"host": {"keys": ["arm_cpu", "cpu"], "kind": "llvm", "mtriple": "aarch64-linux-android", "tag": ""}, "keys": ["opencl", "gpu"], "kind": "opencl", "max_function_args": 128, "max_num_threads": 256, "max_shared_memory_per_block": 16384, "max_threads_per_block": 256, "tag": "", "texture_spatial_limit": 16384, "thread_warp_size": 1}), "tir.noalias": T.bool(True)}) + num_pages, page_size = T.int64(), T.int64(is_size_var=True) + pages = T.match_buffer(var_pages, (num_pages, 2, 12, page_size, 64), "float16", offset_factor=1) + seqlen = T.int64(is_size_var=True) + position_map = T.match_buffer(var_position_map, (seqlen,), "int32", offset_factor=1) + k_data = T.match_buffer(var_k_data, (12, seqlen, 12, 64), "float16") + v_data = T.match_buffer(var_v_data, (12, seqlen, 12, 64), "float16") + # with T.block("root"): + for p, h, d in T.grid(seqlen, 12, 64): + with T.block("copy0"): + vp, vh, vd = T.axis.remap("SSS", [p, h, d]) + T.reads(position_map[vp], pages[T.Cast("int64", position_map[vp]) // page_size, 0:2, vh, T.Cast("int64", position_map[vp]) % page_size, vd]) + T.writes(k_data[layer_id, vp, vh, vd], v_data[layer_id, vp, vh, vd]) + position: T.int32 = position_map[vp] + k_data[layer_id, vp, vh, vd] = pages[T.Cast("int64", position) // page_size, 0, vh, T.Cast("int64", position) % page_size, vd] + v_data[layer_id, vp, vh, vd] = pages[T.Cast("int64", position) // page_size, 1, vh, T.Cast("int64", position) % page_size, vd] + + @T.prim_func + def tir_kv_cache_transpose_append(var_pages: T.handle, var_k_data: T.handle, var_v_data: T.handle, var_position_map: T.handle): + T.func_attr({"target": T.target({"host": {"keys": ["arm_cpu", "cpu"], "kind": "llvm", "mtriple": "aarch64-linux-android", "tag": ""}, "keys": ["opencl", "gpu"], "kind": "opencl", "max_function_args": 128, "max_num_threads": 256, "max_shared_memory_per_block": 16384, "max_threads_per_block": 256, "tag": "", "texture_spatial_limit": 16384, "thread_warp_size": 1}), "tir.noalias": T.bool(True)}) + num_pages = T.int64() + pages = T.match_buffer(var_pages, (num_pages, 2, 12, 16, 64), "float16", offset_factor=1) + ntoken = T.int64(is_size_var=True) + k_data = T.match_buffer(var_k_data, (ntoken, 12, 64), "float16") + v_data = T.match_buffer(var_v_data, (ntoken, 12, 64), "float16") + position_map = T.match_buffer(var_position_map, (ntoken,), "int32", offset_factor=1) + # with T.block("root"): + for global_pos, h, f in T.grid(ntoken, 12, 64): + if position_map[global_pos] != -1: + with T.block("k_transpose_append"): + vgpos, vh, vf = T.axis.remap("SSS", [global_pos, h, f]) + T.reads(position_map[vgpos], k_data[vgpos, vh, vf]) + T.writes(pages[position_map[vgpos] // 16, 0, vh, position_map[vgpos] % 16, vf]) + position: T.int32 = position_map[vgpos] + pages[position // 16, 0, vh, position % 16, vf] = k_data[vgpos, vh, vf] + with T.block("v_transpose_append"): + vgpos, vh, vf = T.axis.remap("SSS", [global_pos, h, f]) + T.reads(position_map[vgpos], v_data[vgpos, vh, vf]) + T.writes(pages[position_map[vgpos] // 16, 1, vh, position_map[vgpos] % 16, vf]) + position: T.int32 = position_map[vgpos] + pages[position // 16, 1, vh, position % 16, vf] = v_data[vgpos, vh, vf] + + @T.prim_func + def tree_attn_paged_kv(_0: T.int32, var_q: T.handle, var_q_indptr: T.handle, var_pages: T.handle, var_page_indptr: T.handle, var_page_values: T.handle, var_length_info: T.handle, var_k_rope_pos_offset: T.handle, var_q_rope_position: T.handle, var_output: T.handle, var_lse: T.handle, rotary_mode: T.int32, rope_scale: T.float32, rope_theta: T.float32, attn_score_scaling_factor: T.float32, tree_order_indptr_handle: T.handle, tree_order_handle: T.handle): + T.func_attr({"target": T.target({"host": {"keys": ["arm_cpu", "cpu"], "kind": "llvm", "mtriple": "aarch64-linux-android", "tag": ""}, "keys": ["opencl", "gpu"], "kind": "opencl", "max_function_args": 128, "max_num_threads": 256, "max_shared_memory_per_block": 16384, "max_threads_per_block": 256, "tag": "", "texture_spatial_limit": 16384, "thread_warp_size": 1}), "tir.is_scheduled": 1}) + total_len = T.int32(is_size_var=True) + q = T.match_buffer(var_q, (total_len, 12, 64), "float16") + batch_size = T.int32(is_size_var=True) + q_indptr = T.match_buffer(var_q_indptr, (batch_size + 1,), "int32", offset_factor=1) + max_num_pages = T.int32(is_size_var=True) + pages = T.match_buffer(var_pages, (max_num_pages, 2, 12, 32, 64), "float16") + page_indptr = T.match_buffer(var_page_indptr, (batch_size + 1,), "int32", offset_factor=1) + nnz_pages = T.int32(is_size_var=True) + page_values = T.match_buffer(var_page_values, (nnz_pages,), "int32", offset_factor=1) + length_info = T.match_buffer(var_length_info, (batch_size,), "int32", offset_factor=1) + k_rope_pos_offset = T.match_buffer(var_k_rope_pos_offset, (batch_size,), "int32", offset_factor=1) + q_rope_position = T.match_buffer(var_q_rope_position, (total_len,), "int32", offset_factor=1) + output = T.match_buffer(var_output, (total_len, 12, 64), "float16") + lse = T.match_buffer(var_lse, (total_len, 12)) + tree_order_indptr = T.match_buffer(tree_order_indptr_handle, (batch_size + 1,), "int32", offset_factor=1) + total_tree_order_len = T.int32(is_size_var=True) + tree_order = T.match_buffer(tree_order_handle, (total_tree_order_len, 2), "int32", offset_factor=1) + # with T.block("root"): + assert rotary_mode == 0, "Inline rotary mode is not supported in tree attention." + for lbx in T.thread_binding(16, thread="blockIdx.x"): + for lby in T.thread_binding(12, thread="blockIdx.y"): + for lty in T.thread_binding(4, thread="threadIdx.y"): + for ltx in T.thread_binding(32, thread="threadIdx.x"): + with T.block("attn"): + bx, by, ty, tx = T.axis.remap("SSSS", [lbx, lby, lty, ltx]) + T.reads() + T.writes() + tile_id = T.alloc_buffer((1,), "int32", scope="local") + batch_idx = T.alloc_buffer((1,), "int32", scope="local") + batch_tiles = T.alloc_buffer((1,), "int32", scope="local") + batch_rows = T.alloc_buffer((1,), "int32", scope="local") + iterator = T.alloc_buffer((1,), "int32", scope="local") + kv_chunk_len = T.alloc_buffer((1,), "int32", scope="local") + Q_smem = T.alloc_buffer((32, 64), "float16", scope="shared") + K_smem = T.alloc_buffer((32, 64), "float16", scope="shared") + V_smem = T.alloc_buffer((32, 64), "float16", scope="shared") + S_smem = T.alloc_buffer((32, 32), scope="shared") + S_local = T.alloc_buffer((32, 32), scope="local") + O_local = T.alloc_buffer((32, 64), scope="local") + m_smem = T.alloc_buffer((32,), scope="shared") + m_prev_smem = T.alloc_buffer((32,), scope="shared") + d_smem = T.alloc_buffer((32,), scope="shared") + m_new = T.alloc_buffer((1,), scope="local") + m_prev = T.alloc_buffer((1,), scope="local") + d_new = T.alloc_buffer((1,), scope="local") + tile_id[0] = bx + batch_idx[0] = 0 + batch_rows[0] = q_indptr[1] - q_indptr[0] + batch_tiles[0] = (batch_rows[0] + 32 - 1) // 32 + while T.tvm_thread_invariant(batch_idx[0] < batch_size): + while tile_id[0] >= batch_tiles[0] and batch_idx[0] < batch_size: + tile_id[0] = tile_id[0] - batch_tiles[0] + batch_idx[0] = batch_idx[0] + 1 + if batch_idx[0] < batch_size: + b_idx: T.int32 = batch_idx[0] + batch_rows[0] = q_indptr[b_idx + 1] - q_indptr[b_idx] + batch_tiles[0] = (batch_rows[0] + 32 - 1) // 32 + if T.tvm_thread_invariant(batch_idx[0] < batch_size): + b_idx: T.int32 = batch_idx[0] + LH_start: T.int32 = tile_id[0] * 32 + q_indptr_val: T.int32 = q_indptr[b_idx] + cur_page_indptr_begin: T.int32 = page_indptr[b_idx] + cur_page_indptr_end: T.int32 = page_indptr[b_idx + 1] + kv_chunk_len[0] = T.if_then_else(cur_page_indptr_begin != cur_page_indptr_end, (cur_page_indptr_end - cur_page_indptr_begin - 1) * 32 + length_info[b_idx], 0) + T.tvm_storage_sync("shared") + for i in range(1): + row: T.int32 = i * 32 * 4 + ty * 32 + tx + if row < 32: + m_smem[row] = T.float32(-50000.0) + d_smem[row] = T.float32(1.0) + for li_0_lj_0_fused_0 in T.thread_binding(4, thread="threadIdx.y"): + for li_0_lj_0_fused_1 in T.thread_binding(32, thread="threadIdx.x"): + for li_1, lj_1 in T.grid(4, 4): + with T.block("O_init"): + i = T.axis.spatial(32, (li_0_lj_0_fused_0 * 32 + li_0_lj_0_fused_1) // 16 * 4 + li_1) + j = T.axis.spatial(64, (li_0_lj_0_fused_0 * 32 + li_0_lj_0_fused_1) % 16 * 4 + lj_1) + T.reads() + T.writes(O_local[i, j]) + O_local[i, j] = T.float32(0.0) + T.tvm_storage_sync("shared") + for li_lj_fused_0 in range(4): + for li_lj_fused_1 in T.thread_binding(4, thread="threadIdx.y"): + for li_lj_fused_2 in T.thread_binding(32, thread="threadIdx.x"): + for li_lj_fused_3 in T.vectorized(4): + with T.block("Q_load"): + i = T.axis.spatial(32, (li_lj_fused_0 * 512 + li_lj_fused_1 * 128 + li_lj_fused_2 * 4 + li_lj_fused_3) // 64) + j = T.axis.spatial(64, (li_lj_fused_0 * 512 + li_lj_fused_1 * 128 + li_lj_fused_2 * 4 + li_lj_fused_3) % 64) + T.reads() + T.writes() + cur_L: T.int32 = q_indptr_val + (LH_start + i) + cur_H_qo: T.int32 = by + if cur_L < q_indptr[b_idx + 1]: + freq = T.float32() + Q_smem[i, j] = T.if_then_else(rotary_mode == 1, T.Let(T.Cast("float16", T.cos(freq) * T.Cast("float32", q[cur_L, cur_H_qo, j]) + T.sin(freq) * T.Cast("float32", T.if_then_else(j < 32, q[cur_L, cur_H_qo, j + 32] * T.float16(-1.0), q[cur_L, cur_H_qo, j - 32]))), where={freq: T.Cast("float32", q_rope_position[cur_L]) * rope_scale / T.pow(rope_theta, T.Cast("float32", j * 2 % 64) / T.float32(64.0))}), q[cur_L, cur_H_qo, j]) + else: + Q_smem[i, j] = T.float16(0.0) + T.tvm_storage_sync("shared") + for iterator_1 in range((kv_chunk_len[0] + 31) // 32): + L_kv_start: T.int32 = iterator_1 * 32 + for lz_ly_fused_0 in range(4): + for lz_ly_fused_1 in T.thread_binding(4, thread="threadIdx.y"): + for lz_ly_fused_2 in T.thread_binding(32, thread="threadIdx.x"): + for lz_ly_fused_3 in T.vectorized(4): + with T.block("K_load"): + i = T.axis.spatial(32, (lz_ly_fused_0 * 512 + lz_ly_fused_1 * 128 + lz_ly_fused_2 * 4 + lz_ly_fused_3) // 64) + j = T.axis.spatial(64, (lz_ly_fused_0 * 512 + lz_ly_fused_1 * 128 + lz_ly_fused_2 * 4 + lz_ly_fused_3) % 64) + T.reads() + T.writes() + cur_L: T.int32 = L_kv_start + i + if cur_L < kv_chunk_len[0]: + seq_offset: T.int32 = cur_L + page_no: T.int32 = page_values[cur_page_indptr_begin + seq_offset // 32] + page_offset: T.int32 = seq_offset % 32 + K_smem[i, j] = pages[page_no, 0, by, page_offset, j] + else: + K_smem[i, j] = T.float16(0.0) + T.tvm_storage_sync("shared") + for lz_ly_fused_0 in range(4): + for lz_ly_fused_1 in T.thread_binding(4, thread="threadIdx.y"): + for lz_ly_fused_2 in T.thread_binding(32, thread="threadIdx.x"): + for lz_ly_fused_3 in T.vectorized(4): + with T.block("V_load"): + i = T.axis.spatial(32, (lz_ly_fused_0 * 512 + lz_ly_fused_1 * 128 + lz_ly_fused_2 * 4 + lz_ly_fused_3) // 64) + j = T.axis.spatial(64, (lz_ly_fused_0 * 512 + lz_ly_fused_1 * 128 + lz_ly_fused_2 * 4 + lz_ly_fused_3) % 64) + T.reads() + T.writes() + cur_L: T.int32 = L_kv_start + i + if cur_L < kv_chunk_len[0]: + seq_offset: T.int32 = cur_L + page_no: T.int32 = page_values[cur_page_indptr_begin + seq_offset // 32] + page_offset: T.int32 = seq_offset % 32 + V_smem[i, j] = pages[page_no, 1, by, page_offset, j] + else: + V_smem[i, j] = T.float16(0.0) + T.tvm_storage_sync("shared") + with T.block(""): + T.reads(Q_smem[0:32, 0:64], K_smem[0:32, 0:64]) + T.writes(S_local[0:32, 0:32]) + for li_0_lj_0_fused_0_init in T.thread_binding(4, thread="threadIdx.y"): + for li_0_lj_0_fused_1_init in T.thread_binding(32, thread="threadIdx.x"): + for li_1_init, lj_1_init in T.grid(2, 4): + with T.block("S_gemm_init"): + i = T.axis.spatial(32, (li_0_lj_0_fused_0_init * 32 + li_0_lj_0_fused_1_init) // 8 * 2 + li_1_init) + j = T.axis.spatial(32, (li_0_lj_0_fused_0_init * 32 + li_0_lj_0_fused_1_init) % 8 * 4 + lj_1_init) + T.reads() + T.writes(S_local[i, j]) + S_local[i, j] = T.float32(0.0) + for li_0_lj_0_fused_0 in T.thread_binding(4, thread="threadIdx.y"): + for li_0_lj_0_fused_1 in T.thread_binding(32, thread="threadIdx.x"): + for lk_0, li_1, lj_1, lk_1 in T.grid(8, 2, 4, 8): + with T.block("S_gemm_update"): + i = T.axis.spatial(32, (li_0_lj_0_fused_0 * 32 + li_0_lj_0_fused_1) // 8 * 2 + li_1) + j = T.axis.spatial(32, (li_0_lj_0_fused_0 * 32 + li_0_lj_0_fused_1) % 8 * 4 + lj_1) + k = T.axis.reduce(64, lk_0 * 8 + lk_1) + T.reads(S_local[i, j], Q_smem[i, k], K_smem[j, k]) + T.writes(S_local[i, j]) + S_local[i, j] = S_local[i, j] + T.Cast("float32", Q_smem[i, k]) * T.Cast("float32", K_smem[j, k]) * attn_score_scaling_factor * T.float32(0.18033688011112042) + T.tvm_storage_sync("shared") + for li_0_lj_0_fused_0 in T.thread_binding(4, thread="threadIdx.y"): + for li_0_lj_0_fused_1 in T.thread_binding(32, thread="threadIdx.x"): + for li_1, lj_1 in T.grid(2, 4): + with T.block("S_store"): + i = T.axis.spatial(32, (li_0_lj_0_fused_0 * 32 + li_0_lj_0_fused_1) // 8 * 2 + li_1) + j = T.axis.spatial(32, (li_0_lj_0_fused_0 * 32 + li_0_lj_0_fused_1) % 8 * 4 + lj_1) + T.reads(S_local[i, j]) + T.writes(S_smem[i, j]) + S_smem[i, j] = S_local[i, j] + T.tvm_storage_sync("shared") + for i in range(1): + row: T.int32 = i * 32 * 4 + ty * 32 + tx + if row < 32: + with T.block("update1"): + T.reads(m_smem[row], kv_chunk_len[0], tree_order_indptr[b_idx:b_idx + 2], tree_order[T.min(LH_start + row + tree_order_indptr[b_idx + 1] + q_indptr[b_idx] - q_indptr[b_idx + 1], L_kv_start + tree_order_indptr[b_idx + 1] - kv_chunk_len[0]):T.min(LH_start + row + tree_order_indptr[b_idx + 1] + q_indptr[b_idx] - q_indptr[b_idx + 1], L_kv_start + tree_order_indptr[b_idx + 1] - kv_chunk_len[0]) + (T.max(LH_start + row + tree_order_indptr[b_idx + 1] + q_indptr[b_idx] - q_indptr[b_idx + 1], L_kv_start + tree_order_indptr[b_idx + 1] + 31 - kv_chunk_len[0]) + 1 - T.min(LH_start + row + tree_order_indptr[b_idx + 1] + q_indptr[b_idx] - q_indptr[b_idx + 1], L_kv_start + tree_order_indptr[b_idx + 1] - kv_chunk_len[0])), 0:2], q_indptr[b_idx:b_idx + 2], m_new[i], S_smem[row, 0:32], d_smem[row], m_prev[i]) + T.writes(m_prev[i], m_new[i], d_new[i]) + m_prev[i] = m_smem[row] + m_new[i] = m_smem[row] + row_: T.int32 = LH_start + row + for j in range(32): + if L_kv_start + j < kv_chunk_len[0] and (L_kv_start + j < kv_chunk_len[0] - (tree_order_indptr[b_idx + 1] - tree_order_indptr[b_idx]) or tree_order[tree_order_indptr[b_idx] + (row_ + (tree_order_indptr[b_idx + 1] - tree_order_indptr[b_idx]) - (q_indptr[b_idx + 1] - q_indptr[b_idx])), 0] >= tree_order[tree_order_indptr[b_idx] + (L_kv_start + j - (kv_chunk_len[0] - (tree_order_indptr[b_idx + 1] - tree_order_indptr[b_idx]))), 0] and tree_order[tree_order_indptr[b_idx] + (row_ + (tree_order_indptr[b_idx + 1] - tree_order_indptr[b_idx]) - (q_indptr[b_idx + 1] - q_indptr[b_idx])), 0] < tree_order[tree_order_indptr[b_idx] + (L_kv_start + j - (kv_chunk_len[0] - (tree_order_indptr[b_idx + 1] - tree_order_indptr[b_idx]))), 1]): + m_new[i] = T.max(m_new[i], S_smem[row, j]) + d_new[i] = d_smem[row] * T.exp2(m_prev[i] - m_new[i]) + for i in range(1): + row: T.int32 = i * 32 * 4 + ty * 32 + tx + with T.block("update"): + T.reads(kv_chunk_len[0], tree_order_indptr[b_idx:b_idx + 2], tree_order[T.min(LH_start + row + tree_order_indptr[b_idx + 1] + q_indptr[b_idx] - q_indptr[b_idx + 1], L_kv_start + tree_order_indptr[b_idx + 1] - kv_chunk_len[0]):T.min(LH_start + row + tree_order_indptr[b_idx + 1] + q_indptr[b_idx] - q_indptr[b_idx + 1], L_kv_start + tree_order_indptr[b_idx + 1] - kv_chunk_len[0]) + (T.max(LH_start + row + tree_order_indptr[b_idx + 1] + q_indptr[b_idx] - q_indptr[b_idx + 1], L_kv_start + tree_order_indptr[b_idx + 1] + 31 - kv_chunk_len[0]) + 1 - T.min(LH_start + row + tree_order_indptr[b_idx + 1] + q_indptr[b_idx] - q_indptr[b_idx + 1], L_kv_start + tree_order_indptr[b_idx + 1] - kv_chunk_len[0])), 0:2], q_indptr[b_idx:b_idx + 2], S_smem[row, 0:32], m_new[i]) + T.writes(S_smem[row, 0:32]) + for j in range(32): + if row < 32: + row_: T.int32 = LH_start + row + if L_kv_start + j < kv_chunk_len[0] and (L_kv_start + j < kv_chunk_len[0] - (tree_order_indptr[b_idx + 1] - tree_order_indptr[b_idx]) or tree_order[tree_order_indptr[b_idx] + (row_ + (tree_order_indptr[b_idx + 1] - tree_order_indptr[b_idx]) - (q_indptr[b_idx + 1] - q_indptr[b_idx])), 0] >= tree_order[tree_order_indptr[b_idx] + (L_kv_start + j - (kv_chunk_len[0] - (tree_order_indptr[b_idx + 1] - tree_order_indptr[b_idx]))), 0] and tree_order[tree_order_indptr[b_idx] + (row_ + (tree_order_indptr[b_idx + 1] - tree_order_indptr[b_idx]) - (q_indptr[b_idx + 1] - q_indptr[b_idx])), 0] < tree_order[tree_order_indptr[b_idx] + (L_kv_start + j - (kv_chunk_len[0] - (tree_order_indptr[b_idx + 1] - tree_order_indptr[b_idx]))), 1]): + S_smem[row, j] = T.exp2(S_smem[row, j] - m_new[i]) + else: + S_smem[row, j] = T.exp2(T.float32(-50000.0) - m_new[i]) + for i in range(1): + row: T.int32 = i * 32 * 4 + ty * 32 + tx + if row < 32: + with T.block("update"): + T.reads(d_new[i], S_smem[row, 0:32], m_new[i], m_prev[i]) + T.writes(d_new[i], m_smem[row], d_smem[row], m_prev_smem[row]) + for j in range(32): + d_new[i] = d_new[i] + S_smem[row, j] + m_smem[row] = m_new[i] + d_smem[row] = d_new[i] + m_prev_smem[row] = m_prev[i] + T.tvm_storage_sync("shared") + with T.block(""): + T.reads(m_prev_smem[0:32], m_smem[0:32], S_smem[0:32, 0:32], V_smem[0:32, 0:64]) + T.writes(O_local[0:32, 0:64]) + for li_0_lj_0_fused_0_init in T.thread_binding(4, thread="threadIdx.y"): + for li_0_lj_0_fused_1_init in T.thread_binding(32, thread="threadIdx.x"): + for li_1_init, lj_1_init in T.grid(4, 4): + with T.block("O_gemm_init"): + i = T.axis.spatial(32, (li_0_lj_0_fused_0_init * 32 + li_0_lj_0_fused_1_init) // 16 * 4 + li_1_init) + j = T.axis.spatial(64, (li_0_lj_0_fused_0_init * 32 + li_0_lj_0_fused_1_init) % 16 * 4 + lj_1_init) + T.reads() + T.writes(O_local[i, j]) + O_local[i, j] = O_local[i, j] * T.exp2(m_prev_smem[i] - m_smem[i]) + for li_0_lj_0_fused_0 in T.thread_binding(4, thread="threadIdx.y"): + for li_0_lj_0_fused_1 in T.thread_binding(32, thread="threadIdx.x"): + for lk_0, lk_1, li_1, lj_1 in T.grid(4, 8, 4, 4): + with T.block("O_gemm_update"): + i = T.axis.spatial(32, (li_0_lj_0_fused_0 * 32 + li_0_lj_0_fused_1) // 16 * 4 + li_1) + j = T.axis.spatial(64, (li_0_lj_0_fused_0 * 32 + li_0_lj_0_fused_1) % 16 * 4 + lj_1) + k = T.axis.reduce(32, lk_0 * 8 + lk_1) + T.reads(O_local[i, j], m_prev_smem[i], m_smem[i], S_smem[i, k], V_smem[k, j]) + T.writes(O_local[i, j]) + O_local[i, j] = O_local[i, j] + S_smem[i, k] * T.Cast("float32", V_smem[k, j]) + for li_0_lj_0_fused_0 in T.thread_binding(4, thread="threadIdx.y"): + for li_0_lj_0_fused_1 in T.thread_binding(32, thread="threadIdx.x"): + for li_1, lj_1 in T.grid(4, 4): + with T.block("O_store"): + i = T.axis.spatial(32, (li_0_lj_0_fused_0 * 32 + li_0_lj_0_fused_1) // 16 * 4 + li_1) + j = T.axis.spatial(64, (li_0_lj_0_fused_0 * 32 + li_0_lj_0_fused_1) % 16 * 4 + lj_1) + T.reads(q_indptr[b_idx:b_idx + 2], O_local[i, j], d_smem[i]) + T.writes(output[q_indptr[b_idx] + (LH_start + i), by, j]) + cur_L: T.int32 = q_indptr[b_idx] + (LH_start + i) + cur_H_qo: T.int32 = by + if cur_L < q_indptr[b_idx + 1]: + output[cur_L, cur_H_qo, j] = T.Cast("float16", O_local[i, j] / d_smem[i]) + for li_0 in range(1): + for li_1 in T.thread_binding(4, thread="threadIdx.y"): + for li_2 in T.thread_binding(32, thread="threadIdx.x"): + with T.block("lse_store"): + i = T.axis.spatial(32, li_0 * 128 + li_1 * 32 + li_2) + T.where((li_0 * 4 + li_1) * 32 + li_2 < 32) + T.reads(q_indptr[b_idx:b_idx + 2], m_smem[i], d_smem[i]) + T.writes(lse[q_indptr[b_idx] + (LH_start + i), by]) + cur_L: T.int32 = q_indptr[b_idx] + (LH_start + i) + cur_H_qo: T.int32 = by + if cur_L < q_indptr[b_idx + 1]: + lse[cur_L, cur_H_qo] = m_smem[i] + T.log2(d_smem[i]) + tile_id[0] = tile_id[0] + 16 + + @R.function + def alloc_embedding_tensor() -> R.Tensor((2048, 768), dtype="float16"): + R.func_attr({"relax.memory_plan_dynamic_func_output": True}) + gv: R.Tensor((2048, 768), dtype="float16") = R.builtin.alloc_tensor(R.shape([2048, 768]), R.dtype("float16"), R.prim_value(0), R.str("global")) + return gv + + @R.function + def batch_decode(input_embeds: R.Tensor(("batch_size", 1, 768), dtype="float16"), paged_kv_cache: R.Object, packed_params: R.Tuple(R.Tensor(("vocab_size", 768), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((2304, 768), dtype="float16"), R.Tensor((2304,), dtype="float16"), R.Tensor((768, 768), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((3072, 768), dtype="float16"), R.Tensor((3072,), dtype="float32"), R.Tensor((768, 3072), dtype="float16"), R.Tensor((768,), dtype="float32"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((2304, 768), dtype="float16"), R.Tensor((2304,), dtype="float16"), R.Tensor((768, 768), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((3072, 768), dtype="float16"), R.Tensor((3072,), dtype="float32"), R.Tensor((768, 3072), dtype="float16"), R.Tensor((768,), dtype="float32"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((2304, 768), dtype="float16"), R.Tensor((2304,), dtype="float16"), R.Tensor((768, 768), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((3072, 768), dtype="float16"), R.Tensor((3072,), dtype="float32"), R.Tensor((768, 3072), dtype="float16"), R.Tensor((768,), dtype="float32"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((2304, 768), dtype="float16"), R.Tensor((2304,), dtype="float16"), R.Tensor((768, 768), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((3072, 768), dtype="float16"), R.Tensor((3072,), dtype="float32"), R.Tensor((768, 3072), dtype="float16"), R.Tensor((768,), dtype="float32"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((2304, 768), dtype="float16"), R.Tensor((2304,), dtype="float16"), R.Tensor((768, 768), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((3072, 768), dtype="float16"), R.Tensor((3072,), dtype="float32"), R.Tensor((768, 3072), dtype="float16"), R.Tensor((768,), dtype="float32"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((2304, 768), dtype="float16"), R.Tensor((2304,), dtype="float16"), R.Tensor((768, 768), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((3072, 768), dtype="float16"), R.Tensor((3072,), dtype="float32"), R.Tensor((768, 3072), dtype="float16"), R.Tensor((768,), dtype="float32"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((2304, 768), dtype="float16"), R.Tensor((2304,), dtype="float16"), R.Tensor((768, 768), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((3072, 768), dtype="float16"), R.Tensor((3072,), dtype="float32"), R.Tensor((768, 3072), dtype="float16"), R.Tensor((768,), dtype="float32"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((2304, 768), dtype="float16"), R.Tensor((2304,), dtype="float16"), R.Tensor((768, 768), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((3072, 768), dtype="float16"), R.Tensor((3072,), dtype="float32"), R.Tensor((768, 3072), dtype="float16"), R.Tensor((768,), dtype="float32"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((2304, 768), dtype="float16"), R.Tensor((2304,), dtype="float16"), R.Tensor((768, 768), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((3072, 768), dtype="float16"), R.Tensor((3072,), dtype="float32"), R.Tensor((768, 3072), dtype="float16"), R.Tensor((768,), dtype="float32"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((2304, 768), dtype="float16"), R.Tensor((2304,), dtype="float16"), R.Tensor((768, 768), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((3072, 768), dtype="float16"), R.Tensor((3072,), dtype="float32"), R.Tensor((768, 3072), dtype="float16"), R.Tensor((768,), dtype="float32"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((2304, 768), dtype="float16"), R.Tensor((2304,), dtype="float16"), R.Tensor((768, 768), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((3072, 768), dtype="float16"), R.Tensor((3072,), dtype="float32"), R.Tensor((768, 3072), dtype="float16"), R.Tensor((768,), dtype="float32"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((2304, 768), dtype="float16"), R.Tensor((2304,), dtype="float16"), R.Tensor((768, 768), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((3072, 768), dtype="float16"), R.Tensor((3072,), dtype="float32"), R.Tensor((768, 3072), dtype="float16"), R.Tensor((768,), dtype="float32"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor(("vocab_size", 768), dtype="float16"))) -> R.Tuple(R.Tensor(("batch_size", 1, "vocab_size"), dtype="float32"), R.Object): + batch_size = T.int64() + vocab_size = T.int64() + R.func_attr({"num_input": 2, "pipeline_parallel_stages": 1, "relax.memory_plan_dynamic_func_output": True, "relax.rewrite_cuda_graph.capture_symbolic_vars": ["batch_size"], "tir_non_negative_var": ["vocab_size"], "tir_var_upper_bound": {"batch_size": 128, "seq_len": 2048, "total_seq_len": 2048}}) + with R.dataflow(): + gpt_neox_embed_in_weight4: R.Tensor((vocab_size, 768), dtype="float16") = packed_params[0] + gpt_neox_layers_0_input_layernorm_weight4: R.Tensor((768,), dtype="float16") = packed_params[1] + gpt_neox_layers_0_input_layernorm_bias4: R.Tensor((768,), dtype="float16") = packed_params[2] + gpt_neox_layers_0_post_attention_layernorm_weight4: R.Tensor((768,), dtype="float16") = packed_params[3] + gpt_neox_layers_0_post_attention_layernorm_bias4: R.Tensor((768,), dtype="float16") = packed_params[4] + gpt_neox_layers_0_attention_query_key_value_weight4: R.Tensor((2304, 768), dtype="float16") = packed_params[5] + gpt_neox_layers_0_attention_query_key_value_bias4: R.Tensor((2304,), dtype="float16") = packed_params[6] + gpt_neox_layers_0_attention_dense_weight4: R.Tensor((768, 768), dtype="float16") = packed_params[7] + gpt_neox_layers_0_attention_dense_bias4: R.Tensor((768,), dtype="float16") = packed_params[8] + gpt_neox_layers_0_mlp_dense_h_to_4h_weight4: R.Tensor((3072, 768), dtype="float16") = packed_params[9] + gpt_neox_layers_0_mlp_dense_h_to_4h_bias4: R.Tensor((3072,), dtype="float32") = packed_params[10] + gpt_neox_layers_0_mlp_dense_4h_to_h_weight4: R.Tensor((768, 3072), dtype="float16") = packed_params[11] + gpt_neox_layers_0_mlp_dense_4h_to_h_bias4: R.Tensor((768,), dtype="float32") = packed_params[12] + gpt_neox_layers_1_input_layernorm_weight4: R.Tensor((768,), dtype="float16") = packed_params[13] + gpt_neox_layers_1_input_layernorm_bias4: R.Tensor((768,), dtype="float16") = packed_params[14] + gpt_neox_layers_1_post_attention_layernorm_weight4: R.Tensor((768,), dtype="float16") = packed_params[15] + gpt_neox_layers_1_post_attention_layernorm_bias4: R.Tensor((768,), dtype="float16") = packed_params[16] + gpt_neox_layers_1_attention_query_key_value_weight4: R.Tensor((2304, 768), dtype="float16") = packed_params[17] + gpt_neox_layers_1_attention_query_key_value_bias4: R.Tensor((2304,), dtype="float16") = packed_params[18] + gpt_neox_layers_1_attention_dense_weight4: R.Tensor((768, 768), dtype="float16") = packed_params[19] + gpt_neox_layers_1_attention_dense_bias4: R.Tensor((768,), dtype="float16") = packed_params[20] + gpt_neox_layers_1_mlp_dense_h_to_4h_weight4: R.Tensor((3072, 768), dtype="float16") = packed_params[21] + gpt_neox_layers_1_mlp_dense_h_to_4h_bias4: R.Tensor((3072,), dtype="float32") = packed_params[22] + gpt_neox_layers_1_mlp_dense_4h_to_h_weight4: R.Tensor((768, 3072), dtype="float16") = packed_params[23] + gpt_neox_layers_1_mlp_dense_4h_to_h_bias4: R.Tensor((768,), dtype="float32") = packed_params[24] + gpt_neox_layers_2_input_layernorm_weight4: R.Tensor((768,), dtype="float16") = packed_params[25] + gpt_neox_layers_2_input_layernorm_bias4: R.Tensor((768,), dtype="float16") = packed_params[26] + gpt_neox_layers_2_post_attention_layernorm_weight4: R.Tensor((768,), dtype="float16") = packed_params[27] + gpt_neox_layers_2_post_attention_layernorm_bias4: R.Tensor((768,), dtype="float16") = packed_params[28] + gpt_neox_layers_2_attention_query_key_value_weight4: R.Tensor((2304, 768), dtype="float16") = packed_params[29] + gpt_neox_layers_2_attention_query_key_value_bias4: R.Tensor((2304,), dtype="float16") = packed_params[30] + gpt_neox_layers_2_attention_dense_weight4: R.Tensor((768, 768), dtype="float16") = packed_params[31] + gpt_neox_layers_2_attention_dense_bias4: R.Tensor((768,), dtype="float16") = packed_params[32] + gpt_neox_layers_2_mlp_dense_h_to_4h_weight4: R.Tensor((3072, 768), dtype="float16") = packed_params[33] + gpt_neox_layers_2_mlp_dense_h_to_4h_bias4: R.Tensor((3072,), dtype="float32") = packed_params[34] + gpt_neox_layers_2_mlp_dense_4h_to_h_weight4: R.Tensor((768, 3072), dtype="float16") = packed_params[35] + gpt_neox_layers_2_mlp_dense_4h_to_h_bias4: R.Tensor((768,), dtype="float32") = packed_params[36] + gpt_neox_layers_3_input_layernorm_weight4: R.Tensor((768,), dtype="float16") = packed_params[37] + gpt_neox_layers_3_input_layernorm_bias4: R.Tensor((768,), dtype="float16") = packed_params[38] + gpt_neox_layers_3_post_attention_layernorm_weight4: R.Tensor((768,), dtype="float16") = packed_params[39] + gpt_neox_layers_3_post_attention_layernorm_bias4: R.Tensor((768,), dtype="float16") = packed_params[40] + gpt_neox_layers_3_attention_query_key_value_weight4: R.Tensor((2304, 768), dtype="float16") = packed_params[41] + gpt_neox_layers_3_attention_query_key_value_bias4: R.Tensor((2304,), dtype="float16") = packed_params[42] + gpt_neox_layers_3_attention_dense_weight4: R.Tensor((768, 768), dtype="float16") = packed_params[43] + gpt_neox_layers_3_attention_dense_bias4: R.Tensor((768,), dtype="float16") = packed_params[44] + gpt_neox_layers_3_mlp_dense_h_to_4h_weight4: R.Tensor((3072, 768), dtype="float16") = packed_params[45] + gpt_neox_layers_3_mlp_dense_h_to_4h_bias4: R.Tensor((3072,), dtype="float32") = packed_params[46] + gpt_neox_layers_3_mlp_dense_4h_to_h_weight4: R.Tensor((768, 3072), dtype="float16") = packed_params[47] + gpt_neox_layers_3_mlp_dense_4h_to_h_bias4: R.Tensor((768,), dtype="float32") = packed_params[48] + gpt_neox_layers_4_input_layernorm_weight4: R.Tensor((768,), dtype="float16") = packed_params[49] + gpt_neox_layers_4_input_layernorm_bias4: R.Tensor((768,), dtype="float16") = packed_params[50] + gpt_neox_layers_4_post_attention_layernorm_weight4: R.Tensor((768,), dtype="float16") = packed_params[51] + gpt_neox_layers_4_post_attention_layernorm_bias4: R.Tensor((768,), dtype="float16") = packed_params[52] + gpt_neox_layers_4_attention_query_key_value_weight4: R.Tensor((2304, 768), dtype="float16") = packed_params[53] + gpt_neox_layers_4_attention_query_key_value_bias4: R.Tensor((2304,), dtype="float16") = packed_params[54] + gpt_neox_layers_4_attention_dense_weight4: R.Tensor((768, 768), dtype="float16") = packed_params[55] + gpt_neox_layers_4_attention_dense_bias4: R.Tensor((768,), dtype="float16") = packed_params[56] + gpt_neox_layers_4_mlp_dense_h_to_4h_weight4: R.Tensor((3072, 768), dtype="float16") = packed_params[57] + gpt_neox_layers_4_mlp_dense_h_to_4h_bias4: R.Tensor((3072,), dtype="float32") = packed_params[58] + gpt_neox_layers_4_mlp_dense_4h_to_h_weight4: R.Tensor((768, 3072), dtype="float16") = packed_params[59] + gpt_neox_layers_4_mlp_dense_4h_to_h_bias4: R.Tensor((768,), dtype="float32") = packed_params[60] + gpt_neox_layers_5_input_layernorm_weight4: R.Tensor((768,), dtype="float16") = packed_params[61] + gpt_neox_layers_5_input_layernorm_bias4: R.Tensor((768,), dtype="float16") = packed_params[62] + gpt_neox_layers_5_post_attention_layernorm_weight4: R.Tensor((768,), dtype="float16") = packed_params[63] + gpt_neox_layers_5_post_attention_layernorm_bias4: R.Tensor((768,), dtype="float16") = packed_params[64] + gpt_neox_layers_5_attention_query_key_value_weight4: R.Tensor((2304, 768), dtype="float16") = packed_params[65] + gpt_neox_layers_5_attention_query_key_value_bias4: R.Tensor((2304,), dtype="float16") = packed_params[66] + gpt_neox_layers_5_attention_dense_weight4: R.Tensor((768, 768), dtype="float16") = packed_params[67] + gpt_neox_layers_5_attention_dense_bias4: R.Tensor((768,), dtype="float16") = packed_params[68] + gpt_neox_layers_5_mlp_dense_h_to_4h_weight4: R.Tensor((3072, 768), dtype="float16") = packed_params[69] + gpt_neox_layers_5_mlp_dense_h_to_4h_bias4: R.Tensor((3072,), dtype="float32") = packed_params[70] + gpt_neox_layers_5_mlp_dense_4h_to_h_weight4: R.Tensor((768, 3072), dtype="float16") = packed_params[71] + gpt_neox_layers_5_mlp_dense_4h_to_h_bias4: R.Tensor((768,), dtype="float32") = packed_params[72] + gpt_neox_layers_6_input_layernorm_weight4: R.Tensor((768,), dtype="float16") = packed_params[73] + gpt_neox_layers_6_input_layernorm_bias4: R.Tensor((768,), dtype="float16") = packed_params[74] + gpt_neox_layers_6_post_attention_layernorm_weight4: R.Tensor((768,), dtype="float16") = packed_params[75] + gpt_neox_layers_6_post_attention_layernorm_bias4: R.Tensor((768,), dtype="float16") = packed_params[76] + gpt_neox_layers_6_attention_query_key_value_weight4: R.Tensor((2304, 768), dtype="float16") = packed_params[77] + gpt_neox_layers_6_attention_query_key_value_bias4: R.Tensor((2304,), dtype="float16") = packed_params[78] + gpt_neox_layers_6_attention_dense_weight4: R.Tensor((768, 768), dtype="float16") = packed_params[79] + gpt_neox_layers_6_attention_dense_bias4: R.Tensor((768,), dtype="float16") = packed_params[80] + gpt_neox_layers_6_mlp_dense_h_to_4h_weight4: R.Tensor((3072, 768), dtype="float16") = packed_params[81] + gpt_neox_layers_6_mlp_dense_h_to_4h_bias4: R.Tensor((3072,), dtype="float32") = packed_params[82] + gpt_neox_layers_6_mlp_dense_4h_to_h_weight4: R.Tensor((768, 3072), dtype="float16") = packed_params[83] + gpt_neox_layers_6_mlp_dense_4h_to_h_bias4: R.Tensor((768,), dtype="float32") = packed_params[84] + gpt_neox_layers_7_input_layernorm_weight4: R.Tensor((768,), dtype="float16") = packed_params[85] + gpt_neox_layers_7_input_layernorm_bias4: R.Tensor((768,), dtype="float16") = packed_params[86] + gpt_neox_layers_7_post_attention_layernorm_weight4: R.Tensor((768,), dtype="float16") = packed_params[87] + gpt_neox_layers_7_post_attention_layernorm_bias4: R.Tensor((768,), dtype="float16") = packed_params[88] + gpt_neox_layers_7_attention_query_key_value_weight4: R.Tensor((2304, 768), dtype="float16") = packed_params[89] + gpt_neox_layers_7_attention_query_key_value_bias4: R.Tensor((2304,), dtype="float16") = packed_params[90] + gpt_neox_layers_7_attention_dense_weight4: R.Tensor((768, 768), dtype="float16") = packed_params[91] + gpt_neox_layers_7_attention_dense_bias4: R.Tensor((768,), dtype="float16") = packed_params[92] + gpt_neox_layers_7_mlp_dense_h_to_4h_weight4: R.Tensor((3072, 768), dtype="float16") = packed_params[93] + gpt_neox_layers_7_mlp_dense_h_to_4h_bias4: R.Tensor((3072,), dtype="float32") = packed_params[94] + gpt_neox_layers_7_mlp_dense_4h_to_h_weight4: R.Tensor((768, 3072), dtype="float16") = packed_params[95] + gpt_neox_layers_7_mlp_dense_4h_to_h_bias4: R.Tensor((768,), dtype="float32") = packed_params[96] + gpt_neox_layers_8_input_layernorm_weight4: R.Tensor((768,), dtype="float16") = packed_params[97] + gpt_neox_layers_8_input_layernorm_bias4: R.Tensor((768,), dtype="float16") = packed_params[98] + gpt_neox_layers_8_post_attention_layernorm_weight4: R.Tensor((768,), dtype="float16") = packed_params[99] + gpt_neox_layers_8_post_attention_layernorm_bias4: R.Tensor((768,), dtype="float16") = packed_params[100] + gpt_neox_layers_8_attention_query_key_value_weight4: R.Tensor((2304, 768), dtype="float16") = packed_params[101] + gpt_neox_layers_8_attention_query_key_value_bias4: R.Tensor((2304,), dtype="float16") = packed_params[102] + gpt_neox_layers_8_attention_dense_weight4: R.Tensor((768, 768), dtype="float16") = packed_params[103] + gpt_neox_layers_8_attention_dense_bias4: R.Tensor((768,), dtype="float16") = packed_params[104] + gpt_neox_layers_8_mlp_dense_h_to_4h_weight4: R.Tensor((3072, 768), dtype="float16") = packed_params[105] + gpt_neox_layers_8_mlp_dense_h_to_4h_bias4: R.Tensor((3072,), dtype="float32") = packed_params[106] + gpt_neox_layers_8_mlp_dense_4h_to_h_weight4: R.Tensor((768, 3072), dtype="float16") = packed_params[107] + gpt_neox_layers_8_mlp_dense_4h_to_h_bias4: R.Tensor((768,), dtype="float32") = packed_params[108] + gpt_neox_layers_9_input_layernorm_weight4: R.Tensor((768,), dtype="float16") = packed_params[109] + gpt_neox_layers_9_input_layernorm_bias4: R.Tensor((768,), dtype="float16") = packed_params[110] + gpt_neox_layers_9_post_attention_layernorm_weight4: R.Tensor((768,), dtype="float16") = packed_params[111] + gpt_neox_layers_9_post_attention_layernorm_bias4: R.Tensor((768,), dtype="float16") = packed_params[112] + gpt_neox_layers_9_attention_query_key_value_weight4: R.Tensor((2304, 768), dtype="float16") = packed_params[113] + gpt_neox_layers_9_attention_query_key_value_bias4: R.Tensor((2304,), dtype="float16") = packed_params[114] + gpt_neox_layers_9_attention_dense_weight4: R.Tensor((768, 768), dtype="float16") = packed_params[115] + gpt_neox_layers_9_attention_dense_bias4: R.Tensor((768,), dtype="float16") = packed_params[116] + gpt_neox_layers_9_mlp_dense_h_to_4h_weight4: R.Tensor((3072, 768), dtype="float16") = packed_params[117] + gpt_neox_layers_9_mlp_dense_h_to_4h_bias4: R.Tensor((3072,), dtype="float32") = packed_params[118] + gpt_neox_layers_9_mlp_dense_4h_to_h_weight4: R.Tensor((768, 3072), dtype="float16") = packed_params[119] + gpt_neox_layers_9_mlp_dense_4h_to_h_bias4: R.Tensor((768,), dtype="float32") = packed_params[120] + gpt_neox_layers_10_input_layernorm_weight4: R.Tensor((768,), dtype="float16") = packed_params[121] + gpt_neox_layers_10_input_layernorm_bias4: R.Tensor((768,), dtype="float16") = packed_params[122] + gpt_neox_layers_10_post_attention_layernorm_weight4: R.Tensor((768,), dtype="float16") = packed_params[123] + gpt_neox_layers_10_post_attention_layernorm_bias4: R.Tensor((768,), dtype="float16") = packed_params[124] + gpt_neox_layers_10_attention_query_key_value_weight4: R.Tensor((2304, 768), dtype="float16") = packed_params[125] + gpt_neox_layers_10_attention_query_key_value_bias4: R.Tensor((2304,), dtype="float16") = packed_params[126] + gpt_neox_layers_10_attention_dense_weight4: R.Tensor((768, 768), dtype="float16") = packed_params[127] + gpt_neox_layers_10_attention_dense_bias4: R.Tensor((768,), dtype="float16") = packed_params[128] + gpt_neox_layers_10_mlp_dense_h_to_4h_weight4: R.Tensor((3072, 768), dtype="float16") = packed_params[129] + gpt_neox_layers_10_mlp_dense_h_to_4h_bias4: R.Tensor((3072,), dtype="float32") = packed_params[130] + gpt_neox_layers_10_mlp_dense_4h_to_h_weight4: R.Tensor((768, 3072), dtype="float16") = packed_params[131] + gpt_neox_layers_10_mlp_dense_4h_to_h_bias4: R.Tensor((768,), dtype="float32") = packed_params[132] + gpt_neox_layers_11_input_layernorm_weight4: R.Tensor((768,), dtype="float16") = packed_params[133] + gpt_neox_layers_11_input_layernorm_bias4: R.Tensor((768,), dtype="float16") = packed_params[134] + gpt_neox_layers_11_post_attention_layernorm_weight4: R.Tensor((768,), dtype="float16") = packed_params[135] + gpt_neox_layers_11_post_attention_layernorm_bias4: R.Tensor((768,), dtype="float16") = packed_params[136] + gpt_neox_layers_11_attention_query_key_value_weight4: R.Tensor((2304, 768), dtype="float16") = packed_params[137] + gpt_neox_layers_11_attention_query_key_value_bias4: R.Tensor((2304,), dtype="float16") = packed_params[138] + gpt_neox_layers_11_attention_dense_weight4: R.Tensor((768, 768), dtype="float16") = packed_params[139] + gpt_neox_layers_11_attention_dense_bias4: R.Tensor((768,), dtype="float16") = packed_params[140] + gpt_neox_layers_11_mlp_dense_h_to_4h_weight4: R.Tensor((3072, 768), dtype="float16") = packed_params[141] + gpt_neox_layers_11_mlp_dense_h_to_4h_bias4: R.Tensor((3072,), dtype="float32") = packed_params[142] + gpt_neox_layers_11_mlp_dense_4h_to_h_weight4: R.Tensor((768, 3072), dtype="float16") = packed_params[143] + gpt_neox_layers_11_mlp_dense_4h_to_h_bias4: R.Tensor((768,), dtype="float32") = packed_params[144] + gpt_neox_final_layer_norm_weight4: R.Tensor((768,), dtype="float16") = packed_params[145] + gpt_neox_final_layer_norm_bias4: R.Tensor((768,), dtype="float16") = packed_params[146] + embed_out_weight4: R.Tensor((vocab_size, 768), dtype="float16") = packed_params[147] + layer_norm75: R.Tensor((batch_size, 1, 768), dtype="float16") = R.nn.layer_norm(input_embeds, gpt_neox_layers_0_input_layernorm_weight4, gpt_neox_layers_0_input_layernorm_bias4, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims147: R.Tensor((768, 2304), dtype="float16") = R.permute_dims(gpt_neox_layers_0_attention_query_key_value_weight4, axes=None) + matmul147: R.Tensor((batch_size, 1, 2304), dtype="float16") = R.matmul(layer_norm75, permute_dims147, out_dtype="void") + add216: R.Tensor((batch_size, 1, 2304), dtype="float16") = R.add(matmul147, gpt_neox_layers_0_attention_query_key_value_bias4) + reshape144: R.Tensor((batch_size, 1, 36, 64), dtype="float16") = R.reshape(add216, R.shape([batch_size, 1, 36, 64])) + reshape145: R.Tensor((batch_size, 36, 64), dtype="float16") = R.reshape(reshape144, R.shape([batch_size, 36, 64])) + lv37 = R.call_dps_packed("vm.builtin.attention_kv_cache_attention_with_fused_qkv", (paged_kv_cache, R.prim_value(0), R.prim_value(T.float32(1.0)), reshape145), out_sinfo=R.Tensor((batch_size, 12, 64), dtype="float16")) + reshape146: R.Tensor((batch_size, 1, 12, 64), dtype="float16") = R.reshape(lv37, R.shape([batch_size, 1, 12, 64])) + reshape147: R.Tensor((batch_size, 1, 768), dtype="float16") = R.reshape(reshape146, R.shape([batch_size, 1, 768])) + permute_dims148: R.Tensor((768, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_0_attention_dense_weight4, axes=None) + matmul148: R.Tensor((batch_size, 1, 768), dtype="float16") = R.matmul(reshape147, permute_dims148, out_dtype="void") + add217: R.Tensor((batch_size, 1, 768), dtype="float16") = R.add(matmul148, gpt_neox_layers_0_attention_dense_bias4) + layer_norm76: R.Tensor((batch_size, 1, 768), dtype="float16") = R.nn.layer_norm(input_embeds, gpt_neox_layers_0_post_attention_layernorm_weight4, gpt_neox_layers_0_post_attention_layernorm_bias4, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims149: R.Tensor((768, 3072), dtype="float16") = R.permute_dims(gpt_neox_layers_0_mlp_dense_h_to_4h_weight4, axes=None) + matmul149: R.Tensor((batch_size, 1, 3072), dtype="float32") = R.matmul(layer_norm76, permute_dims149, out_dtype="float32") + add218: R.Tensor((batch_size, 1, 3072), dtype="float32") = R.add(matmul149, gpt_neox_layers_0_mlp_dense_h_to_4h_bias4) + gelu36: R.Tensor((batch_size, 1, 3072), dtype="float32") = R.nn.gelu(add218) + astype75: R.Tensor((batch_size, 1, 3072), dtype="float16") = R.astype(gelu36, dtype="float16") + permute_dims150: R.Tensor((3072, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_0_mlp_dense_4h_to_h_weight4, axes=None) + matmul150: R.Tensor((batch_size, 1, 768), dtype="float32") = R.matmul(astype75, permute_dims150, out_dtype="float32") + add219: R.Tensor((batch_size, 1, 768), dtype="float32") = R.add(matmul150, gpt_neox_layers_0_mlp_dense_4h_to_h_bias4) + astype76: R.Tensor((batch_size, 1, 768), dtype="float16") = R.astype(add219, dtype="float16") + add220: R.Tensor((batch_size, 1, 768), dtype="float16") = R.add(astype76, add217) + add221: R.Tensor((batch_size, 1, 768), dtype="float16") = R.add(add220, input_embeds) + layer_norm77: R.Tensor((batch_size, 1, 768), dtype="float16") = R.nn.layer_norm(add221, gpt_neox_layers_1_input_layernorm_weight4, gpt_neox_layers_1_input_layernorm_bias4, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims151: R.Tensor((768, 2304), dtype="float16") = R.permute_dims(gpt_neox_layers_1_attention_query_key_value_weight4, axes=None) + matmul151: R.Tensor((batch_size, 1, 2304), dtype="float16") = R.matmul(layer_norm77, permute_dims151, out_dtype="void") + add222: R.Tensor((batch_size, 1, 2304), dtype="float16") = R.add(matmul151, gpt_neox_layers_1_attention_query_key_value_bias4) + reshape148: R.Tensor((batch_size, 1, 36, 64), dtype="float16") = R.reshape(add222, R.shape([batch_size, 1, 36, 64])) + reshape149: R.Tensor((batch_size, 36, 64), dtype="float16") = R.reshape(reshape148, R.shape([batch_size, 36, 64])) + lv38 = R.call_dps_packed("vm.builtin.attention_kv_cache_attention_with_fused_qkv", (paged_kv_cache, R.prim_value(1), R.prim_value(T.float32(1.0)), reshape149), out_sinfo=R.Tensor((batch_size, 12, 64), dtype="float16")) + reshape150: R.Tensor((batch_size, 1, 12, 64), dtype="float16") = R.reshape(lv38, R.shape([batch_size, 1, 12, 64])) + reshape151: R.Tensor((batch_size, 1, 768), dtype="float16") = R.reshape(reshape150, R.shape([batch_size, 1, 768])) + permute_dims152: R.Tensor((768, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_1_attention_dense_weight4, axes=None) + matmul152: R.Tensor((batch_size, 1, 768), dtype="float16") = R.matmul(reshape151, permute_dims152, out_dtype="void") + add223: R.Tensor((batch_size, 1, 768), dtype="float16") = R.add(matmul152, gpt_neox_layers_1_attention_dense_bias4) + layer_norm78: R.Tensor((batch_size, 1, 768), dtype="float16") = R.nn.layer_norm(add221, gpt_neox_layers_1_post_attention_layernorm_weight4, gpt_neox_layers_1_post_attention_layernorm_bias4, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims153: R.Tensor((768, 3072), dtype="float16") = R.permute_dims(gpt_neox_layers_1_mlp_dense_h_to_4h_weight4, axes=None) + matmul153: R.Tensor((batch_size, 1, 3072), dtype="float32") = R.matmul(layer_norm78, permute_dims153, out_dtype="float32") + add224: R.Tensor((batch_size, 1, 3072), dtype="float32") = R.add(matmul153, gpt_neox_layers_1_mlp_dense_h_to_4h_bias4) + gelu37: R.Tensor((batch_size, 1, 3072), dtype="float32") = R.nn.gelu(add224) + astype77: R.Tensor((batch_size, 1, 3072), dtype="float16") = R.astype(gelu37, dtype="float16") + permute_dims154: R.Tensor((3072, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_1_mlp_dense_4h_to_h_weight4, axes=None) + matmul154: R.Tensor((batch_size, 1, 768), dtype="float32") = R.matmul(astype77, permute_dims154, out_dtype="float32") + add225: R.Tensor((batch_size, 1, 768), dtype="float32") = R.add(matmul154, gpt_neox_layers_1_mlp_dense_4h_to_h_bias4) + astype78: R.Tensor((batch_size, 1, 768), dtype="float16") = R.astype(add225, dtype="float16") + add226: R.Tensor((batch_size, 1, 768), dtype="float16") = R.add(astype78, add223) + add227: R.Tensor((batch_size, 1, 768), dtype="float16") = R.add(add226, add221) + layer_norm79: R.Tensor((batch_size, 1, 768), dtype="float16") = R.nn.layer_norm(add227, gpt_neox_layers_2_input_layernorm_weight4, gpt_neox_layers_2_input_layernorm_bias4, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims155: R.Tensor((768, 2304), dtype="float16") = R.permute_dims(gpt_neox_layers_2_attention_query_key_value_weight4, axes=None) + matmul155: R.Tensor((batch_size, 1, 2304), dtype="float16") = R.matmul(layer_norm79, permute_dims155, out_dtype="void") + add228: R.Tensor((batch_size, 1, 2304), dtype="float16") = R.add(matmul155, gpt_neox_layers_2_attention_query_key_value_bias4) + reshape152: R.Tensor((batch_size, 1, 36, 64), dtype="float16") = R.reshape(add228, R.shape([batch_size, 1, 36, 64])) + reshape153: R.Tensor((batch_size, 36, 64), dtype="float16") = R.reshape(reshape152, R.shape([batch_size, 36, 64])) + lv39 = R.call_dps_packed("vm.builtin.attention_kv_cache_attention_with_fused_qkv", (paged_kv_cache, R.prim_value(2), R.prim_value(T.float32(1.0)), reshape153), out_sinfo=R.Tensor((batch_size, 12, 64), dtype="float16")) + reshape154: R.Tensor((batch_size, 1, 12, 64), dtype="float16") = R.reshape(lv39, R.shape([batch_size, 1, 12, 64])) + reshape155: R.Tensor((batch_size, 1, 768), dtype="float16") = R.reshape(reshape154, R.shape([batch_size, 1, 768])) + permute_dims156: R.Tensor((768, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_2_attention_dense_weight4, axes=None) + matmul156: R.Tensor((batch_size, 1, 768), dtype="float16") = R.matmul(reshape155, permute_dims156, out_dtype="void") + add229: R.Tensor((batch_size, 1, 768), dtype="float16") = R.add(matmul156, gpt_neox_layers_2_attention_dense_bias4) + layer_norm80: R.Tensor((batch_size, 1, 768), dtype="float16") = R.nn.layer_norm(add227, gpt_neox_layers_2_post_attention_layernorm_weight4, gpt_neox_layers_2_post_attention_layernorm_bias4, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims157: R.Tensor((768, 3072), dtype="float16") = R.permute_dims(gpt_neox_layers_2_mlp_dense_h_to_4h_weight4, axes=None) + matmul157: R.Tensor((batch_size, 1, 3072), dtype="float32") = R.matmul(layer_norm80, permute_dims157, out_dtype="float32") + add230: R.Tensor((batch_size, 1, 3072), dtype="float32") = R.add(matmul157, gpt_neox_layers_2_mlp_dense_h_to_4h_bias4) + gelu38: R.Tensor((batch_size, 1, 3072), dtype="float32") = R.nn.gelu(add230) + astype79: R.Tensor((batch_size, 1, 3072), dtype="float16") = R.astype(gelu38, dtype="float16") + permute_dims158: R.Tensor((3072, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_2_mlp_dense_4h_to_h_weight4, axes=None) + matmul158: R.Tensor((batch_size, 1, 768), dtype="float32") = R.matmul(astype79, permute_dims158, out_dtype="float32") + add231: R.Tensor((batch_size, 1, 768), dtype="float32") = R.add(matmul158, gpt_neox_layers_2_mlp_dense_4h_to_h_bias4) + astype80: R.Tensor((batch_size, 1, 768), dtype="float16") = R.astype(add231, dtype="float16") + add232: R.Tensor((batch_size, 1, 768), dtype="float16") = R.add(astype80, add229) + add233: R.Tensor((batch_size, 1, 768), dtype="float16") = R.add(add232, add227) + layer_norm81: R.Tensor((batch_size, 1, 768), dtype="float16") = R.nn.layer_norm(add233, gpt_neox_layers_3_input_layernorm_weight4, gpt_neox_layers_3_input_layernorm_bias4, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims159: R.Tensor((768, 2304), dtype="float16") = R.permute_dims(gpt_neox_layers_3_attention_query_key_value_weight4, axes=None) + matmul159: R.Tensor((batch_size, 1, 2304), dtype="float16") = R.matmul(layer_norm81, permute_dims159, out_dtype="void") + add234: R.Tensor((batch_size, 1, 2304), dtype="float16") = R.add(matmul159, gpt_neox_layers_3_attention_query_key_value_bias4) + reshape156: R.Tensor((batch_size, 1, 36, 64), dtype="float16") = R.reshape(add234, R.shape([batch_size, 1, 36, 64])) + reshape157: R.Tensor((batch_size, 36, 64), dtype="float16") = R.reshape(reshape156, R.shape([batch_size, 36, 64])) + lv40 = R.call_dps_packed("vm.builtin.attention_kv_cache_attention_with_fused_qkv", (paged_kv_cache, R.prim_value(3), R.prim_value(T.float32(1.0)), reshape157), out_sinfo=R.Tensor((batch_size, 12, 64), dtype="float16")) + reshape158: R.Tensor((batch_size, 1, 12, 64), dtype="float16") = R.reshape(lv40, R.shape([batch_size, 1, 12, 64])) + reshape159: R.Tensor((batch_size, 1, 768), dtype="float16") = R.reshape(reshape158, R.shape([batch_size, 1, 768])) + permute_dims160: R.Tensor((768, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_3_attention_dense_weight4, axes=None) + matmul160: R.Tensor((batch_size, 1, 768), dtype="float16") = R.matmul(reshape159, permute_dims160, out_dtype="void") + add235: R.Tensor((batch_size, 1, 768), dtype="float16") = R.add(matmul160, gpt_neox_layers_3_attention_dense_bias4) + layer_norm82: R.Tensor((batch_size, 1, 768), dtype="float16") = R.nn.layer_norm(add233, gpt_neox_layers_3_post_attention_layernorm_weight4, gpt_neox_layers_3_post_attention_layernorm_bias4, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims161: R.Tensor((768, 3072), dtype="float16") = R.permute_dims(gpt_neox_layers_3_mlp_dense_h_to_4h_weight4, axes=None) + matmul161: R.Tensor((batch_size, 1, 3072), dtype="float32") = R.matmul(layer_norm82, permute_dims161, out_dtype="float32") + add236: R.Tensor((batch_size, 1, 3072), dtype="float32") = R.add(matmul161, gpt_neox_layers_3_mlp_dense_h_to_4h_bias4) + gelu39: R.Tensor((batch_size, 1, 3072), dtype="float32") = R.nn.gelu(add236) + astype81: R.Tensor((batch_size, 1, 3072), dtype="float16") = R.astype(gelu39, dtype="float16") + permute_dims162: R.Tensor((3072, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_3_mlp_dense_4h_to_h_weight4, axes=None) + matmul162: R.Tensor((batch_size, 1, 768), dtype="float32") = R.matmul(astype81, permute_dims162, out_dtype="float32") + add237: R.Tensor((batch_size, 1, 768), dtype="float32") = R.add(matmul162, gpt_neox_layers_3_mlp_dense_4h_to_h_bias4) + astype82: R.Tensor((batch_size, 1, 768), dtype="float16") = R.astype(add237, dtype="float16") + add238: R.Tensor((batch_size, 1, 768), dtype="float16") = R.add(astype82, add235) + add239: R.Tensor((batch_size, 1, 768), dtype="float16") = R.add(add238, add233) + layer_norm83: R.Tensor((batch_size, 1, 768), dtype="float16") = R.nn.layer_norm(add239, gpt_neox_layers_4_input_layernorm_weight4, gpt_neox_layers_4_input_layernorm_bias4, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims163: R.Tensor((768, 2304), dtype="float16") = R.permute_dims(gpt_neox_layers_4_attention_query_key_value_weight4, axes=None) + matmul163: R.Tensor((batch_size, 1, 2304), dtype="float16") = R.matmul(layer_norm83, permute_dims163, out_dtype="void") + add240: R.Tensor((batch_size, 1, 2304), dtype="float16") = R.add(matmul163, gpt_neox_layers_4_attention_query_key_value_bias4) + reshape160: R.Tensor((batch_size, 1, 36, 64), dtype="float16") = R.reshape(add240, R.shape([batch_size, 1, 36, 64])) + reshape161: R.Tensor((batch_size, 36, 64), dtype="float16") = R.reshape(reshape160, R.shape([batch_size, 36, 64])) + lv41 = R.call_dps_packed("vm.builtin.attention_kv_cache_attention_with_fused_qkv", (paged_kv_cache, R.prim_value(4), R.prim_value(T.float32(1.0)), reshape161), out_sinfo=R.Tensor((batch_size, 12, 64), dtype="float16")) + reshape162: R.Tensor((batch_size, 1, 12, 64), dtype="float16") = R.reshape(lv41, R.shape([batch_size, 1, 12, 64])) + reshape163: R.Tensor((batch_size, 1, 768), dtype="float16") = R.reshape(reshape162, R.shape([batch_size, 1, 768])) + permute_dims164: R.Tensor((768, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_4_attention_dense_weight4, axes=None) + matmul164: R.Tensor((batch_size, 1, 768), dtype="float16") = R.matmul(reshape163, permute_dims164, out_dtype="void") + add241: R.Tensor((batch_size, 1, 768), dtype="float16") = R.add(matmul164, gpt_neox_layers_4_attention_dense_bias4) + layer_norm84: R.Tensor((batch_size, 1, 768), dtype="float16") = R.nn.layer_norm(add239, gpt_neox_layers_4_post_attention_layernorm_weight4, gpt_neox_layers_4_post_attention_layernorm_bias4, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims165: R.Tensor((768, 3072), dtype="float16") = R.permute_dims(gpt_neox_layers_4_mlp_dense_h_to_4h_weight4, axes=None) + matmul165: R.Tensor((batch_size, 1, 3072), dtype="float32") = R.matmul(layer_norm84, permute_dims165, out_dtype="float32") + add242: R.Tensor((batch_size, 1, 3072), dtype="float32") = R.add(matmul165, gpt_neox_layers_4_mlp_dense_h_to_4h_bias4) + gelu40: R.Tensor((batch_size, 1, 3072), dtype="float32") = R.nn.gelu(add242) + astype83: R.Tensor((batch_size, 1, 3072), dtype="float16") = R.astype(gelu40, dtype="float16") + permute_dims166: R.Tensor((3072, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_4_mlp_dense_4h_to_h_weight4, axes=None) + matmul166: R.Tensor((batch_size, 1, 768), dtype="float32") = R.matmul(astype83, permute_dims166, out_dtype="float32") + add243: R.Tensor((batch_size, 1, 768), dtype="float32") = R.add(matmul166, gpt_neox_layers_4_mlp_dense_4h_to_h_bias4) + astype84: R.Tensor((batch_size, 1, 768), dtype="float16") = R.astype(add243, dtype="float16") + add244: R.Tensor((batch_size, 1, 768), dtype="float16") = R.add(astype84, add241) + add245: R.Tensor((batch_size, 1, 768), dtype="float16") = R.add(add244, add239) + layer_norm85: R.Tensor((batch_size, 1, 768), dtype="float16") = R.nn.layer_norm(add245, gpt_neox_layers_5_input_layernorm_weight4, gpt_neox_layers_5_input_layernorm_bias4, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims167: R.Tensor((768, 2304), dtype="float16") = R.permute_dims(gpt_neox_layers_5_attention_query_key_value_weight4, axes=None) + matmul167: R.Tensor((batch_size, 1, 2304), dtype="float16") = R.matmul(layer_norm85, permute_dims167, out_dtype="void") + add246: R.Tensor((batch_size, 1, 2304), dtype="float16") = R.add(matmul167, gpt_neox_layers_5_attention_query_key_value_bias4) + reshape164: R.Tensor((batch_size, 1, 36, 64), dtype="float16") = R.reshape(add246, R.shape([batch_size, 1, 36, 64])) + reshape165: R.Tensor((batch_size, 36, 64), dtype="float16") = R.reshape(reshape164, R.shape([batch_size, 36, 64])) + lv42 = R.call_dps_packed("vm.builtin.attention_kv_cache_attention_with_fused_qkv", (paged_kv_cache, R.prim_value(5), R.prim_value(T.float32(1.0)), reshape165), out_sinfo=R.Tensor((batch_size, 12, 64), dtype="float16")) + reshape166: R.Tensor((batch_size, 1, 12, 64), dtype="float16") = R.reshape(lv42, R.shape([batch_size, 1, 12, 64])) + reshape167: R.Tensor((batch_size, 1, 768), dtype="float16") = R.reshape(reshape166, R.shape([batch_size, 1, 768])) + permute_dims168: R.Tensor((768, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_5_attention_dense_weight4, axes=None) + matmul168: R.Tensor((batch_size, 1, 768), dtype="float16") = R.matmul(reshape167, permute_dims168, out_dtype="void") + add247: R.Tensor((batch_size, 1, 768), dtype="float16") = R.add(matmul168, gpt_neox_layers_5_attention_dense_bias4) + layer_norm86: R.Tensor((batch_size, 1, 768), dtype="float16") = R.nn.layer_norm(add245, gpt_neox_layers_5_post_attention_layernorm_weight4, gpt_neox_layers_5_post_attention_layernorm_bias4, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims169: R.Tensor((768, 3072), dtype="float16") = R.permute_dims(gpt_neox_layers_5_mlp_dense_h_to_4h_weight4, axes=None) + matmul169: R.Tensor((batch_size, 1, 3072), dtype="float32") = R.matmul(layer_norm86, permute_dims169, out_dtype="float32") + add248: R.Tensor((batch_size, 1, 3072), dtype="float32") = R.add(matmul169, gpt_neox_layers_5_mlp_dense_h_to_4h_bias4) + gelu41: R.Tensor((batch_size, 1, 3072), dtype="float32") = R.nn.gelu(add248) + astype85: R.Tensor((batch_size, 1, 3072), dtype="float16") = R.astype(gelu41, dtype="float16") + permute_dims170: R.Tensor((3072, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_5_mlp_dense_4h_to_h_weight4, axes=None) + matmul170: R.Tensor((batch_size, 1, 768), dtype="float32") = R.matmul(astype85, permute_dims170, out_dtype="float32") + add249: R.Tensor((batch_size, 1, 768), dtype="float32") = R.add(matmul170, gpt_neox_layers_5_mlp_dense_4h_to_h_bias4) + astype86: R.Tensor((batch_size, 1, 768), dtype="float16") = R.astype(add249, dtype="float16") + add250: R.Tensor((batch_size, 1, 768), dtype="float16") = R.add(astype86, add247) + add251: R.Tensor((batch_size, 1, 768), dtype="float16") = R.add(add250, add245) + layer_norm87: R.Tensor((batch_size, 1, 768), dtype="float16") = R.nn.layer_norm(add251, gpt_neox_layers_6_input_layernorm_weight4, gpt_neox_layers_6_input_layernorm_bias4, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims171: R.Tensor((768, 2304), dtype="float16") = R.permute_dims(gpt_neox_layers_6_attention_query_key_value_weight4, axes=None) + matmul171: R.Tensor((batch_size, 1, 2304), dtype="float16") = R.matmul(layer_norm87, permute_dims171, out_dtype="void") + add252: R.Tensor((batch_size, 1, 2304), dtype="float16") = R.add(matmul171, gpt_neox_layers_6_attention_query_key_value_bias4) + reshape168: R.Tensor((batch_size, 1, 36, 64), dtype="float16") = R.reshape(add252, R.shape([batch_size, 1, 36, 64])) + reshape169: R.Tensor((batch_size, 36, 64), dtype="float16") = R.reshape(reshape168, R.shape([batch_size, 36, 64])) + lv43 = R.call_dps_packed("vm.builtin.attention_kv_cache_attention_with_fused_qkv", (paged_kv_cache, R.prim_value(6), R.prim_value(T.float32(1.0)), reshape169), out_sinfo=R.Tensor((batch_size, 12, 64), dtype="float16")) + reshape170: R.Tensor((batch_size, 1, 12, 64), dtype="float16") = R.reshape(lv43, R.shape([batch_size, 1, 12, 64])) + reshape171: R.Tensor((batch_size, 1, 768), dtype="float16") = R.reshape(reshape170, R.shape([batch_size, 1, 768])) + permute_dims172: R.Tensor((768, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_6_attention_dense_weight4, axes=None) + matmul172: R.Tensor((batch_size, 1, 768), dtype="float16") = R.matmul(reshape171, permute_dims172, out_dtype="void") + add253: R.Tensor((batch_size, 1, 768), dtype="float16") = R.add(matmul172, gpt_neox_layers_6_attention_dense_bias4) + layer_norm88: R.Tensor((batch_size, 1, 768), dtype="float16") = R.nn.layer_norm(add251, gpt_neox_layers_6_post_attention_layernorm_weight4, gpt_neox_layers_6_post_attention_layernorm_bias4, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims173: R.Tensor((768, 3072), dtype="float16") = R.permute_dims(gpt_neox_layers_6_mlp_dense_h_to_4h_weight4, axes=None) + matmul173: R.Tensor((batch_size, 1, 3072), dtype="float32") = R.matmul(layer_norm88, permute_dims173, out_dtype="float32") + add254: R.Tensor((batch_size, 1, 3072), dtype="float32") = R.add(matmul173, gpt_neox_layers_6_mlp_dense_h_to_4h_bias4) + gelu42: R.Tensor((batch_size, 1, 3072), dtype="float32") = R.nn.gelu(add254) + astype87: R.Tensor((batch_size, 1, 3072), dtype="float16") = R.astype(gelu42, dtype="float16") + permute_dims174: R.Tensor((3072, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_6_mlp_dense_4h_to_h_weight4, axes=None) + matmul174: R.Tensor((batch_size, 1, 768), dtype="float32") = R.matmul(astype87, permute_dims174, out_dtype="float32") + add255: R.Tensor((batch_size, 1, 768), dtype="float32") = R.add(matmul174, gpt_neox_layers_6_mlp_dense_4h_to_h_bias4) + astype88: R.Tensor((batch_size, 1, 768), dtype="float16") = R.astype(add255, dtype="float16") + add256: R.Tensor((batch_size, 1, 768), dtype="float16") = R.add(astype88, add253) + add257: R.Tensor((batch_size, 1, 768), dtype="float16") = R.add(add256, add251) + layer_norm89: R.Tensor((batch_size, 1, 768), dtype="float16") = R.nn.layer_norm(add257, gpt_neox_layers_7_input_layernorm_weight4, gpt_neox_layers_7_input_layernorm_bias4, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims175: R.Tensor((768, 2304), dtype="float16") = R.permute_dims(gpt_neox_layers_7_attention_query_key_value_weight4, axes=None) + matmul175: R.Tensor((batch_size, 1, 2304), dtype="float16") = R.matmul(layer_norm89, permute_dims175, out_dtype="void") + add258: R.Tensor((batch_size, 1, 2304), dtype="float16") = R.add(matmul175, gpt_neox_layers_7_attention_query_key_value_bias4) + reshape172: R.Tensor((batch_size, 1, 36, 64), dtype="float16") = R.reshape(add258, R.shape([batch_size, 1, 36, 64])) + reshape173: R.Tensor((batch_size, 36, 64), dtype="float16") = R.reshape(reshape172, R.shape([batch_size, 36, 64])) + lv44 = R.call_dps_packed("vm.builtin.attention_kv_cache_attention_with_fused_qkv", (paged_kv_cache, R.prim_value(7), R.prim_value(T.float32(1.0)), reshape173), out_sinfo=R.Tensor((batch_size, 12, 64), dtype="float16")) + reshape174: R.Tensor((batch_size, 1, 12, 64), dtype="float16") = R.reshape(lv44, R.shape([batch_size, 1, 12, 64])) + reshape175: R.Tensor((batch_size, 1, 768), dtype="float16") = R.reshape(reshape174, R.shape([batch_size, 1, 768])) + permute_dims176: R.Tensor((768, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_7_attention_dense_weight4, axes=None) + matmul176: R.Tensor((batch_size, 1, 768), dtype="float16") = R.matmul(reshape175, permute_dims176, out_dtype="void") + add259: R.Tensor((batch_size, 1, 768), dtype="float16") = R.add(matmul176, gpt_neox_layers_7_attention_dense_bias4) + layer_norm90: R.Tensor((batch_size, 1, 768), dtype="float16") = R.nn.layer_norm(add257, gpt_neox_layers_7_post_attention_layernorm_weight4, gpt_neox_layers_7_post_attention_layernorm_bias4, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims177: R.Tensor((768, 3072), dtype="float16") = R.permute_dims(gpt_neox_layers_7_mlp_dense_h_to_4h_weight4, axes=None) + matmul177: R.Tensor((batch_size, 1, 3072), dtype="float32") = R.matmul(layer_norm90, permute_dims177, out_dtype="float32") + add260: R.Tensor((batch_size, 1, 3072), dtype="float32") = R.add(matmul177, gpt_neox_layers_7_mlp_dense_h_to_4h_bias4) + gelu43: R.Tensor((batch_size, 1, 3072), dtype="float32") = R.nn.gelu(add260) + astype89: R.Tensor((batch_size, 1, 3072), dtype="float16") = R.astype(gelu43, dtype="float16") + permute_dims178: R.Tensor((3072, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_7_mlp_dense_4h_to_h_weight4, axes=None) + matmul178: R.Tensor((batch_size, 1, 768), dtype="float32") = R.matmul(astype89, permute_dims178, out_dtype="float32") + add261: R.Tensor((batch_size, 1, 768), dtype="float32") = R.add(matmul178, gpt_neox_layers_7_mlp_dense_4h_to_h_bias4) + astype90: R.Tensor((batch_size, 1, 768), dtype="float16") = R.astype(add261, dtype="float16") + add262: R.Tensor((batch_size, 1, 768), dtype="float16") = R.add(astype90, add259) + add263: R.Tensor((batch_size, 1, 768), dtype="float16") = R.add(add262, add257) + layer_norm91: R.Tensor((batch_size, 1, 768), dtype="float16") = R.nn.layer_norm(add263, gpt_neox_layers_8_input_layernorm_weight4, gpt_neox_layers_8_input_layernorm_bias4, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims179: R.Tensor((768, 2304), dtype="float16") = R.permute_dims(gpt_neox_layers_8_attention_query_key_value_weight4, axes=None) + matmul179: R.Tensor((batch_size, 1, 2304), dtype="float16") = R.matmul(layer_norm91, permute_dims179, out_dtype="void") + add264: R.Tensor((batch_size, 1, 2304), dtype="float16") = R.add(matmul179, gpt_neox_layers_8_attention_query_key_value_bias4) + reshape176: R.Tensor((batch_size, 1, 36, 64), dtype="float16") = R.reshape(add264, R.shape([batch_size, 1, 36, 64])) + reshape177: R.Tensor((batch_size, 36, 64), dtype="float16") = R.reshape(reshape176, R.shape([batch_size, 36, 64])) + lv45 = R.call_dps_packed("vm.builtin.attention_kv_cache_attention_with_fused_qkv", (paged_kv_cache, R.prim_value(8), R.prim_value(T.float32(1.0)), reshape177), out_sinfo=R.Tensor((batch_size, 12, 64), dtype="float16")) + reshape178: R.Tensor((batch_size, 1, 12, 64), dtype="float16") = R.reshape(lv45, R.shape([batch_size, 1, 12, 64])) + reshape179: R.Tensor((batch_size, 1, 768), dtype="float16") = R.reshape(reshape178, R.shape([batch_size, 1, 768])) + permute_dims180: R.Tensor((768, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_8_attention_dense_weight4, axes=None) + matmul180: R.Tensor((batch_size, 1, 768), dtype="float16") = R.matmul(reshape179, permute_dims180, out_dtype="void") + add265: R.Tensor((batch_size, 1, 768), dtype="float16") = R.add(matmul180, gpt_neox_layers_8_attention_dense_bias4) + layer_norm92: R.Tensor((batch_size, 1, 768), dtype="float16") = R.nn.layer_norm(add263, gpt_neox_layers_8_post_attention_layernorm_weight4, gpt_neox_layers_8_post_attention_layernorm_bias4, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims181: R.Tensor((768, 3072), dtype="float16") = R.permute_dims(gpt_neox_layers_8_mlp_dense_h_to_4h_weight4, axes=None) + matmul181: R.Tensor((batch_size, 1, 3072), dtype="float32") = R.matmul(layer_norm92, permute_dims181, out_dtype="float32") + add266: R.Tensor((batch_size, 1, 3072), dtype="float32") = R.add(matmul181, gpt_neox_layers_8_mlp_dense_h_to_4h_bias4) + gelu44: R.Tensor((batch_size, 1, 3072), dtype="float32") = R.nn.gelu(add266) + astype91: R.Tensor((batch_size, 1, 3072), dtype="float16") = R.astype(gelu44, dtype="float16") + permute_dims182: R.Tensor((3072, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_8_mlp_dense_4h_to_h_weight4, axes=None) + matmul182: R.Tensor((batch_size, 1, 768), dtype="float32") = R.matmul(astype91, permute_dims182, out_dtype="float32") + add267: R.Tensor((batch_size, 1, 768), dtype="float32") = R.add(matmul182, gpt_neox_layers_8_mlp_dense_4h_to_h_bias4) + astype92: R.Tensor((batch_size, 1, 768), dtype="float16") = R.astype(add267, dtype="float16") + add268: R.Tensor((batch_size, 1, 768), dtype="float16") = R.add(astype92, add265) + add269: R.Tensor((batch_size, 1, 768), dtype="float16") = R.add(add268, add263) + layer_norm93: R.Tensor((batch_size, 1, 768), dtype="float16") = R.nn.layer_norm(add269, gpt_neox_layers_9_input_layernorm_weight4, gpt_neox_layers_9_input_layernorm_bias4, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims183: R.Tensor((768, 2304), dtype="float16") = R.permute_dims(gpt_neox_layers_9_attention_query_key_value_weight4, axes=None) + matmul183: R.Tensor((batch_size, 1, 2304), dtype="float16") = R.matmul(layer_norm93, permute_dims183, out_dtype="void") + add270: R.Tensor((batch_size, 1, 2304), dtype="float16") = R.add(matmul183, gpt_neox_layers_9_attention_query_key_value_bias4) + reshape180: R.Tensor((batch_size, 1, 36, 64), dtype="float16") = R.reshape(add270, R.shape([batch_size, 1, 36, 64])) + reshape181: R.Tensor((batch_size, 36, 64), dtype="float16") = R.reshape(reshape180, R.shape([batch_size, 36, 64])) + lv46 = R.call_dps_packed("vm.builtin.attention_kv_cache_attention_with_fused_qkv", (paged_kv_cache, R.prim_value(9), R.prim_value(T.float32(1.0)), reshape181), out_sinfo=R.Tensor((batch_size, 12, 64), dtype="float16")) + reshape182: R.Tensor((batch_size, 1, 12, 64), dtype="float16") = R.reshape(lv46, R.shape([batch_size, 1, 12, 64])) + reshape183: R.Tensor((batch_size, 1, 768), dtype="float16") = R.reshape(reshape182, R.shape([batch_size, 1, 768])) + permute_dims184: R.Tensor((768, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_9_attention_dense_weight4, axes=None) + matmul184: R.Tensor((batch_size, 1, 768), dtype="float16") = R.matmul(reshape183, permute_dims184, out_dtype="void") + add271: R.Tensor((batch_size, 1, 768), dtype="float16") = R.add(matmul184, gpt_neox_layers_9_attention_dense_bias4) + layer_norm94: R.Tensor((batch_size, 1, 768), dtype="float16") = R.nn.layer_norm(add269, gpt_neox_layers_9_post_attention_layernorm_weight4, gpt_neox_layers_9_post_attention_layernorm_bias4, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims185: R.Tensor((768, 3072), dtype="float16") = R.permute_dims(gpt_neox_layers_9_mlp_dense_h_to_4h_weight4, axes=None) + matmul185: R.Tensor((batch_size, 1, 3072), dtype="float32") = R.matmul(layer_norm94, permute_dims185, out_dtype="float32") + add272: R.Tensor((batch_size, 1, 3072), dtype="float32") = R.add(matmul185, gpt_neox_layers_9_mlp_dense_h_to_4h_bias4) + gelu45: R.Tensor((batch_size, 1, 3072), dtype="float32") = R.nn.gelu(add272) + astype93: R.Tensor((batch_size, 1, 3072), dtype="float16") = R.astype(gelu45, dtype="float16") + permute_dims186: R.Tensor((3072, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_9_mlp_dense_4h_to_h_weight4, axes=None) + matmul186: R.Tensor((batch_size, 1, 768), dtype="float32") = R.matmul(astype93, permute_dims186, out_dtype="float32") + add273: R.Tensor((batch_size, 1, 768), dtype="float32") = R.add(matmul186, gpt_neox_layers_9_mlp_dense_4h_to_h_bias4) + astype94: R.Tensor((batch_size, 1, 768), dtype="float16") = R.astype(add273, dtype="float16") + add274: R.Tensor((batch_size, 1, 768), dtype="float16") = R.add(astype94, add271) + add275: R.Tensor((batch_size, 1, 768), dtype="float16") = R.add(add274, add269) + layer_norm95: R.Tensor((batch_size, 1, 768), dtype="float16") = R.nn.layer_norm(add275, gpt_neox_layers_10_input_layernorm_weight4, gpt_neox_layers_10_input_layernorm_bias4, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims187: R.Tensor((768, 2304), dtype="float16") = R.permute_dims(gpt_neox_layers_10_attention_query_key_value_weight4, axes=None) + matmul187: R.Tensor((batch_size, 1, 2304), dtype="float16") = R.matmul(layer_norm95, permute_dims187, out_dtype="void") + add276: R.Tensor((batch_size, 1, 2304), dtype="float16") = R.add(matmul187, gpt_neox_layers_10_attention_query_key_value_bias4) + reshape184: R.Tensor((batch_size, 1, 36, 64), dtype="float16") = R.reshape(add276, R.shape([batch_size, 1, 36, 64])) + reshape185: R.Tensor((batch_size, 36, 64), dtype="float16") = R.reshape(reshape184, R.shape([batch_size, 36, 64])) + lv47 = R.call_dps_packed("vm.builtin.attention_kv_cache_attention_with_fused_qkv", (paged_kv_cache, R.prim_value(10), R.prim_value(T.float32(1.0)), reshape185), out_sinfo=R.Tensor((batch_size, 12, 64), dtype="float16")) + reshape186: R.Tensor((batch_size, 1, 12, 64), dtype="float16") = R.reshape(lv47, R.shape([batch_size, 1, 12, 64])) + reshape187: R.Tensor((batch_size, 1, 768), dtype="float16") = R.reshape(reshape186, R.shape([batch_size, 1, 768])) + permute_dims188: R.Tensor((768, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_10_attention_dense_weight4, axes=None) + matmul188: R.Tensor((batch_size, 1, 768), dtype="float16") = R.matmul(reshape187, permute_dims188, out_dtype="void") + add277: R.Tensor((batch_size, 1, 768), dtype="float16") = R.add(matmul188, gpt_neox_layers_10_attention_dense_bias4) + layer_norm96: R.Tensor((batch_size, 1, 768), dtype="float16") = R.nn.layer_norm(add275, gpt_neox_layers_10_post_attention_layernorm_weight4, gpt_neox_layers_10_post_attention_layernorm_bias4, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims189: R.Tensor((768, 3072), dtype="float16") = R.permute_dims(gpt_neox_layers_10_mlp_dense_h_to_4h_weight4, axes=None) + matmul189: R.Tensor((batch_size, 1, 3072), dtype="float32") = R.matmul(layer_norm96, permute_dims189, out_dtype="float32") + add278: R.Tensor((batch_size, 1, 3072), dtype="float32") = R.add(matmul189, gpt_neox_layers_10_mlp_dense_h_to_4h_bias4) + gelu46: R.Tensor((batch_size, 1, 3072), dtype="float32") = R.nn.gelu(add278) + astype95: R.Tensor((batch_size, 1, 3072), dtype="float16") = R.astype(gelu46, dtype="float16") + permute_dims190: R.Tensor((3072, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_10_mlp_dense_4h_to_h_weight4, axes=None) + matmul190: R.Tensor((batch_size, 1, 768), dtype="float32") = R.matmul(astype95, permute_dims190, out_dtype="float32") + add279: R.Tensor((batch_size, 1, 768), dtype="float32") = R.add(matmul190, gpt_neox_layers_10_mlp_dense_4h_to_h_bias4) + astype96: R.Tensor((batch_size, 1, 768), dtype="float16") = R.astype(add279, dtype="float16") + add280: R.Tensor((batch_size, 1, 768), dtype="float16") = R.add(astype96, add277) + add281: R.Tensor((batch_size, 1, 768), dtype="float16") = R.add(add280, add275) + layer_norm97: R.Tensor((batch_size, 1, 768), dtype="float16") = R.nn.layer_norm(add281, gpt_neox_layers_11_input_layernorm_weight4, gpt_neox_layers_11_input_layernorm_bias4, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims191: R.Tensor((768, 2304), dtype="float16") = R.permute_dims(gpt_neox_layers_11_attention_query_key_value_weight4, axes=None) + matmul191: R.Tensor((batch_size, 1, 2304), dtype="float16") = R.matmul(layer_norm97, permute_dims191, out_dtype="void") + add282: R.Tensor((batch_size, 1, 2304), dtype="float16") = R.add(matmul191, gpt_neox_layers_11_attention_query_key_value_bias4) + reshape188: R.Tensor((batch_size, 1, 36, 64), dtype="float16") = R.reshape(add282, R.shape([batch_size, 1, 36, 64])) + reshape189: R.Tensor((batch_size, 36, 64), dtype="float16") = R.reshape(reshape188, R.shape([batch_size, 36, 64])) + lv48 = R.call_dps_packed("vm.builtin.attention_kv_cache_attention_with_fused_qkv", (paged_kv_cache, R.prim_value(11), R.prim_value(T.float32(1.0)), reshape189), out_sinfo=R.Tensor((batch_size, 12, 64), dtype="float16")) + reshape190: R.Tensor((batch_size, 1, 12, 64), dtype="float16") = R.reshape(lv48, R.shape([batch_size, 1, 12, 64])) + reshape191: R.Tensor((batch_size, 1, 768), dtype="float16") = R.reshape(reshape190, R.shape([batch_size, 1, 768])) + permute_dims192: R.Tensor((768, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_11_attention_dense_weight4, axes=None) + matmul192: R.Tensor((batch_size, 1, 768), dtype="float16") = R.matmul(reshape191, permute_dims192, out_dtype="void") + add283: R.Tensor((batch_size, 1, 768), dtype="float16") = R.add(matmul192, gpt_neox_layers_11_attention_dense_bias4) + layer_norm98: R.Tensor((batch_size, 1, 768), dtype="float16") = R.nn.layer_norm(add281, gpt_neox_layers_11_post_attention_layernorm_weight4, gpt_neox_layers_11_post_attention_layernorm_bias4, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims193: R.Tensor((768, 3072), dtype="float16") = R.permute_dims(gpt_neox_layers_11_mlp_dense_h_to_4h_weight4, axes=None) + matmul193: R.Tensor((batch_size, 1, 3072), dtype="float32") = R.matmul(layer_norm98, permute_dims193, out_dtype="float32") + add284: R.Tensor((batch_size, 1, 3072), dtype="float32") = R.add(matmul193, gpt_neox_layers_11_mlp_dense_h_to_4h_bias4) + gelu47: R.Tensor((batch_size, 1, 3072), dtype="float32") = R.nn.gelu(add284) + astype97: R.Tensor((batch_size, 1, 3072), dtype="float16") = R.astype(gelu47, dtype="float16") + permute_dims194: R.Tensor((3072, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_11_mlp_dense_4h_to_h_weight4, axes=None) + matmul194: R.Tensor((batch_size, 1, 768), dtype="float32") = R.matmul(astype97, permute_dims194, out_dtype="float32") + add285: R.Tensor((batch_size, 1, 768), dtype="float32") = R.add(matmul194, gpt_neox_layers_11_mlp_dense_4h_to_h_bias4) + astype98: R.Tensor((batch_size, 1, 768), dtype="float16") = R.astype(add285, dtype="float16") + add286: R.Tensor((batch_size, 1, 768), dtype="float16") = R.add(astype98, add283) + add287: R.Tensor((batch_size, 1, 768), dtype="float16") = R.add(add286, add281) + layer_norm99: R.Tensor((batch_size, 1, 768), dtype="float16") = R.nn.layer_norm(add287, gpt_neox_final_layer_norm_weight4, gpt_neox_final_layer_norm_bias4, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims195: R.Tensor((768, vocab_size), dtype="float16") = R.permute_dims(embed_out_weight4, axes=None) + matmul195: R.Tensor((batch_size, 1, vocab_size), dtype="float16") = R.matmul(layer_norm99, permute_dims195, out_dtype="void") + astype99: R.Tensor((batch_size, 1, vocab_size), dtype="float32") = R.astype(matmul195, dtype="float32") + gv4: R.Tuple(R.Tensor((batch_size, 1, vocab_size), dtype="float32"), R.Object) = astype99, paged_kv_cache + R.output(gv4) + return gv4 + + @R.function + def batch_prefill(input_embeds: R.Tensor((1, "seq_len", 768), dtype="float16"), logit_positions: R.Tensor(("batch_size",), dtype="int32"), paged_kv_cache: R.Object, packed_params: R.Tuple(R.Tensor(("vocab_size", 768), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((2304, 768), dtype="float16"), R.Tensor((2304,), dtype="float16"), R.Tensor((768, 768), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((3072, 768), dtype="float16"), R.Tensor((3072,), dtype="float32"), R.Tensor((768, 3072), dtype="float16"), R.Tensor((768,), dtype="float32"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((2304, 768), dtype="float16"), R.Tensor((2304,), dtype="float16"), R.Tensor((768, 768), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((3072, 768), dtype="float16"), R.Tensor((3072,), dtype="float32"), R.Tensor((768, 3072), dtype="float16"), R.Tensor((768,), dtype="float32"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((2304, 768), dtype="float16"), R.Tensor((2304,), dtype="float16"), R.Tensor((768, 768), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((3072, 768), dtype="float16"), R.Tensor((3072,), dtype="float32"), R.Tensor((768, 3072), dtype="float16"), R.Tensor((768,), dtype="float32"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((2304, 768), dtype="float16"), R.Tensor((2304,), dtype="float16"), R.Tensor((768, 768), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((3072, 768), dtype="float16"), R.Tensor((3072,), dtype="float32"), R.Tensor((768, 3072), dtype="float16"), R.Tensor((768,), dtype="float32"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((2304, 768), dtype="float16"), R.Tensor((2304,), dtype="float16"), R.Tensor((768, 768), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((3072, 768), dtype="float16"), R.Tensor((3072,), dtype="float32"), R.Tensor((768, 3072), dtype="float16"), R.Tensor((768,), dtype="float32"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((2304, 768), dtype="float16"), R.Tensor((2304,), dtype="float16"), R.Tensor((768, 768), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((3072, 768), dtype="float16"), R.Tensor((3072,), dtype="float32"), R.Tensor((768, 3072), dtype="float16"), R.Tensor((768,), dtype="float32"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((2304, 768), dtype="float16"), R.Tensor((2304,), dtype="float16"), R.Tensor((768, 768), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((3072, 768), dtype="float16"), R.Tensor((3072,), dtype="float32"), R.Tensor((768, 3072), dtype="float16"), R.Tensor((768,), dtype="float32"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((2304, 768), dtype="float16"), R.Tensor((2304,), dtype="float16"), R.Tensor((768, 768), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((3072, 768), dtype="float16"), R.Tensor((3072,), dtype="float32"), R.Tensor((768, 3072), dtype="float16"), R.Tensor((768,), dtype="float32"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((2304, 768), dtype="float16"), R.Tensor((2304,), dtype="float16"), R.Tensor((768, 768), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((3072, 768), dtype="float16"), R.Tensor((3072,), dtype="float32"), R.Tensor((768, 3072), dtype="float16"), R.Tensor((768,), dtype="float32"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((2304, 768), dtype="float16"), R.Tensor((2304,), dtype="float16"), R.Tensor((768, 768), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((3072, 768), dtype="float16"), R.Tensor((3072,), dtype="float32"), R.Tensor((768, 3072), dtype="float16"), R.Tensor((768,), dtype="float32"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((2304, 768), dtype="float16"), R.Tensor((2304,), dtype="float16"), R.Tensor((768, 768), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((3072, 768), dtype="float16"), R.Tensor((3072,), dtype="float32"), R.Tensor((768, 3072), dtype="float16"), R.Tensor((768,), dtype="float32"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((2304, 768), dtype="float16"), R.Tensor((2304,), dtype="float16"), R.Tensor((768, 768), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((3072, 768), dtype="float16"), R.Tensor((3072,), dtype="float32"), R.Tensor((768, 3072), dtype="float16"), R.Tensor((768,), dtype="float32"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor(("vocab_size", 768), dtype="float16"))) -> R.Tuple(R.Tensor((1, "batch_size", "vocab_size"), dtype="float32"), R.Object): + batch_size = T.int64() + vocab_size = T.int64() + seq_len = T.int64() + R.func_attr({"num_input": 3, "pipeline_parallel_stages": 1, "relax.memory_plan_dynamic_func_output": True, "tir_non_negative_var": ["vocab_size"], "tir_var_upper_bound": {"batch_size": 128, "seq_len": 2048, "total_seq_len": 2048}}) + with R.dataflow(): + gpt_neox_embed_in_weight3: R.Tensor((vocab_size, 768), dtype="float16") = packed_params[0] + gpt_neox_layers_0_input_layernorm_weight3: R.Tensor((768,), dtype="float16") = packed_params[1] + gpt_neox_layers_0_input_layernorm_bias3: R.Tensor((768,), dtype="float16") = packed_params[2] + gpt_neox_layers_0_post_attention_layernorm_weight3: R.Tensor((768,), dtype="float16") = packed_params[3] + gpt_neox_layers_0_post_attention_layernorm_bias3: R.Tensor((768,), dtype="float16") = packed_params[4] + gpt_neox_layers_0_attention_query_key_value_weight3: R.Tensor((2304, 768), dtype="float16") = packed_params[5] + gpt_neox_layers_0_attention_query_key_value_bias3: R.Tensor((2304,), dtype="float16") = packed_params[6] + gpt_neox_layers_0_attention_dense_weight3: R.Tensor((768, 768), dtype="float16") = packed_params[7] + gpt_neox_layers_0_attention_dense_bias3: R.Tensor((768,), dtype="float16") = packed_params[8] + gpt_neox_layers_0_mlp_dense_h_to_4h_weight3: R.Tensor((3072, 768), dtype="float16") = packed_params[9] + gpt_neox_layers_0_mlp_dense_h_to_4h_bias3: R.Tensor((3072,), dtype="float32") = packed_params[10] + gpt_neox_layers_0_mlp_dense_4h_to_h_weight3: R.Tensor((768, 3072), dtype="float16") = packed_params[11] + gpt_neox_layers_0_mlp_dense_4h_to_h_bias3: R.Tensor((768,), dtype="float32") = packed_params[12] + gpt_neox_layers_1_input_layernorm_weight3: R.Tensor((768,), dtype="float16") = packed_params[13] + gpt_neox_layers_1_input_layernorm_bias3: R.Tensor((768,), dtype="float16") = packed_params[14] + gpt_neox_layers_1_post_attention_layernorm_weight3: R.Tensor((768,), dtype="float16") = packed_params[15] + gpt_neox_layers_1_post_attention_layernorm_bias3: R.Tensor((768,), dtype="float16") = packed_params[16] + gpt_neox_layers_1_attention_query_key_value_weight3: R.Tensor((2304, 768), dtype="float16") = packed_params[17] + gpt_neox_layers_1_attention_query_key_value_bias3: R.Tensor((2304,), dtype="float16") = packed_params[18] + gpt_neox_layers_1_attention_dense_weight3: R.Tensor((768, 768), dtype="float16") = packed_params[19] + gpt_neox_layers_1_attention_dense_bias3: R.Tensor((768,), dtype="float16") = packed_params[20] + gpt_neox_layers_1_mlp_dense_h_to_4h_weight3: R.Tensor((3072, 768), dtype="float16") = packed_params[21] + gpt_neox_layers_1_mlp_dense_h_to_4h_bias3: R.Tensor((3072,), dtype="float32") = packed_params[22] + gpt_neox_layers_1_mlp_dense_4h_to_h_weight3: R.Tensor((768, 3072), dtype="float16") = packed_params[23] + gpt_neox_layers_1_mlp_dense_4h_to_h_bias3: R.Tensor((768,), dtype="float32") = packed_params[24] + gpt_neox_layers_2_input_layernorm_weight3: R.Tensor((768,), dtype="float16") = packed_params[25] + gpt_neox_layers_2_input_layernorm_bias3: R.Tensor((768,), dtype="float16") = packed_params[26] + gpt_neox_layers_2_post_attention_layernorm_weight3: R.Tensor((768,), dtype="float16") = packed_params[27] + gpt_neox_layers_2_post_attention_layernorm_bias3: R.Tensor((768,), dtype="float16") = packed_params[28] + gpt_neox_layers_2_attention_query_key_value_weight3: R.Tensor((2304, 768), dtype="float16") = packed_params[29] + gpt_neox_layers_2_attention_query_key_value_bias3: R.Tensor((2304,), dtype="float16") = packed_params[30] + gpt_neox_layers_2_attention_dense_weight3: R.Tensor((768, 768), dtype="float16") = packed_params[31] + gpt_neox_layers_2_attention_dense_bias3: R.Tensor((768,), dtype="float16") = packed_params[32] + gpt_neox_layers_2_mlp_dense_h_to_4h_weight3: R.Tensor((3072, 768), dtype="float16") = packed_params[33] + gpt_neox_layers_2_mlp_dense_h_to_4h_bias3: R.Tensor((3072,), dtype="float32") = packed_params[34] + gpt_neox_layers_2_mlp_dense_4h_to_h_weight3: R.Tensor((768, 3072), dtype="float16") = packed_params[35] + gpt_neox_layers_2_mlp_dense_4h_to_h_bias3: R.Tensor((768,), dtype="float32") = packed_params[36] + gpt_neox_layers_3_input_layernorm_weight3: R.Tensor((768,), dtype="float16") = packed_params[37] + gpt_neox_layers_3_input_layernorm_bias3: R.Tensor((768,), dtype="float16") = packed_params[38] + gpt_neox_layers_3_post_attention_layernorm_weight3: R.Tensor((768,), dtype="float16") = packed_params[39] + gpt_neox_layers_3_post_attention_layernorm_bias3: R.Tensor((768,), dtype="float16") = packed_params[40] + gpt_neox_layers_3_attention_query_key_value_weight3: R.Tensor((2304, 768), dtype="float16") = packed_params[41] + gpt_neox_layers_3_attention_query_key_value_bias3: R.Tensor((2304,), dtype="float16") = packed_params[42] + gpt_neox_layers_3_attention_dense_weight3: R.Tensor((768, 768), dtype="float16") = packed_params[43] + gpt_neox_layers_3_attention_dense_bias3: R.Tensor((768,), dtype="float16") = packed_params[44] + gpt_neox_layers_3_mlp_dense_h_to_4h_weight3: R.Tensor((3072, 768), dtype="float16") = packed_params[45] + gpt_neox_layers_3_mlp_dense_h_to_4h_bias3: R.Tensor((3072,), dtype="float32") = packed_params[46] + gpt_neox_layers_3_mlp_dense_4h_to_h_weight3: R.Tensor((768, 3072), dtype="float16") = packed_params[47] + gpt_neox_layers_3_mlp_dense_4h_to_h_bias3: R.Tensor((768,), dtype="float32") = packed_params[48] + gpt_neox_layers_4_input_layernorm_weight3: R.Tensor((768,), dtype="float16") = packed_params[49] + gpt_neox_layers_4_input_layernorm_bias3: R.Tensor((768,), dtype="float16") = packed_params[50] + gpt_neox_layers_4_post_attention_layernorm_weight3: R.Tensor((768,), dtype="float16") = packed_params[51] + gpt_neox_layers_4_post_attention_layernorm_bias3: R.Tensor((768,), dtype="float16") = packed_params[52] + gpt_neox_layers_4_attention_query_key_value_weight3: R.Tensor((2304, 768), dtype="float16") = packed_params[53] + gpt_neox_layers_4_attention_query_key_value_bias3: R.Tensor((2304,), dtype="float16") = packed_params[54] + gpt_neox_layers_4_attention_dense_weight3: R.Tensor((768, 768), dtype="float16") = packed_params[55] + gpt_neox_layers_4_attention_dense_bias3: R.Tensor((768,), dtype="float16") = packed_params[56] + gpt_neox_layers_4_mlp_dense_h_to_4h_weight3: R.Tensor((3072, 768), dtype="float16") = packed_params[57] + gpt_neox_layers_4_mlp_dense_h_to_4h_bias3: R.Tensor((3072,), dtype="float32") = packed_params[58] + gpt_neox_layers_4_mlp_dense_4h_to_h_weight3: R.Tensor((768, 3072), dtype="float16") = packed_params[59] + gpt_neox_layers_4_mlp_dense_4h_to_h_bias3: R.Tensor((768,), dtype="float32") = packed_params[60] + gpt_neox_layers_5_input_layernorm_weight3: R.Tensor((768,), dtype="float16") = packed_params[61] + gpt_neox_layers_5_input_layernorm_bias3: R.Tensor((768,), dtype="float16") = packed_params[62] + gpt_neox_layers_5_post_attention_layernorm_weight3: R.Tensor((768,), dtype="float16") = packed_params[63] + gpt_neox_layers_5_post_attention_layernorm_bias3: R.Tensor((768,), dtype="float16") = packed_params[64] + gpt_neox_layers_5_attention_query_key_value_weight3: R.Tensor((2304, 768), dtype="float16") = packed_params[65] + gpt_neox_layers_5_attention_query_key_value_bias3: R.Tensor((2304,), dtype="float16") = packed_params[66] + gpt_neox_layers_5_attention_dense_weight3: R.Tensor((768, 768), dtype="float16") = packed_params[67] + gpt_neox_layers_5_attention_dense_bias3: R.Tensor((768,), dtype="float16") = packed_params[68] + gpt_neox_layers_5_mlp_dense_h_to_4h_weight3: R.Tensor((3072, 768), dtype="float16") = packed_params[69] + gpt_neox_layers_5_mlp_dense_h_to_4h_bias3: R.Tensor((3072,), dtype="float32") = packed_params[70] + gpt_neox_layers_5_mlp_dense_4h_to_h_weight3: R.Tensor((768, 3072), dtype="float16") = packed_params[71] + gpt_neox_layers_5_mlp_dense_4h_to_h_bias3: R.Tensor((768,), dtype="float32") = packed_params[72] + gpt_neox_layers_6_input_layernorm_weight3: R.Tensor((768,), dtype="float16") = packed_params[73] + gpt_neox_layers_6_input_layernorm_bias3: R.Tensor((768,), dtype="float16") = packed_params[74] + gpt_neox_layers_6_post_attention_layernorm_weight3: R.Tensor((768,), dtype="float16") = packed_params[75] + gpt_neox_layers_6_post_attention_layernorm_bias3: R.Tensor((768,), dtype="float16") = packed_params[76] + gpt_neox_layers_6_attention_query_key_value_weight3: R.Tensor((2304, 768), dtype="float16") = packed_params[77] + gpt_neox_layers_6_attention_query_key_value_bias3: R.Tensor((2304,), dtype="float16") = packed_params[78] + gpt_neox_layers_6_attention_dense_weight3: R.Tensor((768, 768), dtype="float16") = packed_params[79] + gpt_neox_layers_6_attention_dense_bias3: R.Tensor((768,), dtype="float16") = packed_params[80] + gpt_neox_layers_6_mlp_dense_h_to_4h_weight3: R.Tensor((3072, 768), dtype="float16") = packed_params[81] + gpt_neox_layers_6_mlp_dense_h_to_4h_bias3: R.Tensor((3072,), dtype="float32") = packed_params[82] + gpt_neox_layers_6_mlp_dense_4h_to_h_weight3: R.Tensor((768, 3072), dtype="float16") = packed_params[83] + gpt_neox_layers_6_mlp_dense_4h_to_h_bias3: R.Tensor((768,), dtype="float32") = packed_params[84] + gpt_neox_layers_7_input_layernorm_weight3: R.Tensor((768,), dtype="float16") = packed_params[85] + gpt_neox_layers_7_input_layernorm_bias3: R.Tensor((768,), dtype="float16") = packed_params[86] + gpt_neox_layers_7_post_attention_layernorm_weight3: R.Tensor((768,), dtype="float16") = packed_params[87] + gpt_neox_layers_7_post_attention_layernorm_bias3: R.Tensor((768,), dtype="float16") = packed_params[88] + gpt_neox_layers_7_attention_query_key_value_weight3: R.Tensor((2304, 768), dtype="float16") = packed_params[89] + gpt_neox_layers_7_attention_query_key_value_bias3: R.Tensor((2304,), dtype="float16") = packed_params[90] + gpt_neox_layers_7_attention_dense_weight3: R.Tensor((768, 768), dtype="float16") = packed_params[91] + gpt_neox_layers_7_attention_dense_bias3: R.Tensor((768,), dtype="float16") = packed_params[92] + gpt_neox_layers_7_mlp_dense_h_to_4h_weight3: R.Tensor((3072, 768), dtype="float16") = packed_params[93] + gpt_neox_layers_7_mlp_dense_h_to_4h_bias3: R.Tensor((3072,), dtype="float32") = packed_params[94] + gpt_neox_layers_7_mlp_dense_4h_to_h_weight3: R.Tensor((768, 3072), dtype="float16") = packed_params[95] + gpt_neox_layers_7_mlp_dense_4h_to_h_bias3: R.Tensor((768,), dtype="float32") = packed_params[96] + gpt_neox_layers_8_input_layernorm_weight3: R.Tensor((768,), dtype="float16") = packed_params[97] + gpt_neox_layers_8_input_layernorm_bias3: R.Tensor((768,), dtype="float16") = packed_params[98] + gpt_neox_layers_8_post_attention_layernorm_weight3: R.Tensor((768,), dtype="float16") = packed_params[99] + gpt_neox_layers_8_post_attention_layernorm_bias3: R.Tensor((768,), dtype="float16") = packed_params[100] + gpt_neox_layers_8_attention_query_key_value_weight3: R.Tensor((2304, 768), dtype="float16") = packed_params[101] + gpt_neox_layers_8_attention_query_key_value_bias3: R.Tensor((2304,), dtype="float16") = packed_params[102] + gpt_neox_layers_8_attention_dense_weight3: R.Tensor((768, 768), dtype="float16") = packed_params[103] + gpt_neox_layers_8_attention_dense_bias3: R.Tensor((768,), dtype="float16") = packed_params[104] + gpt_neox_layers_8_mlp_dense_h_to_4h_weight3: R.Tensor((3072, 768), dtype="float16") = packed_params[105] + gpt_neox_layers_8_mlp_dense_h_to_4h_bias3: R.Tensor((3072,), dtype="float32") = packed_params[106] + gpt_neox_layers_8_mlp_dense_4h_to_h_weight3: R.Tensor((768, 3072), dtype="float16") = packed_params[107] + gpt_neox_layers_8_mlp_dense_4h_to_h_bias3: R.Tensor((768,), dtype="float32") = packed_params[108] + gpt_neox_layers_9_input_layernorm_weight3: R.Tensor((768,), dtype="float16") = packed_params[109] + gpt_neox_layers_9_input_layernorm_bias3: R.Tensor((768,), dtype="float16") = packed_params[110] + gpt_neox_layers_9_post_attention_layernorm_weight3: R.Tensor((768,), dtype="float16") = packed_params[111] + gpt_neox_layers_9_post_attention_layernorm_bias3: R.Tensor((768,), dtype="float16") = packed_params[112] + gpt_neox_layers_9_attention_query_key_value_weight3: R.Tensor((2304, 768), dtype="float16") = packed_params[113] + gpt_neox_layers_9_attention_query_key_value_bias3: R.Tensor((2304,), dtype="float16") = packed_params[114] + gpt_neox_layers_9_attention_dense_weight3: R.Tensor((768, 768), dtype="float16") = packed_params[115] + gpt_neox_layers_9_attention_dense_bias3: R.Tensor((768,), dtype="float16") = packed_params[116] + gpt_neox_layers_9_mlp_dense_h_to_4h_weight3: R.Tensor((3072, 768), dtype="float16") = packed_params[117] + gpt_neox_layers_9_mlp_dense_h_to_4h_bias3: R.Tensor((3072,), dtype="float32") = packed_params[118] + gpt_neox_layers_9_mlp_dense_4h_to_h_weight3: R.Tensor((768, 3072), dtype="float16") = packed_params[119] + gpt_neox_layers_9_mlp_dense_4h_to_h_bias3: R.Tensor((768,), dtype="float32") = packed_params[120] + gpt_neox_layers_10_input_layernorm_weight3: R.Tensor((768,), dtype="float16") = packed_params[121] + gpt_neox_layers_10_input_layernorm_bias3: R.Tensor((768,), dtype="float16") = packed_params[122] + gpt_neox_layers_10_post_attention_layernorm_weight3: R.Tensor((768,), dtype="float16") = packed_params[123] + gpt_neox_layers_10_post_attention_layernorm_bias3: R.Tensor((768,), dtype="float16") = packed_params[124] + gpt_neox_layers_10_attention_query_key_value_weight3: R.Tensor((2304, 768), dtype="float16") = packed_params[125] + gpt_neox_layers_10_attention_query_key_value_bias3: R.Tensor((2304,), dtype="float16") = packed_params[126] + gpt_neox_layers_10_attention_dense_weight3: R.Tensor((768, 768), dtype="float16") = packed_params[127] + gpt_neox_layers_10_attention_dense_bias3: R.Tensor((768,), dtype="float16") = packed_params[128] + gpt_neox_layers_10_mlp_dense_h_to_4h_weight3: R.Tensor((3072, 768), dtype="float16") = packed_params[129] + gpt_neox_layers_10_mlp_dense_h_to_4h_bias3: R.Tensor((3072,), dtype="float32") = packed_params[130] + gpt_neox_layers_10_mlp_dense_4h_to_h_weight3: R.Tensor((768, 3072), dtype="float16") = packed_params[131] + gpt_neox_layers_10_mlp_dense_4h_to_h_bias3: R.Tensor((768,), dtype="float32") = packed_params[132] + gpt_neox_layers_11_input_layernorm_weight3: R.Tensor((768,), dtype="float16") = packed_params[133] + gpt_neox_layers_11_input_layernorm_bias3: R.Tensor((768,), dtype="float16") = packed_params[134] + gpt_neox_layers_11_post_attention_layernorm_weight3: R.Tensor((768,), dtype="float16") = packed_params[135] + gpt_neox_layers_11_post_attention_layernorm_bias3: R.Tensor((768,), dtype="float16") = packed_params[136] + gpt_neox_layers_11_attention_query_key_value_weight3: R.Tensor((2304, 768), dtype="float16") = packed_params[137] + gpt_neox_layers_11_attention_query_key_value_bias3: R.Tensor((2304,), dtype="float16") = packed_params[138] + gpt_neox_layers_11_attention_dense_weight3: R.Tensor((768, 768), dtype="float16") = packed_params[139] + gpt_neox_layers_11_attention_dense_bias3: R.Tensor((768,), dtype="float16") = packed_params[140] + gpt_neox_layers_11_mlp_dense_h_to_4h_weight3: R.Tensor((3072, 768), dtype="float16") = packed_params[141] + gpt_neox_layers_11_mlp_dense_h_to_4h_bias3: R.Tensor((3072,), dtype="float32") = packed_params[142] + gpt_neox_layers_11_mlp_dense_4h_to_h_weight3: R.Tensor((768, 3072), dtype="float16") = packed_params[143] + gpt_neox_layers_11_mlp_dense_4h_to_h_bias3: R.Tensor((768,), dtype="float32") = packed_params[144] + gpt_neox_final_layer_norm_weight3: R.Tensor((768,), dtype="float16") = packed_params[145] + gpt_neox_final_layer_norm_bias3: R.Tensor((768,), dtype="float16") = packed_params[146] + embed_out_weight3: R.Tensor((vocab_size, 768), dtype="float16") = packed_params[147] + layer_norm50: R.Tensor((1, seq_len, 768), dtype="float16") = R.nn.layer_norm(input_embeds, gpt_neox_layers_0_input_layernorm_weight3, gpt_neox_layers_0_input_layernorm_bias3, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims98: R.Tensor((768, 2304), dtype="float16") = R.permute_dims(gpt_neox_layers_0_attention_query_key_value_weight3, axes=None) + matmul98: R.Tensor((1, seq_len, 2304), dtype="float16") = R.matmul(layer_norm50, permute_dims98, out_dtype="void") + add144: R.Tensor((1, seq_len, 2304), dtype="float16") = R.add(matmul98, gpt_neox_layers_0_attention_query_key_value_bias3) + reshape96: R.Tensor((1, seq_len, 36, 64), dtype="float16") = R.reshape(add144, R.shape([1, seq_len, 36, 64])) + reshape97: R.Tensor((seq_len, 36, 64), dtype="float16") = R.reshape(reshape96, R.shape([seq_len, 36, 64])) + lv25 = R.call_dps_packed("vm.builtin.attention_kv_cache_attention_with_fused_qkv", (paged_kv_cache, R.prim_value(0), R.prim_value(T.float32(1.0)), reshape97), out_sinfo=R.Tensor((seq_len, 12, 64), dtype="float16")) + reshape98: R.Tensor((1, seq_len, 12, 64), dtype="float16") = R.reshape(lv25, R.shape([1, seq_len, 12, 64])) + reshape99: R.Tensor((1, seq_len, 768), dtype="float16") = R.reshape(reshape98, R.shape([1, seq_len, 768])) + permute_dims99: R.Tensor((768, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_0_attention_dense_weight3, axes=None) + matmul99: R.Tensor((1, seq_len, 768), dtype="float16") = R.matmul(reshape99, permute_dims99, out_dtype="void") + add145: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(matmul99, gpt_neox_layers_0_attention_dense_bias3) + layer_norm51: R.Tensor((1, seq_len, 768), dtype="float16") = R.nn.layer_norm(input_embeds, gpt_neox_layers_0_post_attention_layernorm_weight3, gpt_neox_layers_0_post_attention_layernorm_bias3, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims100: R.Tensor((768, 3072), dtype="float16") = R.permute_dims(gpt_neox_layers_0_mlp_dense_h_to_4h_weight3, axes=None) + matmul100: R.Tensor((1, seq_len, 3072), dtype="float32") = R.matmul(layer_norm51, permute_dims100, out_dtype="float32") + add146: R.Tensor((1, seq_len, 3072), dtype="float32") = R.add(matmul100, gpt_neox_layers_0_mlp_dense_h_to_4h_bias3) + gelu24: R.Tensor((1, seq_len, 3072), dtype="float32") = R.nn.gelu(add146) + astype50: R.Tensor((1, seq_len, 3072), dtype="float16") = R.astype(gelu24, dtype="float16") + permute_dims101: R.Tensor((3072, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_0_mlp_dense_4h_to_h_weight3, axes=None) + matmul101: R.Tensor((1, seq_len, 768), dtype="float32") = R.matmul(astype50, permute_dims101, out_dtype="float32") + add147: R.Tensor((1, seq_len, 768), dtype="float32") = R.add(matmul101, gpt_neox_layers_0_mlp_dense_4h_to_h_bias3) + astype51: R.Tensor((1, seq_len, 768), dtype="float16") = R.astype(add147, dtype="float16") + add148: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(astype51, add145) + add149: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(add148, input_embeds) + layer_norm52: R.Tensor((1, seq_len, 768), dtype="float16") = R.nn.layer_norm(add149, gpt_neox_layers_1_input_layernorm_weight3, gpt_neox_layers_1_input_layernorm_bias3, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims102: R.Tensor((768, 2304), dtype="float16") = R.permute_dims(gpt_neox_layers_1_attention_query_key_value_weight3, axes=None) + matmul102: R.Tensor((1, seq_len, 2304), dtype="float16") = R.matmul(layer_norm52, permute_dims102, out_dtype="void") + add150: R.Tensor((1, seq_len, 2304), dtype="float16") = R.add(matmul102, gpt_neox_layers_1_attention_query_key_value_bias3) + reshape100: R.Tensor((1, seq_len, 36, 64), dtype="float16") = R.reshape(add150, R.shape([1, seq_len, 36, 64])) + reshape101: R.Tensor((seq_len, 36, 64), dtype="float16") = R.reshape(reshape100, R.shape([seq_len, 36, 64])) + lv26 = R.call_dps_packed("vm.builtin.attention_kv_cache_attention_with_fused_qkv", (paged_kv_cache, R.prim_value(1), R.prim_value(T.float32(1.0)), reshape101), out_sinfo=R.Tensor((seq_len, 12, 64), dtype="float16")) + reshape102: R.Tensor((1, seq_len, 12, 64), dtype="float16") = R.reshape(lv26, R.shape([1, seq_len, 12, 64])) + reshape103: R.Tensor((1, seq_len, 768), dtype="float16") = R.reshape(reshape102, R.shape([1, seq_len, 768])) + permute_dims103: R.Tensor((768, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_1_attention_dense_weight3, axes=None) + matmul103: R.Tensor((1, seq_len, 768), dtype="float16") = R.matmul(reshape103, permute_dims103, out_dtype="void") + add151: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(matmul103, gpt_neox_layers_1_attention_dense_bias3) + layer_norm53: R.Tensor((1, seq_len, 768), dtype="float16") = R.nn.layer_norm(add149, gpt_neox_layers_1_post_attention_layernorm_weight3, gpt_neox_layers_1_post_attention_layernorm_bias3, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims104: R.Tensor((768, 3072), dtype="float16") = R.permute_dims(gpt_neox_layers_1_mlp_dense_h_to_4h_weight3, axes=None) + matmul104: R.Tensor((1, seq_len, 3072), dtype="float32") = R.matmul(layer_norm53, permute_dims104, out_dtype="float32") + add152: R.Tensor((1, seq_len, 3072), dtype="float32") = R.add(matmul104, gpt_neox_layers_1_mlp_dense_h_to_4h_bias3) + gelu25: R.Tensor((1, seq_len, 3072), dtype="float32") = R.nn.gelu(add152) + astype52: R.Tensor((1, seq_len, 3072), dtype="float16") = R.astype(gelu25, dtype="float16") + permute_dims105: R.Tensor((3072, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_1_mlp_dense_4h_to_h_weight3, axes=None) + matmul105: R.Tensor((1, seq_len, 768), dtype="float32") = R.matmul(astype52, permute_dims105, out_dtype="float32") + add153: R.Tensor((1, seq_len, 768), dtype="float32") = R.add(matmul105, gpt_neox_layers_1_mlp_dense_4h_to_h_bias3) + astype53: R.Tensor((1, seq_len, 768), dtype="float16") = R.astype(add153, dtype="float16") + add154: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(astype53, add151) + add155: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(add154, add149) + layer_norm54: R.Tensor((1, seq_len, 768), dtype="float16") = R.nn.layer_norm(add155, gpt_neox_layers_2_input_layernorm_weight3, gpt_neox_layers_2_input_layernorm_bias3, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims106: R.Tensor((768, 2304), dtype="float16") = R.permute_dims(gpt_neox_layers_2_attention_query_key_value_weight3, axes=None) + matmul106: R.Tensor((1, seq_len, 2304), dtype="float16") = R.matmul(layer_norm54, permute_dims106, out_dtype="void") + add156: R.Tensor((1, seq_len, 2304), dtype="float16") = R.add(matmul106, gpt_neox_layers_2_attention_query_key_value_bias3) + reshape104: R.Tensor((1, seq_len, 36, 64), dtype="float16") = R.reshape(add156, R.shape([1, seq_len, 36, 64])) + reshape105: R.Tensor((seq_len, 36, 64), dtype="float16") = R.reshape(reshape104, R.shape([seq_len, 36, 64])) + lv27 = R.call_dps_packed("vm.builtin.attention_kv_cache_attention_with_fused_qkv", (paged_kv_cache, R.prim_value(2), R.prim_value(T.float32(1.0)), reshape105), out_sinfo=R.Tensor((seq_len, 12, 64), dtype="float16")) + reshape106: R.Tensor((1, seq_len, 12, 64), dtype="float16") = R.reshape(lv27, R.shape([1, seq_len, 12, 64])) + reshape107: R.Tensor((1, seq_len, 768), dtype="float16") = R.reshape(reshape106, R.shape([1, seq_len, 768])) + permute_dims107: R.Tensor((768, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_2_attention_dense_weight3, axes=None) + matmul107: R.Tensor((1, seq_len, 768), dtype="float16") = R.matmul(reshape107, permute_dims107, out_dtype="void") + add157: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(matmul107, gpt_neox_layers_2_attention_dense_bias3) + layer_norm55: R.Tensor((1, seq_len, 768), dtype="float16") = R.nn.layer_norm(add155, gpt_neox_layers_2_post_attention_layernorm_weight3, gpt_neox_layers_2_post_attention_layernorm_bias3, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims108: R.Tensor((768, 3072), dtype="float16") = R.permute_dims(gpt_neox_layers_2_mlp_dense_h_to_4h_weight3, axes=None) + matmul108: R.Tensor((1, seq_len, 3072), dtype="float32") = R.matmul(layer_norm55, permute_dims108, out_dtype="float32") + add158: R.Tensor((1, seq_len, 3072), dtype="float32") = R.add(matmul108, gpt_neox_layers_2_mlp_dense_h_to_4h_bias3) + gelu26: R.Tensor((1, seq_len, 3072), dtype="float32") = R.nn.gelu(add158) + astype54: R.Tensor((1, seq_len, 3072), dtype="float16") = R.astype(gelu26, dtype="float16") + permute_dims109: R.Tensor((3072, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_2_mlp_dense_4h_to_h_weight3, axes=None) + matmul109: R.Tensor((1, seq_len, 768), dtype="float32") = R.matmul(astype54, permute_dims109, out_dtype="float32") + add159: R.Tensor((1, seq_len, 768), dtype="float32") = R.add(matmul109, gpt_neox_layers_2_mlp_dense_4h_to_h_bias3) + astype55: R.Tensor((1, seq_len, 768), dtype="float16") = R.astype(add159, dtype="float16") + add160: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(astype55, add157) + add161: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(add160, add155) + layer_norm56: R.Tensor((1, seq_len, 768), dtype="float16") = R.nn.layer_norm(add161, gpt_neox_layers_3_input_layernorm_weight3, gpt_neox_layers_3_input_layernorm_bias3, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims110: R.Tensor((768, 2304), dtype="float16") = R.permute_dims(gpt_neox_layers_3_attention_query_key_value_weight3, axes=None) + matmul110: R.Tensor((1, seq_len, 2304), dtype="float16") = R.matmul(layer_norm56, permute_dims110, out_dtype="void") + add162: R.Tensor((1, seq_len, 2304), dtype="float16") = R.add(matmul110, gpt_neox_layers_3_attention_query_key_value_bias3) + reshape108: R.Tensor((1, seq_len, 36, 64), dtype="float16") = R.reshape(add162, R.shape([1, seq_len, 36, 64])) + reshape109: R.Tensor((seq_len, 36, 64), dtype="float16") = R.reshape(reshape108, R.shape([seq_len, 36, 64])) + lv28 = R.call_dps_packed("vm.builtin.attention_kv_cache_attention_with_fused_qkv", (paged_kv_cache, R.prim_value(3), R.prim_value(T.float32(1.0)), reshape109), out_sinfo=R.Tensor((seq_len, 12, 64), dtype="float16")) + reshape110: R.Tensor((1, seq_len, 12, 64), dtype="float16") = R.reshape(lv28, R.shape([1, seq_len, 12, 64])) + reshape111: R.Tensor((1, seq_len, 768), dtype="float16") = R.reshape(reshape110, R.shape([1, seq_len, 768])) + permute_dims111: R.Tensor((768, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_3_attention_dense_weight3, axes=None) + matmul111: R.Tensor((1, seq_len, 768), dtype="float16") = R.matmul(reshape111, permute_dims111, out_dtype="void") + add163: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(matmul111, gpt_neox_layers_3_attention_dense_bias3) + layer_norm57: R.Tensor((1, seq_len, 768), dtype="float16") = R.nn.layer_norm(add161, gpt_neox_layers_3_post_attention_layernorm_weight3, gpt_neox_layers_3_post_attention_layernorm_bias3, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims112: R.Tensor((768, 3072), dtype="float16") = R.permute_dims(gpt_neox_layers_3_mlp_dense_h_to_4h_weight3, axes=None) + matmul112: R.Tensor((1, seq_len, 3072), dtype="float32") = R.matmul(layer_norm57, permute_dims112, out_dtype="float32") + add164: R.Tensor((1, seq_len, 3072), dtype="float32") = R.add(matmul112, gpt_neox_layers_3_mlp_dense_h_to_4h_bias3) + gelu27: R.Tensor((1, seq_len, 3072), dtype="float32") = R.nn.gelu(add164) + astype56: R.Tensor((1, seq_len, 3072), dtype="float16") = R.astype(gelu27, dtype="float16") + permute_dims113: R.Tensor((3072, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_3_mlp_dense_4h_to_h_weight3, axes=None) + matmul113: R.Tensor((1, seq_len, 768), dtype="float32") = R.matmul(astype56, permute_dims113, out_dtype="float32") + add165: R.Tensor((1, seq_len, 768), dtype="float32") = R.add(matmul113, gpt_neox_layers_3_mlp_dense_4h_to_h_bias3) + astype57: R.Tensor((1, seq_len, 768), dtype="float16") = R.astype(add165, dtype="float16") + add166: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(astype57, add163) + add167: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(add166, add161) + layer_norm58: R.Tensor((1, seq_len, 768), dtype="float16") = R.nn.layer_norm(add167, gpt_neox_layers_4_input_layernorm_weight3, gpt_neox_layers_4_input_layernorm_bias3, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims114: R.Tensor((768, 2304), dtype="float16") = R.permute_dims(gpt_neox_layers_4_attention_query_key_value_weight3, axes=None) + matmul114: R.Tensor((1, seq_len, 2304), dtype="float16") = R.matmul(layer_norm58, permute_dims114, out_dtype="void") + add168: R.Tensor((1, seq_len, 2304), dtype="float16") = R.add(matmul114, gpt_neox_layers_4_attention_query_key_value_bias3) + reshape112: R.Tensor((1, seq_len, 36, 64), dtype="float16") = R.reshape(add168, R.shape([1, seq_len, 36, 64])) + reshape113: R.Tensor((seq_len, 36, 64), dtype="float16") = R.reshape(reshape112, R.shape([seq_len, 36, 64])) + lv29 = R.call_dps_packed("vm.builtin.attention_kv_cache_attention_with_fused_qkv", (paged_kv_cache, R.prim_value(4), R.prim_value(T.float32(1.0)), reshape113), out_sinfo=R.Tensor((seq_len, 12, 64), dtype="float16")) + reshape114: R.Tensor((1, seq_len, 12, 64), dtype="float16") = R.reshape(lv29, R.shape([1, seq_len, 12, 64])) + reshape115: R.Tensor((1, seq_len, 768), dtype="float16") = R.reshape(reshape114, R.shape([1, seq_len, 768])) + permute_dims115: R.Tensor((768, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_4_attention_dense_weight3, axes=None) + matmul115: R.Tensor((1, seq_len, 768), dtype="float16") = R.matmul(reshape115, permute_dims115, out_dtype="void") + add169: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(matmul115, gpt_neox_layers_4_attention_dense_bias3) + layer_norm59: R.Tensor((1, seq_len, 768), dtype="float16") = R.nn.layer_norm(add167, gpt_neox_layers_4_post_attention_layernorm_weight3, gpt_neox_layers_4_post_attention_layernorm_bias3, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims116: R.Tensor((768, 3072), dtype="float16") = R.permute_dims(gpt_neox_layers_4_mlp_dense_h_to_4h_weight3, axes=None) + matmul116: R.Tensor((1, seq_len, 3072), dtype="float32") = R.matmul(layer_norm59, permute_dims116, out_dtype="float32") + add170: R.Tensor((1, seq_len, 3072), dtype="float32") = R.add(matmul116, gpt_neox_layers_4_mlp_dense_h_to_4h_bias3) + gelu28: R.Tensor((1, seq_len, 3072), dtype="float32") = R.nn.gelu(add170) + astype58: R.Tensor((1, seq_len, 3072), dtype="float16") = R.astype(gelu28, dtype="float16") + permute_dims117: R.Tensor((3072, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_4_mlp_dense_4h_to_h_weight3, axes=None) + matmul117: R.Tensor((1, seq_len, 768), dtype="float32") = R.matmul(astype58, permute_dims117, out_dtype="float32") + add171: R.Tensor((1, seq_len, 768), dtype="float32") = R.add(matmul117, gpt_neox_layers_4_mlp_dense_4h_to_h_bias3) + astype59: R.Tensor((1, seq_len, 768), dtype="float16") = R.astype(add171, dtype="float16") + add172: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(astype59, add169) + add173: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(add172, add167) + layer_norm60: R.Tensor((1, seq_len, 768), dtype="float16") = R.nn.layer_norm(add173, gpt_neox_layers_5_input_layernorm_weight3, gpt_neox_layers_5_input_layernorm_bias3, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims118: R.Tensor((768, 2304), dtype="float16") = R.permute_dims(gpt_neox_layers_5_attention_query_key_value_weight3, axes=None) + matmul118: R.Tensor((1, seq_len, 2304), dtype="float16") = R.matmul(layer_norm60, permute_dims118, out_dtype="void") + add174: R.Tensor((1, seq_len, 2304), dtype="float16") = R.add(matmul118, gpt_neox_layers_5_attention_query_key_value_bias3) + reshape116: R.Tensor((1, seq_len, 36, 64), dtype="float16") = R.reshape(add174, R.shape([1, seq_len, 36, 64])) + reshape117: R.Tensor((seq_len, 36, 64), dtype="float16") = R.reshape(reshape116, R.shape([seq_len, 36, 64])) + lv30 = R.call_dps_packed("vm.builtin.attention_kv_cache_attention_with_fused_qkv", (paged_kv_cache, R.prim_value(5), R.prim_value(T.float32(1.0)), reshape117), out_sinfo=R.Tensor((seq_len, 12, 64), dtype="float16")) + reshape118: R.Tensor((1, seq_len, 12, 64), dtype="float16") = R.reshape(lv30, R.shape([1, seq_len, 12, 64])) + reshape119: R.Tensor((1, seq_len, 768), dtype="float16") = R.reshape(reshape118, R.shape([1, seq_len, 768])) + permute_dims119: R.Tensor((768, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_5_attention_dense_weight3, axes=None) + matmul119: R.Tensor((1, seq_len, 768), dtype="float16") = R.matmul(reshape119, permute_dims119, out_dtype="void") + add175: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(matmul119, gpt_neox_layers_5_attention_dense_bias3) + layer_norm61: R.Tensor((1, seq_len, 768), dtype="float16") = R.nn.layer_norm(add173, gpt_neox_layers_5_post_attention_layernorm_weight3, gpt_neox_layers_5_post_attention_layernorm_bias3, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims120: R.Tensor((768, 3072), dtype="float16") = R.permute_dims(gpt_neox_layers_5_mlp_dense_h_to_4h_weight3, axes=None) + matmul120: R.Tensor((1, seq_len, 3072), dtype="float32") = R.matmul(layer_norm61, permute_dims120, out_dtype="float32") + add176: R.Tensor((1, seq_len, 3072), dtype="float32") = R.add(matmul120, gpt_neox_layers_5_mlp_dense_h_to_4h_bias3) + gelu29: R.Tensor((1, seq_len, 3072), dtype="float32") = R.nn.gelu(add176) + astype60: R.Tensor((1, seq_len, 3072), dtype="float16") = R.astype(gelu29, dtype="float16") + permute_dims121: R.Tensor((3072, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_5_mlp_dense_4h_to_h_weight3, axes=None) + matmul121: R.Tensor((1, seq_len, 768), dtype="float32") = R.matmul(astype60, permute_dims121, out_dtype="float32") + add177: R.Tensor((1, seq_len, 768), dtype="float32") = R.add(matmul121, gpt_neox_layers_5_mlp_dense_4h_to_h_bias3) + astype61: R.Tensor((1, seq_len, 768), dtype="float16") = R.astype(add177, dtype="float16") + add178: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(astype61, add175) + add179: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(add178, add173) + layer_norm62: R.Tensor((1, seq_len, 768), dtype="float16") = R.nn.layer_norm(add179, gpt_neox_layers_6_input_layernorm_weight3, gpt_neox_layers_6_input_layernorm_bias3, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims122: R.Tensor((768, 2304), dtype="float16") = R.permute_dims(gpt_neox_layers_6_attention_query_key_value_weight3, axes=None) + matmul122: R.Tensor((1, seq_len, 2304), dtype="float16") = R.matmul(layer_norm62, permute_dims122, out_dtype="void") + add180: R.Tensor((1, seq_len, 2304), dtype="float16") = R.add(matmul122, gpt_neox_layers_6_attention_query_key_value_bias3) + reshape120: R.Tensor((1, seq_len, 36, 64), dtype="float16") = R.reshape(add180, R.shape([1, seq_len, 36, 64])) + reshape121: R.Tensor((seq_len, 36, 64), dtype="float16") = R.reshape(reshape120, R.shape([seq_len, 36, 64])) + lv31 = R.call_dps_packed("vm.builtin.attention_kv_cache_attention_with_fused_qkv", (paged_kv_cache, R.prim_value(6), R.prim_value(T.float32(1.0)), reshape121), out_sinfo=R.Tensor((seq_len, 12, 64), dtype="float16")) + reshape122: R.Tensor((1, seq_len, 12, 64), dtype="float16") = R.reshape(lv31, R.shape([1, seq_len, 12, 64])) + reshape123: R.Tensor((1, seq_len, 768), dtype="float16") = R.reshape(reshape122, R.shape([1, seq_len, 768])) + permute_dims123: R.Tensor((768, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_6_attention_dense_weight3, axes=None) + matmul123: R.Tensor((1, seq_len, 768), dtype="float16") = R.matmul(reshape123, permute_dims123, out_dtype="void") + add181: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(matmul123, gpt_neox_layers_6_attention_dense_bias3) + layer_norm63: R.Tensor((1, seq_len, 768), dtype="float16") = R.nn.layer_norm(add179, gpt_neox_layers_6_post_attention_layernorm_weight3, gpt_neox_layers_6_post_attention_layernorm_bias3, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims124: R.Tensor((768, 3072), dtype="float16") = R.permute_dims(gpt_neox_layers_6_mlp_dense_h_to_4h_weight3, axes=None) + matmul124: R.Tensor((1, seq_len, 3072), dtype="float32") = R.matmul(layer_norm63, permute_dims124, out_dtype="float32") + add182: R.Tensor((1, seq_len, 3072), dtype="float32") = R.add(matmul124, gpt_neox_layers_6_mlp_dense_h_to_4h_bias3) + gelu30: R.Tensor((1, seq_len, 3072), dtype="float32") = R.nn.gelu(add182) + astype62: R.Tensor((1, seq_len, 3072), dtype="float16") = R.astype(gelu30, dtype="float16") + permute_dims125: R.Tensor((3072, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_6_mlp_dense_4h_to_h_weight3, axes=None) + matmul125: R.Tensor((1, seq_len, 768), dtype="float32") = R.matmul(astype62, permute_dims125, out_dtype="float32") + add183: R.Tensor((1, seq_len, 768), dtype="float32") = R.add(matmul125, gpt_neox_layers_6_mlp_dense_4h_to_h_bias3) + astype63: R.Tensor((1, seq_len, 768), dtype="float16") = R.astype(add183, dtype="float16") + add184: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(astype63, add181) + add185: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(add184, add179) + layer_norm64: R.Tensor((1, seq_len, 768), dtype="float16") = R.nn.layer_norm(add185, gpt_neox_layers_7_input_layernorm_weight3, gpt_neox_layers_7_input_layernorm_bias3, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims126: R.Tensor((768, 2304), dtype="float16") = R.permute_dims(gpt_neox_layers_7_attention_query_key_value_weight3, axes=None) + matmul126: R.Tensor((1, seq_len, 2304), dtype="float16") = R.matmul(layer_norm64, permute_dims126, out_dtype="void") + add186: R.Tensor((1, seq_len, 2304), dtype="float16") = R.add(matmul126, gpt_neox_layers_7_attention_query_key_value_bias3) + reshape124: R.Tensor((1, seq_len, 36, 64), dtype="float16") = R.reshape(add186, R.shape([1, seq_len, 36, 64])) + reshape125: R.Tensor((seq_len, 36, 64), dtype="float16") = R.reshape(reshape124, R.shape([seq_len, 36, 64])) + lv32 = R.call_dps_packed("vm.builtin.attention_kv_cache_attention_with_fused_qkv", (paged_kv_cache, R.prim_value(7), R.prim_value(T.float32(1.0)), reshape125), out_sinfo=R.Tensor((seq_len, 12, 64), dtype="float16")) + reshape126: R.Tensor((1, seq_len, 12, 64), dtype="float16") = R.reshape(lv32, R.shape([1, seq_len, 12, 64])) + reshape127: R.Tensor((1, seq_len, 768), dtype="float16") = R.reshape(reshape126, R.shape([1, seq_len, 768])) + permute_dims127: R.Tensor((768, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_7_attention_dense_weight3, axes=None) + matmul127: R.Tensor((1, seq_len, 768), dtype="float16") = R.matmul(reshape127, permute_dims127, out_dtype="void") + add187: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(matmul127, gpt_neox_layers_7_attention_dense_bias3) + layer_norm65: R.Tensor((1, seq_len, 768), dtype="float16") = R.nn.layer_norm(add185, gpt_neox_layers_7_post_attention_layernorm_weight3, gpt_neox_layers_7_post_attention_layernorm_bias3, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims128: R.Tensor((768, 3072), dtype="float16") = R.permute_dims(gpt_neox_layers_7_mlp_dense_h_to_4h_weight3, axes=None) + matmul128: R.Tensor((1, seq_len, 3072), dtype="float32") = R.matmul(layer_norm65, permute_dims128, out_dtype="float32") + add188: R.Tensor((1, seq_len, 3072), dtype="float32") = R.add(matmul128, gpt_neox_layers_7_mlp_dense_h_to_4h_bias3) + gelu31: R.Tensor((1, seq_len, 3072), dtype="float32") = R.nn.gelu(add188) + astype64: R.Tensor((1, seq_len, 3072), dtype="float16") = R.astype(gelu31, dtype="float16") + permute_dims129: R.Tensor((3072, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_7_mlp_dense_4h_to_h_weight3, axes=None) + matmul129: R.Tensor((1, seq_len, 768), dtype="float32") = R.matmul(astype64, permute_dims129, out_dtype="float32") + add189: R.Tensor((1, seq_len, 768), dtype="float32") = R.add(matmul129, gpt_neox_layers_7_mlp_dense_4h_to_h_bias3) + astype65: R.Tensor((1, seq_len, 768), dtype="float16") = R.astype(add189, dtype="float16") + add190: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(astype65, add187) + add191: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(add190, add185) + layer_norm66: R.Tensor((1, seq_len, 768), dtype="float16") = R.nn.layer_norm(add191, gpt_neox_layers_8_input_layernorm_weight3, gpt_neox_layers_8_input_layernorm_bias3, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims130: R.Tensor((768, 2304), dtype="float16") = R.permute_dims(gpt_neox_layers_8_attention_query_key_value_weight3, axes=None) + matmul130: R.Tensor((1, seq_len, 2304), dtype="float16") = R.matmul(layer_norm66, permute_dims130, out_dtype="void") + add192: R.Tensor((1, seq_len, 2304), dtype="float16") = R.add(matmul130, gpt_neox_layers_8_attention_query_key_value_bias3) + reshape128: R.Tensor((1, seq_len, 36, 64), dtype="float16") = R.reshape(add192, R.shape([1, seq_len, 36, 64])) + reshape129: R.Tensor((seq_len, 36, 64), dtype="float16") = R.reshape(reshape128, R.shape([seq_len, 36, 64])) + lv33 = R.call_dps_packed("vm.builtin.attention_kv_cache_attention_with_fused_qkv", (paged_kv_cache, R.prim_value(8), R.prim_value(T.float32(1.0)), reshape129), out_sinfo=R.Tensor((seq_len, 12, 64), dtype="float16")) + reshape130: R.Tensor((1, seq_len, 12, 64), dtype="float16") = R.reshape(lv33, R.shape([1, seq_len, 12, 64])) + reshape131: R.Tensor((1, seq_len, 768), dtype="float16") = R.reshape(reshape130, R.shape([1, seq_len, 768])) + permute_dims131: R.Tensor((768, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_8_attention_dense_weight3, axes=None) + matmul131: R.Tensor((1, seq_len, 768), dtype="float16") = R.matmul(reshape131, permute_dims131, out_dtype="void") + add193: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(matmul131, gpt_neox_layers_8_attention_dense_bias3) + layer_norm67: R.Tensor((1, seq_len, 768), dtype="float16") = R.nn.layer_norm(add191, gpt_neox_layers_8_post_attention_layernorm_weight3, gpt_neox_layers_8_post_attention_layernorm_bias3, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims132: R.Tensor((768, 3072), dtype="float16") = R.permute_dims(gpt_neox_layers_8_mlp_dense_h_to_4h_weight3, axes=None) + matmul132: R.Tensor((1, seq_len, 3072), dtype="float32") = R.matmul(layer_norm67, permute_dims132, out_dtype="float32") + add194: R.Tensor((1, seq_len, 3072), dtype="float32") = R.add(matmul132, gpt_neox_layers_8_mlp_dense_h_to_4h_bias3) + gelu32: R.Tensor((1, seq_len, 3072), dtype="float32") = R.nn.gelu(add194) + astype66: R.Tensor((1, seq_len, 3072), dtype="float16") = R.astype(gelu32, dtype="float16") + permute_dims133: R.Tensor((3072, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_8_mlp_dense_4h_to_h_weight3, axes=None) + matmul133: R.Tensor((1, seq_len, 768), dtype="float32") = R.matmul(astype66, permute_dims133, out_dtype="float32") + add195: R.Tensor((1, seq_len, 768), dtype="float32") = R.add(matmul133, gpt_neox_layers_8_mlp_dense_4h_to_h_bias3) + astype67: R.Tensor((1, seq_len, 768), dtype="float16") = R.astype(add195, dtype="float16") + add196: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(astype67, add193) + add197: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(add196, add191) + layer_norm68: R.Tensor((1, seq_len, 768), dtype="float16") = R.nn.layer_norm(add197, gpt_neox_layers_9_input_layernorm_weight3, gpt_neox_layers_9_input_layernorm_bias3, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims134: R.Tensor((768, 2304), dtype="float16") = R.permute_dims(gpt_neox_layers_9_attention_query_key_value_weight3, axes=None) + matmul134: R.Tensor((1, seq_len, 2304), dtype="float16") = R.matmul(layer_norm68, permute_dims134, out_dtype="void") + add198: R.Tensor((1, seq_len, 2304), dtype="float16") = R.add(matmul134, gpt_neox_layers_9_attention_query_key_value_bias3) + reshape132: R.Tensor((1, seq_len, 36, 64), dtype="float16") = R.reshape(add198, R.shape([1, seq_len, 36, 64])) + reshape133: R.Tensor((seq_len, 36, 64), dtype="float16") = R.reshape(reshape132, R.shape([seq_len, 36, 64])) + lv34 = R.call_dps_packed("vm.builtin.attention_kv_cache_attention_with_fused_qkv", (paged_kv_cache, R.prim_value(9), R.prim_value(T.float32(1.0)), reshape133), out_sinfo=R.Tensor((seq_len, 12, 64), dtype="float16")) + reshape134: R.Tensor((1, seq_len, 12, 64), dtype="float16") = R.reshape(lv34, R.shape([1, seq_len, 12, 64])) + reshape135: R.Tensor((1, seq_len, 768), dtype="float16") = R.reshape(reshape134, R.shape([1, seq_len, 768])) + permute_dims135: R.Tensor((768, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_9_attention_dense_weight3, axes=None) + matmul135: R.Tensor((1, seq_len, 768), dtype="float16") = R.matmul(reshape135, permute_dims135, out_dtype="void") + add199: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(matmul135, gpt_neox_layers_9_attention_dense_bias3) + layer_norm69: R.Tensor((1, seq_len, 768), dtype="float16") = R.nn.layer_norm(add197, gpt_neox_layers_9_post_attention_layernorm_weight3, gpt_neox_layers_9_post_attention_layernorm_bias3, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims136: R.Tensor((768, 3072), dtype="float16") = R.permute_dims(gpt_neox_layers_9_mlp_dense_h_to_4h_weight3, axes=None) + matmul136: R.Tensor((1, seq_len, 3072), dtype="float32") = R.matmul(layer_norm69, permute_dims136, out_dtype="float32") + add200: R.Tensor((1, seq_len, 3072), dtype="float32") = R.add(matmul136, gpt_neox_layers_9_mlp_dense_h_to_4h_bias3) + gelu33: R.Tensor((1, seq_len, 3072), dtype="float32") = R.nn.gelu(add200) + astype68: R.Tensor((1, seq_len, 3072), dtype="float16") = R.astype(gelu33, dtype="float16") + permute_dims137: R.Tensor((3072, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_9_mlp_dense_4h_to_h_weight3, axes=None) + matmul137: R.Tensor((1, seq_len, 768), dtype="float32") = R.matmul(astype68, permute_dims137, out_dtype="float32") + add201: R.Tensor((1, seq_len, 768), dtype="float32") = R.add(matmul137, gpt_neox_layers_9_mlp_dense_4h_to_h_bias3) + astype69: R.Tensor((1, seq_len, 768), dtype="float16") = R.astype(add201, dtype="float16") + add202: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(astype69, add199) + add203: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(add202, add197) + layer_norm70: R.Tensor((1, seq_len, 768), dtype="float16") = R.nn.layer_norm(add203, gpt_neox_layers_10_input_layernorm_weight3, gpt_neox_layers_10_input_layernorm_bias3, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims138: R.Tensor((768, 2304), dtype="float16") = R.permute_dims(gpt_neox_layers_10_attention_query_key_value_weight3, axes=None) + matmul138: R.Tensor((1, seq_len, 2304), dtype="float16") = R.matmul(layer_norm70, permute_dims138, out_dtype="void") + add204: R.Tensor((1, seq_len, 2304), dtype="float16") = R.add(matmul138, gpt_neox_layers_10_attention_query_key_value_bias3) + reshape136: R.Tensor((1, seq_len, 36, 64), dtype="float16") = R.reshape(add204, R.shape([1, seq_len, 36, 64])) + reshape137: R.Tensor((seq_len, 36, 64), dtype="float16") = R.reshape(reshape136, R.shape([seq_len, 36, 64])) + lv35 = R.call_dps_packed("vm.builtin.attention_kv_cache_attention_with_fused_qkv", (paged_kv_cache, R.prim_value(10), R.prim_value(T.float32(1.0)), reshape137), out_sinfo=R.Tensor((seq_len, 12, 64), dtype="float16")) + reshape138: R.Tensor((1, seq_len, 12, 64), dtype="float16") = R.reshape(lv35, R.shape([1, seq_len, 12, 64])) + reshape139: R.Tensor((1, seq_len, 768), dtype="float16") = R.reshape(reshape138, R.shape([1, seq_len, 768])) + permute_dims139: R.Tensor((768, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_10_attention_dense_weight3, axes=None) + matmul139: R.Tensor((1, seq_len, 768), dtype="float16") = R.matmul(reshape139, permute_dims139, out_dtype="void") + add205: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(matmul139, gpt_neox_layers_10_attention_dense_bias3) + layer_norm71: R.Tensor((1, seq_len, 768), dtype="float16") = R.nn.layer_norm(add203, gpt_neox_layers_10_post_attention_layernorm_weight3, gpt_neox_layers_10_post_attention_layernorm_bias3, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims140: R.Tensor((768, 3072), dtype="float16") = R.permute_dims(gpt_neox_layers_10_mlp_dense_h_to_4h_weight3, axes=None) + matmul140: R.Tensor((1, seq_len, 3072), dtype="float32") = R.matmul(layer_norm71, permute_dims140, out_dtype="float32") + add206: R.Tensor((1, seq_len, 3072), dtype="float32") = R.add(matmul140, gpt_neox_layers_10_mlp_dense_h_to_4h_bias3) + gelu34: R.Tensor((1, seq_len, 3072), dtype="float32") = R.nn.gelu(add206) + astype70: R.Tensor((1, seq_len, 3072), dtype="float16") = R.astype(gelu34, dtype="float16") + permute_dims141: R.Tensor((3072, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_10_mlp_dense_4h_to_h_weight3, axes=None) + matmul141: R.Tensor((1, seq_len, 768), dtype="float32") = R.matmul(astype70, permute_dims141, out_dtype="float32") + add207: R.Tensor((1, seq_len, 768), dtype="float32") = R.add(matmul141, gpt_neox_layers_10_mlp_dense_4h_to_h_bias3) + astype71: R.Tensor((1, seq_len, 768), dtype="float16") = R.astype(add207, dtype="float16") + add208: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(astype71, add205) + add209: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(add208, add203) + layer_norm72: R.Tensor((1, seq_len, 768), dtype="float16") = R.nn.layer_norm(add209, gpt_neox_layers_11_input_layernorm_weight3, gpt_neox_layers_11_input_layernorm_bias3, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims142: R.Tensor((768, 2304), dtype="float16") = R.permute_dims(gpt_neox_layers_11_attention_query_key_value_weight3, axes=None) + matmul142: R.Tensor((1, seq_len, 2304), dtype="float16") = R.matmul(layer_norm72, permute_dims142, out_dtype="void") + add210: R.Tensor((1, seq_len, 2304), dtype="float16") = R.add(matmul142, gpt_neox_layers_11_attention_query_key_value_bias3) + reshape140: R.Tensor((1, seq_len, 36, 64), dtype="float16") = R.reshape(add210, R.shape([1, seq_len, 36, 64])) + reshape141: R.Tensor((seq_len, 36, 64), dtype="float16") = R.reshape(reshape140, R.shape([seq_len, 36, 64])) + lv36 = R.call_dps_packed("vm.builtin.attention_kv_cache_attention_with_fused_qkv", (paged_kv_cache, R.prim_value(11), R.prim_value(T.float32(1.0)), reshape141), out_sinfo=R.Tensor((seq_len, 12, 64), dtype="float16")) + reshape142: R.Tensor((1, seq_len, 12, 64), dtype="float16") = R.reshape(lv36, R.shape([1, seq_len, 12, 64])) + reshape143: R.Tensor((1, seq_len, 768), dtype="float16") = R.reshape(reshape142, R.shape([1, seq_len, 768])) + permute_dims143: R.Tensor((768, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_11_attention_dense_weight3, axes=None) + matmul143: R.Tensor((1, seq_len, 768), dtype="float16") = R.matmul(reshape143, permute_dims143, out_dtype="void") + add211: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(matmul143, gpt_neox_layers_11_attention_dense_bias3) + layer_norm73: R.Tensor((1, seq_len, 768), dtype="float16") = R.nn.layer_norm(add209, gpt_neox_layers_11_post_attention_layernorm_weight3, gpt_neox_layers_11_post_attention_layernorm_bias3, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims144: R.Tensor((768, 3072), dtype="float16") = R.permute_dims(gpt_neox_layers_11_mlp_dense_h_to_4h_weight3, axes=None) + matmul144: R.Tensor((1, seq_len, 3072), dtype="float32") = R.matmul(layer_norm73, permute_dims144, out_dtype="float32") + add212: R.Tensor((1, seq_len, 3072), dtype="float32") = R.add(matmul144, gpt_neox_layers_11_mlp_dense_h_to_4h_bias3) + gelu35: R.Tensor((1, seq_len, 3072), dtype="float32") = R.nn.gelu(add212) + astype72: R.Tensor((1, seq_len, 3072), dtype="float16") = R.astype(gelu35, dtype="float16") + permute_dims145: R.Tensor((3072, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_11_mlp_dense_4h_to_h_weight3, axes=None) + matmul145: R.Tensor((1, seq_len, 768), dtype="float32") = R.matmul(astype72, permute_dims145, out_dtype="float32") + add213: R.Tensor((1, seq_len, 768), dtype="float32") = R.add(matmul145, gpt_neox_layers_11_mlp_dense_4h_to_h_bias3) + astype73: R.Tensor((1, seq_len, 768), dtype="float16") = R.astype(add213, dtype="float16") + add214: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(astype73, add211) + add215: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(add214, add209) + layer_norm74: R.Tensor((1, seq_len, 768), dtype="float16") = R.nn.layer_norm(add215, gpt_neox_final_layer_norm_weight3, gpt_neox_final_layer_norm_bias3, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + take1: R.Tensor((1, batch_size, 768), dtype="float16") = R.take(layer_norm74, logit_positions, axis=1) + permute_dims146: R.Tensor((768, vocab_size), dtype="float16") = R.permute_dims(embed_out_weight3, axes=None) + matmul146: R.Tensor((1, batch_size, vocab_size), dtype="float16") = R.matmul(take1, permute_dims146, out_dtype="void") + astype74: R.Tensor((1, batch_size, vocab_size), dtype="float32") = R.astype(matmul146, dtype="float32") + gv3: R.Tuple(R.Tensor((1, batch_size, vocab_size), dtype="float32"), R.Object) = astype74, paged_kv_cache + R.output(gv3) + return gv3 + + @R.function + def batch_verify(input_embeds: R.Tensor((1, "seq_len", 768), dtype="float16"), paged_kv_cache: R.Object, packed_params: R.Tuple(R.Tensor(("vocab_size", 768), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((2304, 768), dtype="float16"), R.Tensor((2304,), dtype="float16"), R.Tensor((768, 768), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((3072, 768), dtype="float16"), R.Tensor((3072,), dtype="float32"), R.Tensor((768, 3072), dtype="float16"), R.Tensor((768,), dtype="float32"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((2304, 768), dtype="float16"), R.Tensor((2304,), dtype="float16"), R.Tensor((768, 768), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((3072, 768), dtype="float16"), R.Tensor((3072,), dtype="float32"), R.Tensor((768, 3072), dtype="float16"), R.Tensor((768,), dtype="float32"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((2304, 768), dtype="float16"), R.Tensor((2304,), dtype="float16"), R.Tensor((768, 768), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((3072, 768), dtype="float16"), R.Tensor((3072,), dtype="float32"), R.Tensor((768, 3072), dtype="float16"), R.Tensor((768,), dtype="float32"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((2304, 768), dtype="float16"), R.Tensor((2304,), dtype="float16"), R.Tensor((768, 768), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((3072, 768), dtype="float16"), R.Tensor((3072,), dtype="float32"), R.Tensor((768, 3072), dtype="float16"), R.Tensor((768,), dtype="float32"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((2304, 768), dtype="float16"), R.Tensor((2304,), dtype="float16"), R.Tensor((768, 768), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((3072, 768), dtype="float16"), R.Tensor((3072,), dtype="float32"), R.Tensor((768, 3072), dtype="float16"), R.Tensor((768,), dtype="float32"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((2304, 768), dtype="float16"), R.Tensor((2304,), dtype="float16"), R.Tensor((768, 768), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((3072, 768), dtype="float16"), R.Tensor((3072,), dtype="float32"), R.Tensor((768, 3072), dtype="float16"), R.Tensor((768,), dtype="float32"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((2304, 768), dtype="float16"), R.Tensor((2304,), dtype="float16"), R.Tensor((768, 768), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((3072, 768), dtype="float16"), R.Tensor((3072,), dtype="float32"), R.Tensor((768, 3072), dtype="float16"), R.Tensor((768,), dtype="float32"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((2304, 768), dtype="float16"), R.Tensor((2304,), dtype="float16"), R.Tensor((768, 768), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((3072, 768), dtype="float16"), R.Tensor((3072,), dtype="float32"), R.Tensor((768, 3072), dtype="float16"), R.Tensor((768,), dtype="float32"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((2304, 768), dtype="float16"), R.Tensor((2304,), dtype="float16"), R.Tensor((768, 768), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((3072, 768), dtype="float16"), R.Tensor((3072,), dtype="float32"), R.Tensor((768, 3072), dtype="float16"), R.Tensor((768,), dtype="float32"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((2304, 768), dtype="float16"), R.Tensor((2304,), dtype="float16"), R.Tensor((768, 768), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((3072, 768), dtype="float16"), R.Tensor((3072,), dtype="float32"), R.Tensor((768, 3072), dtype="float16"), R.Tensor((768,), dtype="float32"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((2304, 768), dtype="float16"), R.Tensor((2304,), dtype="float16"), R.Tensor((768, 768), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((3072, 768), dtype="float16"), R.Tensor((3072,), dtype="float32"), R.Tensor((768, 3072), dtype="float16"), R.Tensor((768,), dtype="float32"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((2304, 768), dtype="float16"), R.Tensor((2304,), dtype="float16"), R.Tensor((768, 768), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((3072, 768), dtype="float16"), R.Tensor((3072,), dtype="float32"), R.Tensor((768, 3072), dtype="float16"), R.Tensor((768,), dtype="float32"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor(("vocab_size", 768), dtype="float16"))) -> R.Tuple(R.Tensor((1, "seq_len", "vocab_size"), dtype="float32"), R.Object): + seq_len = T.int64() + vocab_size = T.int64() + R.func_attr({"num_input": 2, "pipeline_parallel_stages": 1, "relax.memory_plan_dynamic_func_output": True, "relax.rewrite_cuda_graph.capture_symbolic_vars": ["batch_size", "seq_len"], "tir_non_negative_var": ["vocab_size"], "tir_var_upper_bound": {"batch_size": 128, "seq_len": 2048, "total_seq_len": 2048}}) + with R.dataflow(): + gpt_neox_embed_in_weight5: R.Tensor((vocab_size, 768), dtype="float16") = packed_params[0] + gpt_neox_layers_0_input_layernorm_weight5: R.Tensor((768,), dtype="float16") = packed_params[1] + gpt_neox_layers_0_input_layernorm_bias5: R.Tensor((768,), dtype="float16") = packed_params[2] + gpt_neox_layers_0_post_attention_layernorm_weight5: R.Tensor((768,), dtype="float16") = packed_params[3] + gpt_neox_layers_0_post_attention_layernorm_bias5: R.Tensor((768,), dtype="float16") = packed_params[4] + gpt_neox_layers_0_attention_query_key_value_weight5: R.Tensor((2304, 768), dtype="float16") = packed_params[5] + gpt_neox_layers_0_attention_query_key_value_bias5: R.Tensor((2304,), dtype="float16") = packed_params[6] + gpt_neox_layers_0_attention_dense_weight5: R.Tensor((768, 768), dtype="float16") = packed_params[7] + gpt_neox_layers_0_attention_dense_bias5: R.Tensor((768,), dtype="float16") = packed_params[8] + gpt_neox_layers_0_mlp_dense_h_to_4h_weight5: R.Tensor((3072, 768), dtype="float16") = packed_params[9] + gpt_neox_layers_0_mlp_dense_h_to_4h_bias5: R.Tensor((3072,), dtype="float32") = packed_params[10] + gpt_neox_layers_0_mlp_dense_4h_to_h_weight5: R.Tensor((768, 3072), dtype="float16") = packed_params[11] + gpt_neox_layers_0_mlp_dense_4h_to_h_bias5: R.Tensor((768,), dtype="float32") = packed_params[12] + gpt_neox_layers_1_input_layernorm_weight5: R.Tensor((768,), dtype="float16") = packed_params[13] + gpt_neox_layers_1_input_layernorm_bias5: R.Tensor((768,), dtype="float16") = packed_params[14] + gpt_neox_layers_1_post_attention_layernorm_weight5: R.Tensor((768,), dtype="float16") = packed_params[15] + gpt_neox_layers_1_post_attention_layernorm_bias5: R.Tensor((768,), dtype="float16") = packed_params[16] + gpt_neox_layers_1_attention_query_key_value_weight5: R.Tensor((2304, 768), dtype="float16") = packed_params[17] + gpt_neox_layers_1_attention_query_key_value_bias5: R.Tensor((2304,), dtype="float16") = packed_params[18] + gpt_neox_layers_1_attention_dense_weight5: R.Tensor((768, 768), dtype="float16") = packed_params[19] + gpt_neox_layers_1_attention_dense_bias5: R.Tensor((768,), dtype="float16") = packed_params[20] + gpt_neox_layers_1_mlp_dense_h_to_4h_weight5: R.Tensor((3072, 768), dtype="float16") = packed_params[21] + gpt_neox_layers_1_mlp_dense_h_to_4h_bias5: R.Tensor((3072,), dtype="float32") = packed_params[22] + gpt_neox_layers_1_mlp_dense_4h_to_h_weight5: R.Tensor((768, 3072), dtype="float16") = packed_params[23] + gpt_neox_layers_1_mlp_dense_4h_to_h_bias5: R.Tensor((768,), dtype="float32") = packed_params[24] + gpt_neox_layers_2_input_layernorm_weight5: R.Tensor((768,), dtype="float16") = packed_params[25] + gpt_neox_layers_2_input_layernorm_bias5: R.Tensor((768,), dtype="float16") = packed_params[26] + gpt_neox_layers_2_post_attention_layernorm_weight5: R.Tensor((768,), dtype="float16") = packed_params[27] + gpt_neox_layers_2_post_attention_layernorm_bias5: R.Tensor((768,), dtype="float16") = packed_params[28] + gpt_neox_layers_2_attention_query_key_value_weight5: R.Tensor((2304, 768), dtype="float16") = packed_params[29] + gpt_neox_layers_2_attention_query_key_value_bias5: R.Tensor((2304,), dtype="float16") = packed_params[30] + gpt_neox_layers_2_attention_dense_weight5: R.Tensor((768, 768), dtype="float16") = packed_params[31] + gpt_neox_layers_2_attention_dense_bias5: R.Tensor((768,), dtype="float16") = packed_params[32] + gpt_neox_layers_2_mlp_dense_h_to_4h_weight5: R.Tensor((3072, 768), dtype="float16") = packed_params[33] + gpt_neox_layers_2_mlp_dense_h_to_4h_bias5: R.Tensor((3072,), dtype="float32") = packed_params[34] + gpt_neox_layers_2_mlp_dense_4h_to_h_weight5: R.Tensor((768, 3072), dtype="float16") = packed_params[35] + gpt_neox_layers_2_mlp_dense_4h_to_h_bias5: R.Tensor((768,), dtype="float32") = packed_params[36] + gpt_neox_layers_3_input_layernorm_weight5: R.Tensor((768,), dtype="float16") = packed_params[37] + gpt_neox_layers_3_input_layernorm_bias5: R.Tensor((768,), dtype="float16") = packed_params[38] + gpt_neox_layers_3_post_attention_layernorm_weight5: R.Tensor((768,), dtype="float16") = packed_params[39] + gpt_neox_layers_3_post_attention_layernorm_bias5: R.Tensor((768,), dtype="float16") = packed_params[40] + gpt_neox_layers_3_attention_query_key_value_weight5: R.Tensor((2304, 768), dtype="float16") = packed_params[41] + gpt_neox_layers_3_attention_query_key_value_bias5: R.Tensor((2304,), dtype="float16") = packed_params[42] + gpt_neox_layers_3_attention_dense_weight5: R.Tensor((768, 768), dtype="float16") = packed_params[43] + gpt_neox_layers_3_attention_dense_bias5: R.Tensor((768,), dtype="float16") = packed_params[44] + gpt_neox_layers_3_mlp_dense_h_to_4h_weight5: R.Tensor((3072, 768), dtype="float16") = packed_params[45] + gpt_neox_layers_3_mlp_dense_h_to_4h_bias5: R.Tensor((3072,), dtype="float32") = packed_params[46] + gpt_neox_layers_3_mlp_dense_4h_to_h_weight5: R.Tensor((768, 3072), dtype="float16") = packed_params[47] + gpt_neox_layers_3_mlp_dense_4h_to_h_bias5: R.Tensor((768,), dtype="float32") = packed_params[48] + gpt_neox_layers_4_input_layernorm_weight5: R.Tensor((768,), dtype="float16") = packed_params[49] + gpt_neox_layers_4_input_layernorm_bias5: R.Tensor((768,), dtype="float16") = packed_params[50] + gpt_neox_layers_4_post_attention_layernorm_weight5: R.Tensor((768,), dtype="float16") = packed_params[51] + gpt_neox_layers_4_post_attention_layernorm_bias5: R.Tensor((768,), dtype="float16") = packed_params[52] + gpt_neox_layers_4_attention_query_key_value_weight5: R.Tensor((2304, 768), dtype="float16") = packed_params[53] + gpt_neox_layers_4_attention_query_key_value_bias5: R.Tensor((2304,), dtype="float16") = packed_params[54] + gpt_neox_layers_4_attention_dense_weight5: R.Tensor((768, 768), dtype="float16") = packed_params[55] + gpt_neox_layers_4_attention_dense_bias5: R.Tensor((768,), dtype="float16") = packed_params[56] + gpt_neox_layers_4_mlp_dense_h_to_4h_weight5: R.Tensor((3072, 768), dtype="float16") = packed_params[57] + gpt_neox_layers_4_mlp_dense_h_to_4h_bias5: R.Tensor((3072,), dtype="float32") = packed_params[58] + gpt_neox_layers_4_mlp_dense_4h_to_h_weight5: R.Tensor((768, 3072), dtype="float16") = packed_params[59] + gpt_neox_layers_4_mlp_dense_4h_to_h_bias5: R.Tensor((768,), dtype="float32") = packed_params[60] + gpt_neox_layers_5_input_layernorm_weight5: R.Tensor((768,), dtype="float16") = packed_params[61] + gpt_neox_layers_5_input_layernorm_bias5: R.Tensor((768,), dtype="float16") = packed_params[62] + gpt_neox_layers_5_post_attention_layernorm_weight5: R.Tensor((768,), dtype="float16") = packed_params[63] + gpt_neox_layers_5_post_attention_layernorm_bias5: R.Tensor((768,), dtype="float16") = packed_params[64] + gpt_neox_layers_5_attention_query_key_value_weight5: R.Tensor((2304, 768), dtype="float16") = packed_params[65] + gpt_neox_layers_5_attention_query_key_value_bias5: R.Tensor((2304,), dtype="float16") = packed_params[66] + gpt_neox_layers_5_attention_dense_weight5: R.Tensor((768, 768), dtype="float16") = packed_params[67] + gpt_neox_layers_5_attention_dense_bias5: R.Tensor((768,), dtype="float16") = packed_params[68] + gpt_neox_layers_5_mlp_dense_h_to_4h_weight5: R.Tensor((3072, 768), dtype="float16") = packed_params[69] + gpt_neox_layers_5_mlp_dense_h_to_4h_bias5: R.Tensor((3072,), dtype="float32") = packed_params[70] + gpt_neox_layers_5_mlp_dense_4h_to_h_weight5: R.Tensor((768, 3072), dtype="float16") = packed_params[71] + gpt_neox_layers_5_mlp_dense_4h_to_h_bias5: R.Tensor((768,), dtype="float32") = packed_params[72] + gpt_neox_layers_6_input_layernorm_weight5: R.Tensor((768,), dtype="float16") = packed_params[73] + gpt_neox_layers_6_input_layernorm_bias5: R.Tensor((768,), dtype="float16") = packed_params[74] + gpt_neox_layers_6_post_attention_layernorm_weight5: R.Tensor((768,), dtype="float16") = packed_params[75] + gpt_neox_layers_6_post_attention_layernorm_bias5: R.Tensor((768,), dtype="float16") = packed_params[76] + gpt_neox_layers_6_attention_query_key_value_weight5: R.Tensor((2304, 768), dtype="float16") = packed_params[77] + gpt_neox_layers_6_attention_query_key_value_bias5: R.Tensor((2304,), dtype="float16") = packed_params[78] + gpt_neox_layers_6_attention_dense_weight5: R.Tensor((768, 768), dtype="float16") = packed_params[79] + gpt_neox_layers_6_attention_dense_bias5: R.Tensor((768,), dtype="float16") = packed_params[80] + gpt_neox_layers_6_mlp_dense_h_to_4h_weight5: R.Tensor((3072, 768), dtype="float16") = packed_params[81] + gpt_neox_layers_6_mlp_dense_h_to_4h_bias5: R.Tensor((3072,), dtype="float32") = packed_params[82] + gpt_neox_layers_6_mlp_dense_4h_to_h_weight5: R.Tensor((768, 3072), dtype="float16") = packed_params[83] + gpt_neox_layers_6_mlp_dense_4h_to_h_bias5: R.Tensor((768,), dtype="float32") = packed_params[84] + gpt_neox_layers_7_input_layernorm_weight5: R.Tensor((768,), dtype="float16") = packed_params[85] + gpt_neox_layers_7_input_layernorm_bias5: R.Tensor((768,), dtype="float16") = packed_params[86] + gpt_neox_layers_7_post_attention_layernorm_weight5: R.Tensor((768,), dtype="float16") = packed_params[87] + gpt_neox_layers_7_post_attention_layernorm_bias5: R.Tensor((768,), dtype="float16") = packed_params[88] + gpt_neox_layers_7_attention_query_key_value_weight5: R.Tensor((2304, 768), dtype="float16") = packed_params[89] + gpt_neox_layers_7_attention_query_key_value_bias5: R.Tensor((2304,), dtype="float16") = packed_params[90] + gpt_neox_layers_7_attention_dense_weight5: R.Tensor((768, 768), dtype="float16") = packed_params[91] + gpt_neox_layers_7_attention_dense_bias5: R.Tensor((768,), dtype="float16") = packed_params[92] + gpt_neox_layers_7_mlp_dense_h_to_4h_weight5: R.Tensor((3072, 768), dtype="float16") = packed_params[93] + gpt_neox_layers_7_mlp_dense_h_to_4h_bias5: R.Tensor((3072,), dtype="float32") = packed_params[94] + gpt_neox_layers_7_mlp_dense_4h_to_h_weight5: R.Tensor((768, 3072), dtype="float16") = packed_params[95] + gpt_neox_layers_7_mlp_dense_4h_to_h_bias5: R.Tensor((768,), dtype="float32") = packed_params[96] + gpt_neox_layers_8_input_layernorm_weight5: R.Tensor((768,), dtype="float16") = packed_params[97] + gpt_neox_layers_8_input_layernorm_bias5: R.Tensor((768,), dtype="float16") = packed_params[98] + gpt_neox_layers_8_post_attention_layernorm_weight5: R.Tensor((768,), dtype="float16") = packed_params[99] + gpt_neox_layers_8_post_attention_layernorm_bias5: R.Tensor((768,), dtype="float16") = packed_params[100] + gpt_neox_layers_8_attention_query_key_value_weight5: R.Tensor((2304, 768), dtype="float16") = packed_params[101] + gpt_neox_layers_8_attention_query_key_value_bias5: R.Tensor((2304,), dtype="float16") = packed_params[102] + gpt_neox_layers_8_attention_dense_weight5: R.Tensor((768, 768), dtype="float16") = packed_params[103] + gpt_neox_layers_8_attention_dense_bias5: R.Tensor((768,), dtype="float16") = packed_params[104] + gpt_neox_layers_8_mlp_dense_h_to_4h_weight5: R.Tensor((3072, 768), dtype="float16") = packed_params[105] + gpt_neox_layers_8_mlp_dense_h_to_4h_bias5: R.Tensor((3072,), dtype="float32") = packed_params[106] + gpt_neox_layers_8_mlp_dense_4h_to_h_weight5: R.Tensor((768, 3072), dtype="float16") = packed_params[107] + gpt_neox_layers_8_mlp_dense_4h_to_h_bias5: R.Tensor((768,), dtype="float32") = packed_params[108] + gpt_neox_layers_9_input_layernorm_weight5: R.Tensor((768,), dtype="float16") = packed_params[109] + gpt_neox_layers_9_input_layernorm_bias5: R.Tensor((768,), dtype="float16") = packed_params[110] + gpt_neox_layers_9_post_attention_layernorm_weight5: R.Tensor((768,), dtype="float16") = packed_params[111] + gpt_neox_layers_9_post_attention_layernorm_bias5: R.Tensor((768,), dtype="float16") = packed_params[112] + gpt_neox_layers_9_attention_query_key_value_weight5: R.Tensor((2304, 768), dtype="float16") = packed_params[113] + gpt_neox_layers_9_attention_query_key_value_bias5: R.Tensor((2304,), dtype="float16") = packed_params[114] + gpt_neox_layers_9_attention_dense_weight5: R.Tensor((768, 768), dtype="float16") = packed_params[115] + gpt_neox_layers_9_attention_dense_bias5: R.Tensor((768,), dtype="float16") = packed_params[116] + gpt_neox_layers_9_mlp_dense_h_to_4h_weight5: R.Tensor((3072, 768), dtype="float16") = packed_params[117] + gpt_neox_layers_9_mlp_dense_h_to_4h_bias5: R.Tensor((3072,), dtype="float32") = packed_params[118] + gpt_neox_layers_9_mlp_dense_4h_to_h_weight5: R.Tensor((768, 3072), dtype="float16") = packed_params[119] + gpt_neox_layers_9_mlp_dense_4h_to_h_bias5: R.Tensor((768,), dtype="float32") = packed_params[120] + gpt_neox_layers_10_input_layernorm_weight5: R.Tensor((768,), dtype="float16") = packed_params[121] + gpt_neox_layers_10_input_layernorm_bias5: R.Tensor((768,), dtype="float16") = packed_params[122] + gpt_neox_layers_10_post_attention_layernorm_weight5: R.Tensor((768,), dtype="float16") = packed_params[123] + gpt_neox_layers_10_post_attention_layernorm_bias5: R.Tensor((768,), dtype="float16") = packed_params[124] + gpt_neox_layers_10_attention_query_key_value_weight5: R.Tensor((2304, 768), dtype="float16") = packed_params[125] + gpt_neox_layers_10_attention_query_key_value_bias5: R.Tensor((2304,), dtype="float16") = packed_params[126] + gpt_neox_layers_10_attention_dense_weight5: R.Tensor((768, 768), dtype="float16") = packed_params[127] + gpt_neox_layers_10_attention_dense_bias5: R.Tensor((768,), dtype="float16") = packed_params[128] + gpt_neox_layers_10_mlp_dense_h_to_4h_weight5: R.Tensor((3072, 768), dtype="float16") = packed_params[129] + gpt_neox_layers_10_mlp_dense_h_to_4h_bias5: R.Tensor((3072,), dtype="float32") = packed_params[130] + gpt_neox_layers_10_mlp_dense_4h_to_h_weight5: R.Tensor((768, 3072), dtype="float16") = packed_params[131] + gpt_neox_layers_10_mlp_dense_4h_to_h_bias5: R.Tensor((768,), dtype="float32") = packed_params[132] + gpt_neox_layers_11_input_layernorm_weight5: R.Tensor((768,), dtype="float16") = packed_params[133] + gpt_neox_layers_11_input_layernorm_bias5: R.Tensor((768,), dtype="float16") = packed_params[134] + gpt_neox_layers_11_post_attention_layernorm_weight5: R.Tensor((768,), dtype="float16") = packed_params[135] + gpt_neox_layers_11_post_attention_layernorm_bias5: R.Tensor((768,), dtype="float16") = packed_params[136] + gpt_neox_layers_11_attention_query_key_value_weight5: R.Tensor((2304, 768), dtype="float16") = packed_params[137] + gpt_neox_layers_11_attention_query_key_value_bias5: R.Tensor((2304,), dtype="float16") = packed_params[138] + gpt_neox_layers_11_attention_dense_weight5: R.Tensor((768, 768), dtype="float16") = packed_params[139] + gpt_neox_layers_11_attention_dense_bias5: R.Tensor((768,), dtype="float16") = packed_params[140] + gpt_neox_layers_11_mlp_dense_h_to_4h_weight5: R.Tensor((3072, 768), dtype="float16") = packed_params[141] + gpt_neox_layers_11_mlp_dense_h_to_4h_bias5: R.Tensor((3072,), dtype="float32") = packed_params[142] + gpt_neox_layers_11_mlp_dense_4h_to_h_weight5: R.Tensor((768, 3072), dtype="float16") = packed_params[143] + gpt_neox_layers_11_mlp_dense_4h_to_h_bias5: R.Tensor((768,), dtype="float32") = packed_params[144] + gpt_neox_final_layer_norm_weight5: R.Tensor((768,), dtype="float16") = packed_params[145] + gpt_neox_final_layer_norm_bias5: R.Tensor((768,), dtype="float16") = packed_params[146] + embed_out_weight5: R.Tensor((vocab_size, 768), dtype="float16") = packed_params[147] + layer_norm100: R.Tensor((1, seq_len, 768), dtype="float16") = R.nn.layer_norm(input_embeds, gpt_neox_layers_0_input_layernorm_weight5, gpt_neox_layers_0_input_layernorm_bias5, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims196: R.Tensor((768, 2304), dtype="float16") = R.permute_dims(gpt_neox_layers_0_attention_query_key_value_weight5, axes=None) + matmul196: R.Tensor((1, seq_len, 2304), dtype="float16") = R.matmul(layer_norm100, permute_dims196, out_dtype="void") + add288: R.Tensor((1, seq_len, 2304), dtype="float16") = R.add(matmul196, gpt_neox_layers_0_attention_query_key_value_bias5) + reshape192: R.Tensor((1, seq_len, 36, 64), dtype="float16") = R.reshape(add288, R.shape([1, seq_len, 36, 64])) + reshape193: R.Tensor((seq_len, 36, 64), dtype="float16") = R.reshape(reshape192, R.shape([seq_len, 36, 64])) + lv49 = R.call_dps_packed("vm.builtin.attention_kv_cache_attention_with_fused_qkv", (paged_kv_cache, R.prim_value(0), R.prim_value(T.float32(1.0)), reshape193), out_sinfo=R.Tensor((seq_len, 12, 64), dtype="float16")) + reshape194: R.Tensor((1, seq_len, 12, 64), dtype="float16") = R.reshape(lv49, R.shape([1, seq_len, 12, 64])) + reshape195: R.Tensor((1, seq_len, 768), dtype="float16") = R.reshape(reshape194, R.shape([1, seq_len, 768])) + permute_dims197: R.Tensor((768, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_0_attention_dense_weight5, axes=None) + matmul197: R.Tensor((1, seq_len, 768), dtype="float16") = R.matmul(reshape195, permute_dims197, out_dtype="void") + add289: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(matmul197, gpt_neox_layers_0_attention_dense_bias5) + layer_norm101: R.Tensor((1, seq_len, 768), dtype="float16") = R.nn.layer_norm(input_embeds, gpt_neox_layers_0_post_attention_layernorm_weight5, gpt_neox_layers_0_post_attention_layernorm_bias5, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims198: R.Tensor((768, 3072), dtype="float16") = R.permute_dims(gpt_neox_layers_0_mlp_dense_h_to_4h_weight5, axes=None) + matmul198: R.Tensor((1, seq_len, 3072), dtype="float32") = R.matmul(layer_norm101, permute_dims198, out_dtype="float32") + add290: R.Tensor((1, seq_len, 3072), dtype="float32") = R.add(matmul198, gpt_neox_layers_0_mlp_dense_h_to_4h_bias5) + gelu48: R.Tensor((1, seq_len, 3072), dtype="float32") = R.nn.gelu(add290) + astype100: R.Tensor((1, seq_len, 3072), dtype="float16") = R.astype(gelu48, dtype="float16") + permute_dims199: R.Tensor((3072, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_0_mlp_dense_4h_to_h_weight5, axes=None) + matmul199: R.Tensor((1, seq_len, 768), dtype="float32") = R.matmul(astype100, permute_dims199, out_dtype="float32") + add291: R.Tensor((1, seq_len, 768), dtype="float32") = R.add(matmul199, gpt_neox_layers_0_mlp_dense_4h_to_h_bias5) + astype101: R.Tensor((1, seq_len, 768), dtype="float16") = R.astype(add291, dtype="float16") + add292: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(astype101, add289) + add293: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(add292, input_embeds) + layer_norm102: R.Tensor((1, seq_len, 768), dtype="float16") = R.nn.layer_norm(add293, gpt_neox_layers_1_input_layernorm_weight5, gpt_neox_layers_1_input_layernorm_bias5, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims200: R.Tensor((768, 2304), dtype="float16") = R.permute_dims(gpt_neox_layers_1_attention_query_key_value_weight5, axes=None) + matmul200: R.Tensor((1, seq_len, 2304), dtype="float16") = R.matmul(layer_norm102, permute_dims200, out_dtype="void") + add294: R.Tensor((1, seq_len, 2304), dtype="float16") = R.add(matmul200, gpt_neox_layers_1_attention_query_key_value_bias5) + reshape196: R.Tensor((1, seq_len, 36, 64), dtype="float16") = R.reshape(add294, R.shape([1, seq_len, 36, 64])) + reshape197: R.Tensor((seq_len, 36, 64), dtype="float16") = R.reshape(reshape196, R.shape([seq_len, 36, 64])) + lv50 = R.call_dps_packed("vm.builtin.attention_kv_cache_attention_with_fused_qkv", (paged_kv_cache, R.prim_value(1), R.prim_value(T.float32(1.0)), reshape197), out_sinfo=R.Tensor((seq_len, 12, 64), dtype="float16")) + reshape198: R.Tensor((1, seq_len, 12, 64), dtype="float16") = R.reshape(lv50, R.shape([1, seq_len, 12, 64])) + reshape199: R.Tensor((1, seq_len, 768), dtype="float16") = R.reshape(reshape198, R.shape([1, seq_len, 768])) + permute_dims201: R.Tensor((768, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_1_attention_dense_weight5, axes=None) + matmul201: R.Tensor((1, seq_len, 768), dtype="float16") = R.matmul(reshape199, permute_dims201, out_dtype="void") + add295: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(matmul201, gpt_neox_layers_1_attention_dense_bias5) + layer_norm103: R.Tensor((1, seq_len, 768), dtype="float16") = R.nn.layer_norm(add293, gpt_neox_layers_1_post_attention_layernorm_weight5, gpt_neox_layers_1_post_attention_layernorm_bias5, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims202: R.Tensor((768, 3072), dtype="float16") = R.permute_dims(gpt_neox_layers_1_mlp_dense_h_to_4h_weight5, axes=None) + matmul202: R.Tensor((1, seq_len, 3072), dtype="float32") = R.matmul(layer_norm103, permute_dims202, out_dtype="float32") + add296: R.Tensor((1, seq_len, 3072), dtype="float32") = R.add(matmul202, gpt_neox_layers_1_mlp_dense_h_to_4h_bias5) + gelu49: R.Tensor((1, seq_len, 3072), dtype="float32") = R.nn.gelu(add296) + astype102: R.Tensor((1, seq_len, 3072), dtype="float16") = R.astype(gelu49, dtype="float16") + permute_dims203: R.Tensor((3072, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_1_mlp_dense_4h_to_h_weight5, axes=None) + matmul203: R.Tensor((1, seq_len, 768), dtype="float32") = R.matmul(astype102, permute_dims203, out_dtype="float32") + add297: R.Tensor((1, seq_len, 768), dtype="float32") = R.add(matmul203, gpt_neox_layers_1_mlp_dense_4h_to_h_bias5) + astype103: R.Tensor((1, seq_len, 768), dtype="float16") = R.astype(add297, dtype="float16") + add298: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(astype103, add295) + add299: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(add298, add293) + layer_norm104: R.Tensor((1, seq_len, 768), dtype="float16") = R.nn.layer_norm(add299, gpt_neox_layers_2_input_layernorm_weight5, gpt_neox_layers_2_input_layernorm_bias5, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims204: R.Tensor((768, 2304), dtype="float16") = R.permute_dims(gpt_neox_layers_2_attention_query_key_value_weight5, axes=None) + matmul204: R.Tensor((1, seq_len, 2304), dtype="float16") = R.matmul(layer_norm104, permute_dims204, out_dtype="void") + add300: R.Tensor((1, seq_len, 2304), dtype="float16") = R.add(matmul204, gpt_neox_layers_2_attention_query_key_value_bias5) + reshape200: R.Tensor((1, seq_len, 36, 64), dtype="float16") = R.reshape(add300, R.shape([1, seq_len, 36, 64])) + reshape201: R.Tensor((seq_len, 36, 64), dtype="float16") = R.reshape(reshape200, R.shape([seq_len, 36, 64])) + lv51 = R.call_dps_packed("vm.builtin.attention_kv_cache_attention_with_fused_qkv", (paged_kv_cache, R.prim_value(2), R.prim_value(T.float32(1.0)), reshape201), out_sinfo=R.Tensor((seq_len, 12, 64), dtype="float16")) + reshape202: R.Tensor((1, seq_len, 12, 64), dtype="float16") = R.reshape(lv51, R.shape([1, seq_len, 12, 64])) + reshape203: R.Tensor((1, seq_len, 768), dtype="float16") = R.reshape(reshape202, R.shape([1, seq_len, 768])) + permute_dims205: R.Tensor((768, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_2_attention_dense_weight5, axes=None) + matmul205: R.Tensor((1, seq_len, 768), dtype="float16") = R.matmul(reshape203, permute_dims205, out_dtype="void") + add301: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(matmul205, gpt_neox_layers_2_attention_dense_bias5) + layer_norm105: R.Tensor((1, seq_len, 768), dtype="float16") = R.nn.layer_norm(add299, gpt_neox_layers_2_post_attention_layernorm_weight5, gpt_neox_layers_2_post_attention_layernorm_bias5, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims206: R.Tensor((768, 3072), dtype="float16") = R.permute_dims(gpt_neox_layers_2_mlp_dense_h_to_4h_weight5, axes=None) + matmul206: R.Tensor((1, seq_len, 3072), dtype="float32") = R.matmul(layer_norm105, permute_dims206, out_dtype="float32") + add302: R.Tensor((1, seq_len, 3072), dtype="float32") = R.add(matmul206, gpt_neox_layers_2_mlp_dense_h_to_4h_bias5) + gelu50: R.Tensor((1, seq_len, 3072), dtype="float32") = R.nn.gelu(add302) + astype104: R.Tensor((1, seq_len, 3072), dtype="float16") = R.astype(gelu50, dtype="float16") + permute_dims207: R.Tensor((3072, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_2_mlp_dense_4h_to_h_weight5, axes=None) + matmul207: R.Tensor((1, seq_len, 768), dtype="float32") = R.matmul(astype104, permute_dims207, out_dtype="float32") + add303: R.Tensor((1, seq_len, 768), dtype="float32") = R.add(matmul207, gpt_neox_layers_2_mlp_dense_4h_to_h_bias5) + astype105: R.Tensor((1, seq_len, 768), dtype="float16") = R.astype(add303, dtype="float16") + add304: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(astype105, add301) + add305: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(add304, add299) + layer_norm106: R.Tensor((1, seq_len, 768), dtype="float16") = R.nn.layer_norm(add305, gpt_neox_layers_3_input_layernorm_weight5, gpt_neox_layers_3_input_layernorm_bias5, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims208: R.Tensor((768, 2304), dtype="float16") = R.permute_dims(gpt_neox_layers_3_attention_query_key_value_weight5, axes=None) + matmul208: R.Tensor((1, seq_len, 2304), dtype="float16") = R.matmul(layer_norm106, permute_dims208, out_dtype="void") + add306: R.Tensor((1, seq_len, 2304), dtype="float16") = R.add(matmul208, gpt_neox_layers_3_attention_query_key_value_bias5) + reshape204: R.Tensor((1, seq_len, 36, 64), dtype="float16") = R.reshape(add306, R.shape([1, seq_len, 36, 64])) + reshape205: R.Tensor((seq_len, 36, 64), dtype="float16") = R.reshape(reshape204, R.shape([seq_len, 36, 64])) + lv52 = R.call_dps_packed("vm.builtin.attention_kv_cache_attention_with_fused_qkv", (paged_kv_cache, R.prim_value(3), R.prim_value(T.float32(1.0)), reshape205), out_sinfo=R.Tensor((seq_len, 12, 64), dtype="float16")) + reshape206: R.Tensor((1, seq_len, 12, 64), dtype="float16") = R.reshape(lv52, R.shape([1, seq_len, 12, 64])) + reshape207: R.Tensor((1, seq_len, 768), dtype="float16") = R.reshape(reshape206, R.shape([1, seq_len, 768])) + permute_dims209: R.Tensor((768, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_3_attention_dense_weight5, axes=None) + matmul209: R.Tensor((1, seq_len, 768), dtype="float16") = R.matmul(reshape207, permute_dims209, out_dtype="void") + add307: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(matmul209, gpt_neox_layers_3_attention_dense_bias5) + layer_norm107: R.Tensor((1, seq_len, 768), dtype="float16") = R.nn.layer_norm(add305, gpt_neox_layers_3_post_attention_layernorm_weight5, gpt_neox_layers_3_post_attention_layernorm_bias5, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims210: R.Tensor((768, 3072), dtype="float16") = R.permute_dims(gpt_neox_layers_3_mlp_dense_h_to_4h_weight5, axes=None) + matmul210: R.Tensor((1, seq_len, 3072), dtype="float32") = R.matmul(layer_norm107, permute_dims210, out_dtype="float32") + add308: R.Tensor((1, seq_len, 3072), dtype="float32") = R.add(matmul210, gpt_neox_layers_3_mlp_dense_h_to_4h_bias5) + gelu51: R.Tensor((1, seq_len, 3072), dtype="float32") = R.nn.gelu(add308) + astype106: R.Tensor((1, seq_len, 3072), dtype="float16") = R.astype(gelu51, dtype="float16") + permute_dims211: R.Tensor((3072, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_3_mlp_dense_4h_to_h_weight5, axes=None) + matmul211: R.Tensor((1, seq_len, 768), dtype="float32") = R.matmul(astype106, permute_dims211, out_dtype="float32") + add309: R.Tensor((1, seq_len, 768), dtype="float32") = R.add(matmul211, gpt_neox_layers_3_mlp_dense_4h_to_h_bias5) + astype107: R.Tensor((1, seq_len, 768), dtype="float16") = R.astype(add309, dtype="float16") + add310: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(astype107, add307) + add311: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(add310, add305) + layer_norm108: R.Tensor((1, seq_len, 768), dtype="float16") = R.nn.layer_norm(add311, gpt_neox_layers_4_input_layernorm_weight5, gpt_neox_layers_4_input_layernorm_bias5, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims212: R.Tensor((768, 2304), dtype="float16") = R.permute_dims(gpt_neox_layers_4_attention_query_key_value_weight5, axes=None) + matmul212: R.Tensor((1, seq_len, 2304), dtype="float16") = R.matmul(layer_norm108, permute_dims212, out_dtype="void") + add312: R.Tensor((1, seq_len, 2304), dtype="float16") = R.add(matmul212, gpt_neox_layers_4_attention_query_key_value_bias5) + reshape208: R.Tensor((1, seq_len, 36, 64), dtype="float16") = R.reshape(add312, R.shape([1, seq_len, 36, 64])) + reshape209: R.Tensor((seq_len, 36, 64), dtype="float16") = R.reshape(reshape208, R.shape([seq_len, 36, 64])) + lv53 = R.call_dps_packed("vm.builtin.attention_kv_cache_attention_with_fused_qkv", (paged_kv_cache, R.prim_value(4), R.prim_value(T.float32(1.0)), reshape209), out_sinfo=R.Tensor((seq_len, 12, 64), dtype="float16")) + reshape210: R.Tensor((1, seq_len, 12, 64), dtype="float16") = R.reshape(lv53, R.shape([1, seq_len, 12, 64])) + reshape211: R.Tensor((1, seq_len, 768), dtype="float16") = R.reshape(reshape210, R.shape([1, seq_len, 768])) + permute_dims213: R.Tensor((768, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_4_attention_dense_weight5, axes=None) + matmul213: R.Tensor((1, seq_len, 768), dtype="float16") = R.matmul(reshape211, permute_dims213, out_dtype="void") + add313: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(matmul213, gpt_neox_layers_4_attention_dense_bias5) + layer_norm109: R.Tensor((1, seq_len, 768), dtype="float16") = R.nn.layer_norm(add311, gpt_neox_layers_4_post_attention_layernorm_weight5, gpt_neox_layers_4_post_attention_layernorm_bias5, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims214: R.Tensor((768, 3072), dtype="float16") = R.permute_dims(gpt_neox_layers_4_mlp_dense_h_to_4h_weight5, axes=None) + matmul214: R.Tensor((1, seq_len, 3072), dtype="float32") = R.matmul(layer_norm109, permute_dims214, out_dtype="float32") + add314: R.Tensor((1, seq_len, 3072), dtype="float32") = R.add(matmul214, gpt_neox_layers_4_mlp_dense_h_to_4h_bias5) + gelu52: R.Tensor((1, seq_len, 3072), dtype="float32") = R.nn.gelu(add314) + astype108: R.Tensor((1, seq_len, 3072), dtype="float16") = R.astype(gelu52, dtype="float16") + permute_dims215: R.Tensor((3072, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_4_mlp_dense_4h_to_h_weight5, axes=None) + matmul215: R.Tensor((1, seq_len, 768), dtype="float32") = R.matmul(astype108, permute_dims215, out_dtype="float32") + add315: R.Tensor((1, seq_len, 768), dtype="float32") = R.add(matmul215, gpt_neox_layers_4_mlp_dense_4h_to_h_bias5) + astype109: R.Tensor((1, seq_len, 768), dtype="float16") = R.astype(add315, dtype="float16") + add316: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(astype109, add313) + add317: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(add316, add311) + layer_norm110: R.Tensor((1, seq_len, 768), dtype="float16") = R.nn.layer_norm(add317, gpt_neox_layers_5_input_layernorm_weight5, gpt_neox_layers_5_input_layernorm_bias5, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims216: R.Tensor((768, 2304), dtype="float16") = R.permute_dims(gpt_neox_layers_5_attention_query_key_value_weight5, axes=None) + matmul216: R.Tensor((1, seq_len, 2304), dtype="float16") = R.matmul(layer_norm110, permute_dims216, out_dtype="void") + add318: R.Tensor((1, seq_len, 2304), dtype="float16") = R.add(matmul216, gpt_neox_layers_5_attention_query_key_value_bias5) + reshape212: R.Tensor((1, seq_len, 36, 64), dtype="float16") = R.reshape(add318, R.shape([1, seq_len, 36, 64])) + reshape213: R.Tensor((seq_len, 36, 64), dtype="float16") = R.reshape(reshape212, R.shape([seq_len, 36, 64])) + lv54 = R.call_dps_packed("vm.builtin.attention_kv_cache_attention_with_fused_qkv", (paged_kv_cache, R.prim_value(5), R.prim_value(T.float32(1.0)), reshape213), out_sinfo=R.Tensor((seq_len, 12, 64), dtype="float16")) + reshape214: R.Tensor((1, seq_len, 12, 64), dtype="float16") = R.reshape(lv54, R.shape([1, seq_len, 12, 64])) + reshape215: R.Tensor((1, seq_len, 768), dtype="float16") = R.reshape(reshape214, R.shape([1, seq_len, 768])) + permute_dims217: R.Tensor((768, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_5_attention_dense_weight5, axes=None) + matmul217: R.Tensor((1, seq_len, 768), dtype="float16") = R.matmul(reshape215, permute_dims217, out_dtype="void") + add319: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(matmul217, gpt_neox_layers_5_attention_dense_bias5) + layer_norm111: R.Tensor((1, seq_len, 768), dtype="float16") = R.nn.layer_norm(add317, gpt_neox_layers_5_post_attention_layernorm_weight5, gpt_neox_layers_5_post_attention_layernorm_bias5, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims218: R.Tensor((768, 3072), dtype="float16") = R.permute_dims(gpt_neox_layers_5_mlp_dense_h_to_4h_weight5, axes=None) + matmul218: R.Tensor((1, seq_len, 3072), dtype="float32") = R.matmul(layer_norm111, permute_dims218, out_dtype="float32") + add320: R.Tensor((1, seq_len, 3072), dtype="float32") = R.add(matmul218, gpt_neox_layers_5_mlp_dense_h_to_4h_bias5) + gelu53: R.Tensor((1, seq_len, 3072), dtype="float32") = R.nn.gelu(add320) + astype110: R.Tensor((1, seq_len, 3072), dtype="float16") = R.astype(gelu53, dtype="float16") + permute_dims219: R.Tensor((3072, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_5_mlp_dense_4h_to_h_weight5, axes=None) + matmul219: R.Tensor((1, seq_len, 768), dtype="float32") = R.matmul(astype110, permute_dims219, out_dtype="float32") + add321: R.Tensor((1, seq_len, 768), dtype="float32") = R.add(matmul219, gpt_neox_layers_5_mlp_dense_4h_to_h_bias5) + astype111: R.Tensor((1, seq_len, 768), dtype="float16") = R.astype(add321, dtype="float16") + add322: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(astype111, add319) + add323: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(add322, add317) + layer_norm112: R.Tensor((1, seq_len, 768), dtype="float16") = R.nn.layer_norm(add323, gpt_neox_layers_6_input_layernorm_weight5, gpt_neox_layers_6_input_layernorm_bias5, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims220: R.Tensor((768, 2304), dtype="float16") = R.permute_dims(gpt_neox_layers_6_attention_query_key_value_weight5, axes=None) + matmul220: R.Tensor((1, seq_len, 2304), dtype="float16") = R.matmul(layer_norm112, permute_dims220, out_dtype="void") + add324: R.Tensor((1, seq_len, 2304), dtype="float16") = R.add(matmul220, gpt_neox_layers_6_attention_query_key_value_bias5) + reshape216: R.Tensor((1, seq_len, 36, 64), dtype="float16") = R.reshape(add324, R.shape([1, seq_len, 36, 64])) + reshape217: R.Tensor((seq_len, 36, 64), dtype="float16") = R.reshape(reshape216, R.shape([seq_len, 36, 64])) + lv55 = R.call_dps_packed("vm.builtin.attention_kv_cache_attention_with_fused_qkv", (paged_kv_cache, R.prim_value(6), R.prim_value(T.float32(1.0)), reshape217), out_sinfo=R.Tensor((seq_len, 12, 64), dtype="float16")) + reshape218: R.Tensor((1, seq_len, 12, 64), dtype="float16") = R.reshape(lv55, R.shape([1, seq_len, 12, 64])) + reshape219: R.Tensor((1, seq_len, 768), dtype="float16") = R.reshape(reshape218, R.shape([1, seq_len, 768])) + permute_dims221: R.Tensor((768, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_6_attention_dense_weight5, axes=None) + matmul221: R.Tensor((1, seq_len, 768), dtype="float16") = R.matmul(reshape219, permute_dims221, out_dtype="void") + add325: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(matmul221, gpt_neox_layers_6_attention_dense_bias5) + layer_norm113: R.Tensor((1, seq_len, 768), dtype="float16") = R.nn.layer_norm(add323, gpt_neox_layers_6_post_attention_layernorm_weight5, gpt_neox_layers_6_post_attention_layernorm_bias5, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims222: R.Tensor((768, 3072), dtype="float16") = R.permute_dims(gpt_neox_layers_6_mlp_dense_h_to_4h_weight5, axes=None) + matmul222: R.Tensor((1, seq_len, 3072), dtype="float32") = R.matmul(layer_norm113, permute_dims222, out_dtype="float32") + add326: R.Tensor((1, seq_len, 3072), dtype="float32") = R.add(matmul222, gpt_neox_layers_6_mlp_dense_h_to_4h_bias5) + gelu54: R.Tensor((1, seq_len, 3072), dtype="float32") = R.nn.gelu(add326) + astype112: R.Tensor((1, seq_len, 3072), dtype="float16") = R.astype(gelu54, dtype="float16") + permute_dims223: R.Tensor((3072, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_6_mlp_dense_4h_to_h_weight5, axes=None) + matmul223: R.Tensor((1, seq_len, 768), dtype="float32") = R.matmul(astype112, permute_dims223, out_dtype="float32") + add327: R.Tensor((1, seq_len, 768), dtype="float32") = R.add(matmul223, gpt_neox_layers_6_mlp_dense_4h_to_h_bias5) + astype113: R.Tensor((1, seq_len, 768), dtype="float16") = R.astype(add327, dtype="float16") + add328: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(astype113, add325) + add329: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(add328, add323) + layer_norm114: R.Tensor((1, seq_len, 768), dtype="float16") = R.nn.layer_norm(add329, gpt_neox_layers_7_input_layernorm_weight5, gpt_neox_layers_7_input_layernorm_bias5, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims224: R.Tensor((768, 2304), dtype="float16") = R.permute_dims(gpt_neox_layers_7_attention_query_key_value_weight5, axes=None) + matmul224: R.Tensor((1, seq_len, 2304), dtype="float16") = R.matmul(layer_norm114, permute_dims224, out_dtype="void") + add330: R.Tensor((1, seq_len, 2304), dtype="float16") = R.add(matmul224, gpt_neox_layers_7_attention_query_key_value_bias5) + reshape220: R.Tensor((1, seq_len, 36, 64), dtype="float16") = R.reshape(add330, R.shape([1, seq_len, 36, 64])) + reshape221: R.Tensor((seq_len, 36, 64), dtype="float16") = R.reshape(reshape220, R.shape([seq_len, 36, 64])) + lv56 = R.call_dps_packed("vm.builtin.attention_kv_cache_attention_with_fused_qkv", (paged_kv_cache, R.prim_value(7), R.prim_value(T.float32(1.0)), reshape221), out_sinfo=R.Tensor((seq_len, 12, 64), dtype="float16")) + reshape222: R.Tensor((1, seq_len, 12, 64), dtype="float16") = R.reshape(lv56, R.shape([1, seq_len, 12, 64])) + reshape223: R.Tensor((1, seq_len, 768), dtype="float16") = R.reshape(reshape222, R.shape([1, seq_len, 768])) + permute_dims225: R.Tensor((768, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_7_attention_dense_weight5, axes=None) + matmul225: R.Tensor((1, seq_len, 768), dtype="float16") = R.matmul(reshape223, permute_dims225, out_dtype="void") + add331: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(matmul225, gpt_neox_layers_7_attention_dense_bias5) + layer_norm115: R.Tensor((1, seq_len, 768), dtype="float16") = R.nn.layer_norm(add329, gpt_neox_layers_7_post_attention_layernorm_weight5, gpt_neox_layers_7_post_attention_layernorm_bias5, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims226: R.Tensor((768, 3072), dtype="float16") = R.permute_dims(gpt_neox_layers_7_mlp_dense_h_to_4h_weight5, axes=None) + matmul226: R.Tensor((1, seq_len, 3072), dtype="float32") = R.matmul(layer_norm115, permute_dims226, out_dtype="float32") + add332: R.Tensor((1, seq_len, 3072), dtype="float32") = R.add(matmul226, gpt_neox_layers_7_mlp_dense_h_to_4h_bias5) + gelu55: R.Tensor((1, seq_len, 3072), dtype="float32") = R.nn.gelu(add332) + astype114: R.Tensor((1, seq_len, 3072), dtype="float16") = R.astype(gelu55, dtype="float16") + permute_dims227: R.Tensor((3072, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_7_mlp_dense_4h_to_h_weight5, axes=None) + matmul227: R.Tensor((1, seq_len, 768), dtype="float32") = R.matmul(astype114, permute_dims227, out_dtype="float32") + add333: R.Tensor((1, seq_len, 768), dtype="float32") = R.add(matmul227, gpt_neox_layers_7_mlp_dense_4h_to_h_bias5) + astype115: R.Tensor((1, seq_len, 768), dtype="float16") = R.astype(add333, dtype="float16") + add334: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(astype115, add331) + add335: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(add334, add329) + layer_norm116: R.Tensor((1, seq_len, 768), dtype="float16") = R.nn.layer_norm(add335, gpt_neox_layers_8_input_layernorm_weight5, gpt_neox_layers_8_input_layernorm_bias5, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims228: R.Tensor((768, 2304), dtype="float16") = R.permute_dims(gpt_neox_layers_8_attention_query_key_value_weight5, axes=None) + matmul228: R.Tensor((1, seq_len, 2304), dtype="float16") = R.matmul(layer_norm116, permute_dims228, out_dtype="void") + add336: R.Tensor((1, seq_len, 2304), dtype="float16") = R.add(matmul228, gpt_neox_layers_8_attention_query_key_value_bias5) + reshape224: R.Tensor((1, seq_len, 36, 64), dtype="float16") = R.reshape(add336, R.shape([1, seq_len, 36, 64])) + reshape225: R.Tensor((seq_len, 36, 64), dtype="float16") = R.reshape(reshape224, R.shape([seq_len, 36, 64])) + lv57 = R.call_dps_packed("vm.builtin.attention_kv_cache_attention_with_fused_qkv", (paged_kv_cache, R.prim_value(8), R.prim_value(T.float32(1.0)), reshape225), out_sinfo=R.Tensor((seq_len, 12, 64), dtype="float16")) + reshape226: R.Tensor((1, seq_len, 12, 64), dtype="float16") = R.reshape(lv57, R.shape([1, seq_len, 12, 64])) + reshape227: R.Tensor((1, seq_len, 768), dtype="float16") = R.reshape(reshape226, R.shape([1, seq_len, 768])) + permute_dims229: R.Tensor((768, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_8_attention_dense_weight5, axes=None) + matmul229: R.Tensor((1, seq_len, 768), dtype="float16") = R.matmul(reshape227, permute_dims229, out_dtype="void") + add337: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(matmul229, gpt_neox_layers_8_attention_dense_bias5) + layer_norm117: R.Tensor((1, seq_len, 768), dtype="float16") = R.nn.layer_norm(add335, gpt_neox_layers_8_post_attention_layernorm_weight5, gpt_neox_layers_8_post_attention_layernorm_bias5, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims230: R.Tensor((768, 3072), dtype="float16") = R.permute_dims(gpt_neox_layers_8_mlp_dense_h_to_4h_weight5, axes=None) + matmul230: R.Tensor((1, seq_len, 3072), dtype="float32") = R.matmul(layer_norm117, permute_dims230, out_dtype="float32") + add338: R.Tensor((1, seq_len, 3072), dtype="float32") = R.add(matmul230, gpt_neox_layers_8_mlp_dense_h_to_4h_bias5) + gelu56: R.Tensor((1, seq_len, 3072), dtype="float32") = R.nn.gelu(add338) + astype116: R.Tensor((1, seq_len, 3072), dtype="float16") = R.astype(gelu56, dtype="float16") + permute_dims231: R.Tensor((3072, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_8_mlp_dense_4h_to_h_weight5, axes=None) + matmul231: R.Tensor((1, seq_len, 768), dtype="float32") = R.matmul(astype116, permute_dims231, out_dtype="float32") + add339: R.Tensor((1, seq_len, 768), dtype="float32") = R.add(matmul231, gpt_neox_layers_8_mlp_dense_4h_to_h_bias5) + astype117: R.Tensor((1, seq_len, 768), dtype="float16") = R.astype(add339, dtype="float16") + add340: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(astype117, add337) + add341: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(add340, add335) + layer_norm118: R.Tensor((1, seq_len, 768), dtype="float16") = R.nn.layer_norm(add341, gpt_neox_layers_9_input_layernorm_weight5, gpt_neox_layers_9_input_layernorm_bias5, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims232: R.Tensor((768, 2304), dtype="float16") = R.permute_dims(gpt_neox_layers_9_attention_query_key_value_weight5, axes=None) + matmul232: R.Tensor((1, seq_len, 2304), dtype="float16") = R.matmul(layer_norm118, permute_dims232, out_dtype="void") + add342: R.Tensor((1, seq_len, 2304), dtype="float16") = R.add(matmul232, gpt_neox_layers_9_attention_query_key_value_bias5) + reshape228: R.Tensor((1, seq_len, 36, 64), dtype="float16") = R.reshape(add342, R.shape([1, seq_len, 36, 64])) + reshape229: R.Tensor((seq_len, 36, 64), dtype="float16") = R.reshape(reshape228, R.shape([seq_len, 36, 64])) + lv58 = R.call_dps_packed("vm.builtin.attention_kv_cache_attention_with_fused_qkv", (paged_kv_cache, R.prim_value(9), R.prim_value(T.float32(1.0)), reshape229), out_sinfo=R.Tensor((seq_len, 12, 64), dtype="float16")) + reshape230: R.Tensor((1, seq_len, 12, 64), dtype="float16") = R.reshape(lv58, R.shape([1, seq_len, 12, 64])) + reshape231: R.Tensor((1, seq_len, 768), dtype="float16") = R.reshape(reshape230, R.shape([1, seq_len, 768])) + permute_dims233: R.Tensor((768, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_9_attention_dense_weight5, axes=None) + matmul233: R.Tensor((1, seq_len, 768), dtype="float16") = R.matmul(reshape231, permute_dims233, out_dtype="void") + add343: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(matmul233, gpt_neox_layers_9_attention_dense_bias5) + layer_norm119: R.Tensor((1, seq_len, 768), dtype="float16") = R.nn.layer_norm(add341, gpt_neox_layers_9_post_attention_layernorm_weight5, gpt_neox_layers_9_post_attention_layernorm_bias5, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims234: R.Tensor((768, 3072), dtype="float16") = R.permute_dims(gpt_neox_layers_9_mlp_dense_h_to_4h_weight5, axes=None) + matmul234: R.Tensor((1, seq_len, 3072), dtype="float32") = R.matmul(layer_norm119, permute_dims234, out_dtype="float32") + add344: R.Tensor((1, seq_len, 3072), dtype="float32") = R.add(matmul234, gpt_neox_layers_9_mlp_dense_h_to_4h_bias5) + gelu57: R.Tensor((1, seq_len, 3072), dtype="float32") = R.nn.gelu(add344) + astype118: R.Tensor((1, seq_len, 3072), dtype="float16") = R.astype(gelu57, dtype="float16") + permute_dims235: R.Tensor((3072, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_9_mlp_dense_4h_to_h_weight5, axes=None) + matmul235: R.Tensor((1, seq_len, 768), dtype="float32") = R.matmul(astype118, permute_dims235, out_dtype="float32") + add345: R.Tensor((1, seq_len, 768), dtype="float32") = R.add(matmul235, gpt_neox_layers_9_mlp_dense_4h_to_h_bias5) + astype119: R.Tensor((1, seq_len, 768), dtype="float16") = R.astype(add345, dtype="float16") + add346: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(astype119, add343) + add347: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(add346, add341) + layer_norm120: R.Tensor((1, seq_len, 768), dtype="float16") = R.nn.layer_norm(add347, gpt_neox_layers_10_input_layernorm_weight5, gpt_neox_layers_10_input_layernorm_bias5, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims236: R.Tensor((768, 2304), dtype="float16") = R.permute_dims(gpt_neox_layers_10_attention_query_key_value_weight5, axes=None) + matmul236: R.Tensor((1, seq_len, 2304), dtype="float16") = R.matmul(layer_norm120, permute_dims236, out_dtype="void") + add348: R.Tensor((1, seq_len, 2304), dtype="float16") = R.add(matmul236, gpt_neox_layers_10_attention_query_key_value_bias5) + reshape232: R.Tensor((1, seq_len, 36, 64), dtype="float16") = R.reshape(add348, R.shape([1, seq_len, 36, 64])) + reshape233: R.Tensor((seq_len, 36, 64), dtype="float16") = R.reshape(reshape232, R.shape([seq_len, 36, 64])) + lv59 = R.call_dps_packed("vm.builtin.attention_kv_cache_attention_with_fused_qkv", (paged_kv_cache, R.prim_value(10), R.prim_value(T.float32(1.0)), reshape233), out_sinfo=R.Tensor((seq_len, 12, 64), dtype="float16")) + reshape234: R.Tensor((1, seq_len, 12, 64), dtype="float16") = R.reshape(lv59, R.shape([1, seq_len, 12, 64])) + reshape235: R.Tensor((1, seq_len, 768), dtype="float16") = R.reshape(reshape234, R.shape([1, seq_len, 768])) + permute_dims237: R.Tensor((768, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_10_attention_dense_weight5, axes=None) + matmul237: R.Tensor((1, seq_len, 768), dtype="float16") = R.matmul(reshape235, permute_dims237, out_dtype="void") + add349: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(matmul237, gpt_neox_layers_10_attention_dense_bias5) + layer_norm121: R.Tensor((1, seq_len, 768), dtype="float16") = R.nn.layer_norm(add347, gpt_neox_layers_10_post_attention_layernorm_weight5, gpt_neox_layers_10_post_attention_layernorm_bias5, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims238: R.Tensor((768, 3072), dtype="float16") = R.permute_dims(gpt_neox_layers_10_mlp_dense_h_to_4h_weight5, axes=None) + matmul238: R.Tensor((1, seq_len, 3072), dtype="float32") = R.matmul(layer_norm121, permute_dims238, out_dtype="float32") + add350: R.Tensor((1, seq_len, 3072), dtype="float32") = R.add(matmul238, gpt_neox_layers_10_mlp_dense_h_to_4h_bias5) + gelu58: R.Tensor((1, seq_len, 3072), dtype="float32") = R.nn.gelu(add350) + astype120: R.Tensor((1, seq_len, 3072), dtype="float16") = R.astype(gelu58, dtype="float16") + permute_dims239: R.Tensor((3072, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_10_mlp_dense_4h_to_h_weight5, axes=None) + matmul239: R.Tensor((1, seq_len, 768), dtype="float32") = R.matmul(astype120, permute_dims239, out_dtype="float32") + add351: R.Tensor((1, seq_len, 768), dtype="float32") = R.add(matmul239, gpt_neox_layers_10_mlp_dense_4h_to_h_bias5) + astype121: R.Tensor((1, seq_len, 768), dtype="float16") = R.astype(add351, dtype="float16") + add352: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(astype121, add349) + add353: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(add352, add347) + layer_norm122: R.Tensor((1, seq_len, 768), dtype="float16") = R.nn.layer_norm(add353, gpt_neox_layers_11_input_layernorm_weight5, gpt_neox_layers_11_input_layernorm_bias5, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims240: R.Tensor((768, 2304), dtype="float16") = R.permute_dims(gpt_neox_layers_11_attention_query_key_value_weight5, axes=None) + matmul240: R.Tensor((1, seq_len, 2304), dtype="float16") = R.matmul(layer_norm122, permute_dims240, out_dtype="void") + add354: R.Tensor((1, seq_len, 2304), dtype="float16") = R.add(matmul240, gpt_neox_layers_11_attention_query_key_value_bias5) + reshape236: R.Tensor((1, seq_len, 36, 64), dtype="float16") = R.reshape(add354, R.shape([1, seq_len, 36, 64])) + reshape237: R.Tensor((seq_len, 36, 64), dtype="float16") = R.reshape(reshape236, R.shape([seq_len, 36, 64])) + lv60 = R.call_dps_packed("vm.builtin.attention_kv_cache_attention_with_fused_qkv", (paged_kv_cache, R.prim_value(11), R.prim_value(T.float32(1.0)), reshape237), out_sinfo=R.Tensor((seq_len, 12, 64), dtype="float16")) + reshape238: R.Tensor((1, seq_len, 12, 64), dtype="float16") = R.reshape(lv60, R.shape([1, seq_len, 12, 64])) + reshape239: R.Tensor((1, seq_len, 768), dtype="float16") = R.reshape(reshape238, R.shape([1, seq_len, 768])) + permute_dims241: R.Tensor((768, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_11_attention_dense_weight5, axes=None) + matmul241: R.Tensor((1, seq_len, 768), dtype="float16") = R.matmul(reshape239, permute_dims241, out_dtype="void") + add355: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(matmul241, gpt_neox_layers_11_attention_dense_bias5) + layer_norm123: R.Tensor((1, seq_len, 768), dtype="float16") = R.nn.layer_norm(add353, gpt_neox_layers_11_post_attention_layernorm_weight5, gpt_neox_layers_11_post_attention_layernorm_bias5, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims242: R.Tensor((768, 3072), dtype="float16") = R.permute_dims(gpt_neox_layers_11_mlp_dense_h_to_4h_weight5, axes=None) + matmul242: R.Tensor((1, seq_len, 3072), dtype="float32") = R.matmul(layer_norm123, permute_dims242, out_dtype="float32") + add356: R.Tensor((1, seq_len, 3072), dtype="float32") = R.add(matmul242, gpt_neox_layers_11_mlp_dense_h_to_4h_bias5) + gelu59: R.Tensor((1, seq_len, 3072), dtype="float32") = R.nn.gelu(add356) + astype122: R.Tensor((1, seq_len, 3072), dtype="float16") = R.astype(gelu59, dtype="float16") + permute_dims243: R.Tensor((3072, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_11_mlp_dense_4h_to_h_weight5, axes=None) + matmul243: R.Tensor((1, seq_len, 768), dtype="float32") = R.matmul(astype122, permute_dims243, out_dtype="float32") + add357: R.Tensor((1, seq_len, 768), dtype="float32") = R.add(matmul243, gpt_neox_layers_11_mlp_dense_4h_to_h_bias5) + astype123: R.Tensor((1, seq_len, 768), dtype="float16") = R.astype(add357, dtype="float16") + add358: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(astype123, add355) + add359: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(add358, add353) + layer_norm124: R.Tensor((1, seq_len, 768), dtype="float16") = R.nn.layer_norm(add359, gpt_neox_final_layer_norm_weight5, gpt_neox_final_layer_norm_bias5, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims244: R.Tensor((768, vocab_size), dtype="float16") = R.permute_dims(embed_out_weight5, axes=None) + matmul244: R.Tensor((1, seq_len, vocab_size), dtype="float16") = R.matmul(layer_norm124, permute_dims244, out_dtype="void") + astype124: R.Tensor((1, seq_len, vocab_size), dtype="float32") = R.astype(matmul244, dtype="float32") + gv5: R.Tuple(R.Tensor((1, seq_len, vocab_size), dtype="float32"), R.Object) = astype124, paged_kv_cache + R.output(gv5) + return gv5 + + @R.function + def create_tir_paged_kv_cache(max_batch_size_: R.Shape(["max_batch_size"]), max_total_seq_len_: R.Shape(["max_total_seq_len"]), prefill_chunk_size_: R.Shape(["prefill_chunk_size"]), page_size_: R.Shape(["page_size"]), support_sliding_window_: R.Shape(["support_sliding_window"])) -> R.Object: + max_batch_size = T.int64() + max_total_seq_len = T.int64() + prefill_chunk_size = T.int64() + page_size = T.int64() + support_sliding_window = T.int64() + R.func_attr({"relax.memory_plan_dynamic_func_output": True, "tir_non_negative_var": ["vocab_size"], "tir_var_upper_bound": {"batch_size": 128, "seq_len": 2048, "total_seq_len": 2048}}) + cls = Module + gv: R.Tensor((), dtype="float16") = R.zeros(R.shape([]), dtype="float16") + paged_kv_cache: R.Object = R.call_pure_packed("vm.builtin.paged_attention_kv_cache_create_reduced", R.shape([max_batch_size, max_total_seq_len, prefill_chunk_size, page_size, support_sliding_window]), R.shape([0, 12]), R.prim_value(12), R.prim_value(12), R.prim_value(64), R.prim_value(1), R.prim_value(1), R.prim_value(10000), gv, cls.tir_kv_cache_transpose_append, cls.batch_prefill_paged_kv, cls.batch_decode_paged_kv, cls.batch_prefill_paged_kv_sliding_window, cls.batch_decode_paged_kv_sliding_window, cls.batch_prefill_ragged_kv, cls.merge_state_inplace, cls.fused_rope, cls.copy_single_page, cls.tir_kv_cache_debug_get_kv, cls.compact_kv_copy, cls.batch_tree_attn, cls.tree_attn_paged_kv, R.prim_value(0), R.prim_value(0), sinfo_args=(R.Object,)) + return paged_kv_cache + + @R.function + def decode(input_embed: R.Tensor((1, 1, 768), dtype="float16"), paged_kv_cache: R.Object, packed_params: R.Tuple(R.Tensor(("vocab_size", 768), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((2304, 768), dtype="float16"), R.Tensor((2304,), dtype="float16"), R.Tensor((768, 768), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((3072, 768), dtype="float16"), R.Tensor((3072,), dtype="float32"), R.Tensor((768, 3072), dtype="float16"), R.Tensor((768,), dtype="float32"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((2304, 768), dtype="float16"), R.Tensor((2304,), dtype="float16"), R.Tensor((768, 768), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((3072, 768), dtype="float16"), R.Tensor((3072,), dtype="float32"), R.Tensor((768, 3072), dtype="float16"), R.Tensor((768,), dtype="float32"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((2304, 768), dtype="float16"), R.Tensor((2304,), dtype="float16"), R.Tensor((768, 768), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((3072, 768), dtype="float16"), R.Tensor((3072,), dtype="float32"), R.Tensor((768, 3072), dtype="float16"), R.Tensor((768,), dtype="float32"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((2304, 768), dtype="float16"), R.Tensor((2304,), dtype="float16"), R.Tensor((768, 768), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((3072, 768), dtype="float16"), R.Tensor((3072,), dtype="float32"), R.Tensor((768, 3072), dtype="float16"), R.Tensor((768,), dtype="float32"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((2304, 768), dtype="float16"), R.Tensor((2304,), dtype="float16"), R.Tensor((768, 768), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((3072, 768), dtype="float16"), R.Tensor((3072,), dtype="float32"), R.Tensor((768, 3072), dtype="float16"), R.Tensor((768,), dtype="float32"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((2304, 768), dtype="float16"), R.Tensor((2304,), dtype="float16"), R.Tensor((768, 768), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((3072, 768), dtype="float16"), R.Tensor((3072,), dtype="float32"), R.Tensor((768, 3072), dtype="float16"), R.Tensor((768,), dtype="float32"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((2304, 768), dtype="float16"), R.Tensor((2304,), dtype="float16"), R.Tensor((768, 768), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((3072, 768), dtype="float16"), R.Tensor((3072,), dtype="float32"), R.Tensor((768, 3072), dtype="float16"), R.Tensor((768,), dtype="float32"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((2304, 768), dtype="float16"), R.Tensor((2304,), dtype="float16"), R.Tensor((768, 768), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((3072, 768), dtype="float16"), R.Tensor((3072,), dtype="float32"), R.Tensor((768, 3072), dtype="float16"), R.Tensor((768,), dtype="float32"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((2304, 768), dtype="float16"), R.Tensor((2304,), dtype="float16"), R.Tensor((768, 768), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((3072, 768), dtype="float16"), R.Tensor((3072,), dtype="float32"), R.Tensor((768, 3072), dtype="float16"), R.Tensor((768,), dtype="float32"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((2304, 768), dtype="float16"), R.Tensor((2304,), dtype="float16"), R.Tensor((768, 768), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((3072, 768), dtype="float16"), R.Tensor((3072,), dtype="float32"), R.Tensor((768, 3072), dtype="float16"), R.Tensor((768,), dtype="float32"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((2304, 768), dtype="float16"), R.Tensor((2304,), dtype="float16"), R.Tensor((768, 768), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((3072, 768), dtype="float16"), R.Tensor((3072,), dtype="float32"), R.Tensor((768, 3072), dtype="float16"), R.Tensor((768,), dtype="float32"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((2304, 768), dtype="float16"), R.Tensor((2304,), dtype="float16"), R.Tensor((768, 768), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((3072, 768), dtype="float16"), R.Tensor((3072,), dtype="float32"), R.Tensor((768, 3072), dtype="float16"), R.Tensor((768,), dtype="float32"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor(("vocab_size", 768), dtype="float16"))) -> R.Tuple(R.Tensor((1, 1, "vocab_size"), dtype="float32"), R.Object): + vocab_size = T.int64() + R.func_attr({"num_input": 2, "pipeline_parallel_stages": 1, "relax.memory_plan_dynamic_func_output": True, "tir_non_negative_var": ["vocab_size"], "tir_var_upper_bound": {"batch_size": 128, "seq_len": 2048, "total_seq_len": 2048}}) + with R.dataflow(): + gpt_neox_embed_in_weight2: R.Tensor((vocab_size, 768), dtype="float16") = packed_params[0] + gpt_neox_layers_0_input_layernorm_weight2: R.Tensor((768,), dtype="float16") = packed_params[1] + gpt_neox_layers_0_input_layernorm_bias2: R.Tensor((768,), dtype="float16") = packed_params[2] + gpt_neox_layers_0_post_attention_layernorm_weight2: R.Tensor((768,), dtype="float16") = packed_params[3] + gpt_neox_layers_0_post_attention_layernorm_bias2: R.Tensor((768,), dtype="float16") = packed_params[4] + gpt_neox_layers_0_attention_query_key_value_weight2: R.Tensor((2304, 768), dtype="float16") = packed_params[5] + gpt_neox_layers_0_attention_query_key_value_bias2: R.Tensor((2304,), dtype="float16") = packed_params[6] + gpt_neox_layers_0_attention_dense_weight2: R.Tensor((768, 768), dtype="float16") = packed_params[7] + gpt_neox_layers_0_attention_dense_bias2: R.Tensor((768,), dtype="float16") = packed_params[8] + gpt_neox_layers_0_mlp_dense_h_to_4h_weight2: R.Tensor((3072, 768), dtype="float16") = packed_params[9] + gpt_neox_layers_0_mlp_dense_h_to_4h_bias2: R.Tensor((3072,), dtype="float32") = packed_params[10] + gpt_neox_layers_0_mlp_dense_4h_to_h_weight2: R.Tensor((768, 3072), dtype="float16") = packed_params[11] + gpt_neox_layers_0_mlp_dense_4h_to_h_bias2: R.Tensor((768,), dtype="float32") = packed_params[12] + gpt_neox_layers_1_input_layernorm_weight2: R.Tensor((768,), dtype="float16") = packed_params[13] + gpt_neox_layers_1_input_layernorm_bias2: R.Tensor((768,), dtype="float16") = packed_params[14] + gpt_neox_layers_1_post_attention_layernorm_weight2: R.Tensor((768,), dtype="float16") = packed_params[15] + gpt_neox_layers_1_post_attention_layernorm_bias2: R.Tensor((768,), dtype="float16") = packed_params[16] + gpt_neox_layers_1_attention_query_key_value_weight2: R.Tensor((2304, 768), dtype="float16") = packed_params[17] + gpt_neox_layers_1_attention_query_key_value_bias2: R.Tensor((2304,), dtype="float16") = packed_params[18] + gpt_neox_layers_1_attention_dense_weight2: R.Tensor((768, 768), dtype="float16") = packed_params[19] + gpt_neox_layers_1_attention_dense_bias2: R.Tensor((768,), dtype="float16") = packed_params[20] + gpt_neox_layers_1_mlp_dense_h_to_4h_weight2: R.Tensor((3072, 768), dtype="float16") = packed_params[21] + gpt_neox_layers_1_mlp_dense_h_to_4h_bias2: R.Tensor((3072,), dtype="float32") = packed_params[22] + gpt_neox_layers_1_mlp_dense_4h_to_h_weight2: R.Tensor((768, 3072), dtype="float16") = packed_params[23] + gpt_neox_layers_1_mlp_dense_4h_to_h_bias2: R.Tensor((768,), dtype="float32") = packed_params[24] + gpt_neox_layers_2_input_layernorm_weight2: R.Tensor((768,), dtype="float16") = packed_params[25] + gpt_neox_layers_2_input_layernorm_bias2: R.Tensor((768,), dtype="float16") = packed_params[26] + gpt_neox_layers_2_post_attention_layernorm_weight2: R.Tensor((768,), dtype="float16") = packed_params[27] + gpt_neox_layers_2_post_attention_layernorm_bias2: R.Tensor((768,), dtype="float16") = packed_params[28] + gpt_neox_layers_2_attention_query_key_value_weight2: R.Tensor((2304, 768), dtype="float16") = packed_params[29] + gpt_neox_layers_2_attention_query_key_value_bias2: R.Tensor((2304,), dtype="float16") = packed_params[30] + gpt_neox_layers_2_attention_dense_weight2: R.Tensor((768, 768), dtype="float16") = packed_params[31] + gpt_neox_layers_2_attention_dense_bias2: R.Tensor((768,), dtype="float16") = packed_params[32] + gpt_neox_layers_2_mlp_dense_h_to_4h_weight2: R.Tensor((3072, 768), dtype="float16") = packed_params[33] + gpt_neox_layers_2_mlp_dense_h_to_4h_bias2: R.Tensor((3072,), dtype="float32") = packed_params[34] + gpt_neox_layers_2_mlp_dense_4h_to_h_weight2: R.Tensor((768, 3072), dtype="float16") = packed_params[35] + gpt_neox_layers_2_mlp_dense_4h_to_h_bias2: R.Tensor((768,), dtype="float32") = packed_params[36] + gpt_neox_layers_3_input_layernorm_weight2: R.Tensor((768,), dtype="float16") = packed_params[37] + gpt_neox_layers_3_input_layernorm_bias2: R.Tensor((768,), dtype="float16") = packed_params[38] + gpt_neox_layers_3_post_attention_layernorm_weight2: R.Tensor((768,), dtype="float16") = packed_params[39] + gpt_neox_layers_3_post_attention_layernorm_bias2: R.Tensor((768,), dtype="float16") = packed_params[40] + gpt_neox_layers_3_attention_query_key_value_weight2: R.Tensor((2304, 768), dtype="float16") = packed_params[41] + gpt_neox_layers_3_attention_query_key_value_bias2: R.Tensor((2304,), dtype="float16") = packed_params[42] + gpt_neox_layers_3_attention_dense_weight2: R.Tensor((768, 768), dtype="float16") = packed_params[43] + gpt_neox_layers_3_attention_dense_bias2: R.Tensor((768,), dtype="float16") = packed_params[44] + gpt_neox_layers_3_mlp_dense_h_to_4h_weight2: R.Tensor((3072, 768), dtype="float16") = packed_params[45] + gpt_neox_layers_3_mlp_dense_h_to_4h_bias2: R.Tensor((3072,), dtype="float32") = packed_params[46] + gpt_neox_layers_3_mlp_dense_4h_to_h_weight2: R.Tensor((768, 3072), dtype="float16") = packed_params[47] + gpt_neox_layers_3_mlp_dense_4h_to_h_bias2: R.Tensor((768,), dtype="float32") = packed_params[48] + gpt_neox_layers_4_input_layernorm_weight2: R.Tensor((768,), dtype="float16") = packed_params[49] + gpt_neox_layers_4_input_layernorm_bias2: R.Tensor((768,), dtype="float16") = packed_params[50] + gpt_neox_layers_4_post_attention_layernorm_weight2: R.Tensor((768,), dtype="float16") = packed_params[51] + gpt_neox_layers_4_post_attention_layernorm_bias2: R.Tensor((768,), dtype="float16") = packed_params[52] + gpt_neox_layers_4_attention_query_key_value_weight2: R.Tensor((2304, 768), dtype="float16") = packed_params[53] + gpt_neox_layers_4_attention_query_key_value_bias2: R.Tensor((2304,), dtype="float16") = packed_params[54] + gpt_neox_layers_4_attention_dense_weight2: R.Tensor((768, 768), dtype="float16") = packed_params[55] + gpt_neox_layers_4_attention_dense_bias2: R.Tensor((768,), dtype="float16") = packed_params[56] + gpt_neox_layers_4_mlp_dense_h_to_4h_weight2: R.Tensor((3072, 768), dtype="float16") = packed_params[57] + gpt_neox_layers_4_mlp_dense_h_to_4h_bias2: R.Tensor((3072,), dtype="float32") = packed_params[58] + gpt_neox_layers_4_mlp_dense_4h_to_h_weight2: R.Tensor((768, 3072), dtype="float16") = packed_params[59] + gpt_neox_layers_4_mlp_dense_4h_to_h_bias2: R.Tensor((768,), dtype="float32") = packed_params[60] + gpt_neox_layers_5_input_layernorm_weight2: R.Tensor((768,), dtype="float16") = packed_params[61] + gpt_neox_layers_5_input_layernorm_bias2: R.Tensor((768,), dtype="float16") = packed_params[62] + gpt_neox_layers_5_post_attention_layernorm_weight2: R.Tensor((768,), dtype="float16") = packed_params[63] + gpt_neox_layers_5_post_attention_layernorm_bias2: R.Tensor((768,), dtype="float16") = packed_params[64] + gpt_neox_layers_5_attention_query_key_value_weight2: R.Tensor((2304, 768), dtype="float16") = packed_params[65] + gpt_neox_layers_5_attention_query_key_value_bias2: R.Tensor((2304,), dtype="float16") = packed_params[66] + gpt_neox_layers_5_attention_dense_weight2: R.Tensor((768, 768), dtype="float16") = packed_params[67] + gpt_neox_layers_5_attention_dense_bias2: R.Tensor((768,), dtype="float16") = packed_params[68] + gpt_neox_layers_5_mlp_dense_h_to_4h_weight2: R.Tensor((3072, 768), dtype="float16") = packed_params[69] + gpt_neox_layers_5_mlp_dense_h_to_4h_bias2: R.Tensor((3072,), dtype="float32") = packed_params[70] + gpt_neox_layers_5_mlp_dense_4h_to_h_weight2: R.Tensor((768, 3072), dtype="float16") = packed_params[71] + gpt_neox_layers_5_mlp_dense_4h_to_h_bias2: R.Tensor((768,), dtype="float32") = packed_params[72] + gpt_neox_layers_6_input_layernorm_weight2: R.Tensor((768,), dtype="float16") = packed_params[73] + gpt_neox_layers_6_input_layernorm_bias2: R.Tensor((768,), dtype="float16") = packed_params[74] + gpt_neox_layers_6_post_attention_layernorm_weight2: R.Tensor((768,), dtype="float16") = packed_params[75] + gpt_neox_layers_6_post_attention_layernorm_bias2: R.Tensor((768,), dtype="float16") = packed_params[76] + gpt_neox_layers_6_attention_query_key_value_weight2: R.Tensor((2304, 768), dtype="float16") = packed_params[77] + gpt_neox_layers_6_attention_query_key_value_bias2: R.Tensor((2304,), dtype="float16") = packed_params[78] + gpt_neox_layers_6_attention_dense_weight2: R.Tensor((768, 768), dtype="float16") = packed_params[79] + gpt_neox_layers_6_attention_dense_bias2: R.Tensor((768,), dtype="float16") = packed_params[80] + gpt_neox_layers_6_mlp_dense_h_to_4h_weight2: R.Tensor((3072, 768), dtype="float16") = packed_params[81] + gpt_neox_layers_6_mlp_dense_h_to_4h_bias2: R.Tensor((3072,), dtype="float32") = packed_params[82] + gpt_neox_layers_6_mlp_dense_4h_to_h_weight2: R.Tensor((768, 3072), dtype="float16") = packed_params[83] + gpt_neox_layers_6_mlp_dense_4h_to_h_bias2: R.Tensor((768,), dtype="float32") = packed_params[84] + gpt_neox_layers_7_input_layernorm_weight2: R.Tensor((768,), dtype="float16") = packed_params[85] + gpt_neox_layers_7_input_layernorm_bias2: R.Tensor((768,), dtype="float16") = packed_params[86] + gpt_neox_layers_7_post_attention_layernorm_weight2: R.Tensor((768,), dtype="float16") = packed_params[87] + gpt_neox_layers_7_post_attention_layernorm_bias2: R.Tensor((768,), dtype="float16") = packed_params[88] + gpt_neox_layers_7_attention_query_key_value_weight2: R.Tensor((2304, 768), dtype="float16") = packed_params[89] + gpt_neox_layers_7_attention_query_key_value_bias2: R.Tensor((2304,), dtype="float16") = packed_params[90] + gpt_neox_layers_7_attention_dense_weight2: R.Tensor((768, 768), dtype="float16") = packed_params[91] + gpt_neox_layers_7_attention_dense_bias2: R.Tensor((768,), dtype="float16") = packed_params[92] + gpt_neox_layers_7_mlp_dense_h_to_4h_weight2: R.Tensor((3072, 768), dtype="float16") = packed_params[93] + gpt_neox_layers_7_mlp_dense_h_to_4h_bias2: R.Tensor((3072,), dtype="float32") = packed_params[94] + gpt_neox_layers_7_mlp_dense_4h_to_h_weight2: R.Tensor((768, 3072), dtype="float16") = packed_params[95] + gpt_neox_layers_7_mlp_dense_4h_to_h_bias2: R.Tensor((768,), dtype="float32") = packed_params[96] + gpt_neox_layers_8_input_layernorm_weight2: R.Tensor((768,), dtype="float16") = packed_params[97] + gpt_neox_layers_8_input_layernorm_bias2: R.Tensor((768,), dtype="float16") = packed_params[98] + gpt_neox_layers_8_post_attention_layernorm_weight2: R.Tensor((768,), dtype="float16") = packed_params[99] + gpt_neox_layers_8_post_attention_layernorm_bias2: R.Tensor((768,), dtype="float16") = packed_params[100] + gpt_neox_layers_8_attention_query_key_value_weight2: R.Tensor((2304, 768), dtype="float16") = packed_params[101] + gpt_neox_layers_8_attention_query_key_value_bias2: R.Tensor((2304,), dtype="float16") = packed_params[102] + gpt_neox_layers_8_attention_dense_weight2: R.Tensor((768, 768), dtype="float16") = packed_params[103] + gpt_neox_layers_8_attention_dense_bias2: R.Tensor((768,), dtype="float16") = packed_params[104] + gpt_neox_layers_8_mlp_dense_h_to_4h_weight2: R.Tensor((3072, 768), dtype="float16") = packed_params[105] + gpt_neox_layers_8_mlp_dense_h_to_4h_bias2: R.Tensor((3072,), dtype="float32") = packed_params[106] + gpt_neox_layers_8_mlp_dense_4h_to_h_weight2: R.Tensor((768, 3072), dtype="float16") = packed_params[107] + gpt_neox_layers_8_mlp_dense_4h_to_h_bias2: R.Tensor((768,), dtype="float32") = packed_params[108] + gpt_neox_layers_9_input_layernorm_weight2: R.Tensor((768,), dtype="float16") = packed_params[109] + gpt_neox_layers_9_input_layernorm_bias2: R.Tensor((768,), dtype="float16") = packed_params[110] + gpt_neox_layers_9_post_attention_layernorm_weight2: R.Tensor((768,), dtype="float16") = packed_params[111] + gpt_neox_layers_9_post_attention_layernorm_bias2: R.Tensor((768,), dtype="float16") = packed_params[112] + gpt_neox_layers_9_attention_query_key_value_weight2: R.Tensor((2304, 768), dtype="float16") = packed_params[113] + gpt_neox_layers_9_attention_query_key_value_bias2: R.Tensor((2304,), dtype="float16") = packed_params[114] + gpt_neox_layers_9_attention_dense_weight2: R.Tensor((768, 768), dtype="float16") = packed_params[115] + gpt_neox_layers_9_attention_dense_bias2: R.Tensor((768,), dtype="float16") = packed_params[116] + gpt_neox_layers_9_mlp_dense_h_to_4h_weight2: R.Tensor((3072, 768), dtype="float16") = packed_params[117] + gpt_neox_layers_9_mlp_dense_h_to_4h_bias2: R.Tensor((3072,), dtype="float32") = packed_params[118] + gpt_neox_layers_9_mlp_dense_4h_to_h_weight2: R.Tensor((768, 3072), dtype="float16") = packed_params[119] + gpt_neox_layers_9_mlp_dense_4h_to_h_bias2: R.Tensor((768,), dtype="float32") = packed_params[120] + gpt_neox_layers_10_input_layernorm_weight2: R.Tensor((768,), dtype="float16") = packed_params[121] + gpt_neox_layers_10_input_layernorm_bias2: R.Tensor((768,), dtype="float16") = packed_params[122] + gpt_neox_layers_10_post_attention_layernorm_weight2: R.Tensor((768,), dtype="float16") = packed_params[123] + gpt_neox_layers_10_post_attention_layernorm_bias2: R.Tensor((768,), dtype="float16") = packed_params[124] + gpt_neox_layers_10_attention_query_key_value_weight2: R.Tensor((2304, 768), dtype="float16") = packed_params[125] + gpt_neox_layers_10_attention_query_key_value_bias2: R.Tensor((2304,), dtype="float16") = packed_params[126] + gpt_neox_layers_10_attention_dense_weight2: R.Tensor((768, 768), dtype="float16") = packed_params[127] + gpt_neox_layers_10_attention_dense_bias2: R.Tensor((768,), dtype="float16") = packed_params[128] + gpt_neox_layers_10_mlp_dense_h_to_4h_weight2: R.Tensor((3072, 768), dtype="float16") = packed_params[129] + gpt_neox_layers_10_mlp_dense_h_to_4h_bias2: R.Tensor((3072,), dtype="float32") = packed_params[130] + gpt_neox_layers_10_mlp_dense_4h_to_h_weight2: R.Tensor((768, 3072), dtype="float16") = packed_params[131] + gpt_neox_layers_10_mlp_dense_4h_to_h_bias2: R.Tensor((768,), dtype="float32") = packed_params[132] + gpt_neox_layers_11_input_layernorm_weight2: R.Tensor((768,), dtype="float16") = packed_params[133] + gpt_neox_layers_11_input_layernorm_bias2: R.Tensor((768,), dtype="float16") = packed_params[134] + gpt_neox_layers_11_post_attention_layernorm_weight2: R.Tensor((768,), dtype="float16") = packed_params[135] + gpt_neox_layers_11_post_attention_layernorm_bias2: R.Tensor((768,), dtype="float16") = packed_params[136] + gpt_neox_layers_11_attention_query_key_value_weight2: R.Tensor((2304, 768), dtype="float16") = packed_params[137] + gpt_neox_layers_11_attention_query_key_value_bias2: R.Tensor((2304,), dtype="float16") = packed_params[138] + gpt_neox_layers_11_attention_dense_weight2: R.Tensor((768, 768), dtype="float16") = packed_params[139] + gpt_neox_layers_11_attention_dense_bias2: R.Tensor((768,), dtype="float16") = packed_params[140] + gpt_neox_layers_11_mlp_dense_h_to_4h_weight2: R.Tensor((3072, 768), dtype="float16") = packed_params[141] + gpt_neox_layers_11_mlp_dense_h_to_4h_bias2: R.Tensor((3072,), dtype="float32") = packed_params[142] + gpt_neox_layers_11_mlp_dense_4h_to_h_weight2: R.Tensor((768, 3072), dtype="float16") = packed_params[143] + gpt_neox_layers_11_mlp_dense_4h_to_h_bias2: R.Tensor((768,), dtype="float32") = packed_params[144] + gpt_neox_final_layer_norm_weight2: R.Tensor((768,), dtype="float16") = packed_params[145] + gpt_neox_final_layer_norm_bias2: R.Tensor((768,), dtype="float16") = packed_params[146] + embed_out_weight2: R.Tensor((vocab_size, 768), dtype="float16") = packed_params[147] + layer_norm25: R.Tensor((1, 1, 768), dtype="float16") = R.nn.layer_norm(input_embed, gpt_neox_layers_0_input_layernorm_weight2, gpt_neox_layers_0_input_layernorm_bias2, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims49: R.Tensor((768, 2304), dtype="float16") = R.permute_dims(gpt_neox_layers_0_attention_query_key_value_weight2, axes=None) + matmul49: R.Tensor((1, 1, 2304), dtype="float16") = R.matmul(layer_norm25, permute_dims49, out_dtype="void") + add72: R.Tensor((1, 1, 2304), dtype="float16") = R.add(matmul49, gpt_neox_layers_0_attention_query_key_value_bias2) + reshape48: R.Tensor((1, 1, 36, 64), dtype="float16") = R.reshape(add72, R.shape([1, 1, 36, 64])) + reshape49: R.Tensor((1, 36, 64), dtype="float16") = R.reshape(reshape48, R.shape([1, 36, 64])) + lv13 = R.call_dps_packed("vm.builtin.attention_kv_cache_attention_with_fused_qkv", (paged_kv_cache, R.prim_value(0), R.prim_value(T.float32(1.0)), reshape49), out_sinfo=R.Tensor((1, 12, 64), dtype="float16")) + reshape50: R.Tensor((1, 1, 12, 64), dtype="float16") = R.reshape(lv13, R.shape([1, 1, 12, 64])) + reshape51: R.Tensor((1, 1, 768), dtype="float16") = R.reshape(reshape50, R.shape([1, 1, 768])) + permute_dims50: R.Tensor((768, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_0_attention_dense_weight2, axes=None) + matmul50: R.Tensor((1, 1, 768), dtype="float16") = R.matmul(reshape51, permute_dims50, out_dtype="void") + add73: R.Tensor((1, 1, 768), dtype="float16") = R.add(matmul50, gpt_neox_layers_0_attention_dense_bias2) + layer_norm26: R.Tensor((1, 1, 768), dtype="float16") = R.nn.layer_norm(input_embed, gpt_neox_layers_0_post_attention_layernorm_weight2, gpt_neox_layers_0_post_attention_layernorm_bias2, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims51: R.Tensor((768, 3072), dtype="float16") = R.permute_dims(gpt_neox_layers_0_mlp_dense_h_to_4h_weight2, axes=None) + matmul51: R.Tensor((1, 1, 3072), dtype="float32") = R.matmul(layer_norm26, permute_dims51, out_dtype="float32") + add74: R.Tensor((1, 1, 3072), dtype="float32") = R.add(matmul51, gpt_neox_layers_0_mlp_dense_h_to_4h_bias2) + gelu12: R.Tensor((1, 1, 3072), dtype="float32") = R.nn.gelu(add74) + astype25: R.Tensor((1, 1, 3072), dtype="float16") = R.astype(gelu12, dtype="float16") + permute_dims52: R.Tensor((3072, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_0_mlp_dense_4h_to_h_weight2, axes=None) + matmul52: R.Tensor((1, 1, 768), dtype="float32") = R.matmul(astype25, permute_dims52, out_dtype="float32") + add75: R.Tensor((1, 1, 768), dtype="float32") = R.add(matmul52, gpt_neox_layers_0_mlp_dense_4h_to_h_bias2) + astype26: R.Tensor((1, 1, 768), dtype="float16") = R.astype(add75, dtype="float16") + add76: R.Tensor((1, 1, 768), dtype="float16") = R.add(astype26, add73) + add77: R.Tensor((1, 1, 768), dtype="float16") = R.add(add76, input_embed) + layer_norm27: R.Tensor((1, 1, 768), dtype="float16") = R.nn.layer_norm(add77, gpt_neox_layers_1_input_layernorm_weight2, gpt_neox_layers_1_input_layernorm_bias2, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims53: R.Tensor((768, 2304), dtype="float16") = R.permute_dims(gpt_neox_layers_1_attention_query_key_value_weight2, axes=None) + matmul53: R.Tensor((1, 1, 2304), dtype="float16") = R.matmul(layer_norm27, permute_dims53, out_dtype="void") + add78: R.Tensor((1, 1, 2304), dtype="float16") = R.add(matmul53, gpt_neox_layers_1_attention_query_key_value_bias2) + reshape52: R.Tensor((1, 1, 36, 64), dtype="float16") = R.reshape(add78, R.shape([1, 1, 36, 64])) + reshape53: R.Tensor((1, 36, 64), dtype="float16") = R.reshape(reshape52, R.shape([1, 36, 64])) + lv14 = R.call_dps_packed("vm.builtin.attention_kv_cache_attention_with_fused_qkv", (paged_kv_cache, R.prim_value(1), R.prim_value(T.float32(1.0)), reshape53), out_sinfo=R.Tensor((1, 12, 64), dtype="float16")) + reshape54: R.Tensor((1, 1, 12, 64), dtype="float16") = R.reshape(lv14, R.shape([1, 1, 12, 64])) + reshape55: R.Tensor((1, 1, 768), dtype="float16") = R.reshape(reshape54, R.shape([1, 1, 768])) + permute_dims54: R.Tensor((768, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_1_attention_dense_weight2, axes=None) + matmul54: R.Tensor((1, 1, 768), dtype="float16") = R.matmul(reshape55, permute_dims54, out_dtype="void") + add79: R.Tensor((1, 1, 768), dtype="float16") = R.add(matmul54, gpt_neox_layers_1_attention_dense_bias2) + layer_norm28: R.Tensor((1, 1, 768), dtype="float16") = R.nn.layer_norm(add77, gpt_neox_layers_1_post_attention_layernorm_weight2, gpt_neox_layers_1_post_attention_layernorm_bias2, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims55: R.Tensor((768, 3072), dtype="float16") = R.permute_dims(gpt_neox_layers_1_mlp_dense_h_to_4h_weight2, axes=None) + matmul55: R.Tensor((1, 1, 3072), dtype="float32") = R.matmul(layer_norm28, permute_dims55, out_dtype="float32") + add80: R.Tensor((1, 1, 3072), dtype="float32") = R.add(matmul55, gpt_neox_layers_1_mlp_dense_h_to_4h_bias2) + gelu13: R.Tensor((1, 1, 3072), dtype="float32") = R.nn.gelu(add80) + astype27: R.Tensor((1, 1, 3072), dtype="float16") = R.astype(gelu13, dtype="float16") + permute_dims56: R.Tensor((3072, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_1_mlp_dense_4h_to_h_weight2, axes=None) + matmul56: R.Tensor((1, 1, 768), dtype="float32") = R.matmul(astype27, permute_dims56, out_dtype="float32") + add81: R.Tensor((1, 1, 768), dtype="float32") = R.add(matmul56, gpt_neox_layers_1_mlp_dense_4h_to_h_bias2) + astype28: R.Tensor((1, 1, 768), dtype="float16") = R.astype(add81, dtype="float16") + add82: R.Tensor((1, 1, 768), dtype="float16") = R.add(astype28, add79) + add83: R.Tensor((1, 1, 768), dtype="float16") = R.add(add82, add77) + layer_norm29: R.Tensor((1, 1, 768), dtype="float16") = R.nn.layer_norm(add83, gpt_neox_layers_2_input_layernorm_weight2, gpt_neox_layers_2_input_layernorm_bias2, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims57: R.Tensor((768, 2304), dtype="float16") = R.permute_dims(gpt_neox_layers_2_attention_query_key_value_weight2, axes=None) + matmul57: R.Tensor((1, 1, 2304), dtype="float16") = R.matmul(layer_norm29, permute_dims57, out_dtype="void") + add84: R.Tensor((1, 1, 2304), dtype="float16") = R.add(matmul57, gpt_neox_layers_2_attention_query_key_value_bias2) + reshape56: R.Tensor((1, 1, 36, 64), dtype="float16") = R.reshape(add84, R.shape([1, 1, 36, 64])) + reshape57: R.Tensor((1, 36, 64), dtype="float16") = R.reshape(reshape56, R.shape([1, 36, 64])) + lv15 = R.call_dps_packed("vm.builtin.attention_kv_cache_attention_with_fused_qkv", (paged_kv_cache, R.prim_value(2), R.prim_value(T.float32(1.0)), reshape57), out_sinfo=R.Tensor((1, 12, 64), dtype="float16")) + reshape58: R.Tensor((1, 1, 12, 64), dtype="float16") = R.reshape(lv15, R.shape([1, 1, 12, 64])) + reshape59: R.Tensor((1, 1, 768), dtype="float16") = R.reshape(reshape58, R.shape([1, 1, 768])) + permute_dims58: R.Tensor((768, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_2_attention_dense_weight2, axes=None) + matmul58: R.Tensor((1, 1, 768), dtype="float16") = R.matmul(reshape59, permute_dims58, out_dtype="void") + add85: R.Tensor((1, 1, 768), dtype="float16") = R.add(matmul58, gpt_neox_layers_2_attention_dense_bias2) + layer_norm30: R.Tensor((1, 1, 768), dtype="float16") = R.nn.layer_norm(add83, gpt_neox_layers_2_post_attention_layernorm_weight2, gpt_neox_layers_2_post_attention_layernorm_bias2, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims59: R.Tensor((768, 3072), dtype="float16") = R.permute_dims(gpt_neox_layers_2_mlp_dense_h_to_4h_weight2, axes=None) + matmul59: R.Tensor((1, 1, 3072), dtype="float32") = R.matmul(layer_norm30, permute_dims59, out_dtype="float32") + add86: R.Tensor((1, 1, 3072), dtype="float32") = R.add(matmul59, gpt_neox_layers_2_mlp_dense_h_to_4h_bias2) + gelu14: R.Tensor((1, 1, 3072), dtype="float32") = R.nn.gelu(add86) + astype29: R.Tensor((1, 1, 3072), dtype="float16") = R.astype(gelu14, dtype="float16") + permute_dims60: R.Tensor((3072, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_2_mlp_dense_4h_to_h_weight2, axes=None) + matmul60: R.Tensor((1, 1, 768), dtype="float32") = R.matmul(astype29, permute_dims60, out_dtype="float32") + add87: R.Tensor((1, 1, 768), dtype="float32") = R.add(matmul60, gpt_neox_layers_2_mlp_dense_4h_to_h_bias2) + astype30: R.Tensor((1, 1, 768), dtype="float16") = R.astype(add87, dtype="float16") + add88: R.Tensor((1, 1, 768), dtype="float16") = R.add(astype30, add85) + add89: R.Tensor((1, 1, 768), dtype="float16") = R.add(add88, add83) + layer_norm31: R.Tensor((1, 1, 768), dtype="float16") = R.nn.layer_norm(add89, gpt_neox_layers_3_input_layernorm_weight2, gpt_neox_layers_3_input_layernorm_bias2, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims61: R.Tensor((768, 2304), dtype="float16") = R.permute_dims(gpt_neox_layers_3_attention_query_key_value_weight2, axes=None) + matmul61: R.Tensor((1, 1, 2304), dtype="float16") = R.matmul(layer_norm31, permute_dims61, out_dtype="void") + add90: R.Tensor((1, 1, 2304), dtype="float16") = R.add(matmul61, gpt_neox_layers_3_attention_query_key_value_bias2) + reshape60: R.Tensor((1, 1, 36, 64), dtype="float16") = R.reshape(add90, R.shape([1, 1, 36, 64])) + reshape61: R.Tensor((1, 36, 64), dtype="float16") = R.reshape(reshape60, R.shape([1, 36, 64])) + lv16 = R.call_dps_packed("vm.builtin.attention_kv_cache_attention_with_fused_qkv", (paged_kv_cache, R.prim_value(3), R.prim_value(T.float32(1.0)), reshape61), out_sinfo=R.Tensor((1, 12, 64), dtype="float16")) + reshape62: R.Tensor((1, 1, 12, 64), dtype="float16") = R.reshape(lv16, R.shape([1, 1, 12, 64])) + reshape63: R.Tensor((1, 1, 768), dtype="float16") = R.reshape(reshape62, R.shape([1, 1, 768])) + permute_dims62: R.Tensor((768, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_3_attention_dense_weight2, axes=None) + matmul62: R.Tensor((1, 1, 768), dtype="float16") = R.matmul(reshape63, permute_dims62, out_dtype="void") + add91: R.Tensor((1, 1, 768), dtype="float16") = R.add(matmul62, gpt_neox_layers_3_attention_dense_bias2) + layer_norm32: R.Tensor((1, 1, 768), dtype="float16") = R.nn.layer_norm(add89, gpt_neox_layers_3_post_attention_layernorm_weight2, gpt_neox_layers_3_post_attention_layernorm_bias2, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims63: R.Tensor((768, 3072), dtype="float16") = R.permute_dims(gpt_neox_layers_3_mlp_dense_h_to_4h_weight2, axes=None) + matmul63: R.Tensor((1, 1, 3072), dtype="float32") = R.matmul(layer_norm32, permute_dims63, out_dtype="float32") + add92: R.Tensor((1, 1, 3072), dtype="float32") = R.add(matmul63, gpt_neox_layers_3_mlp_dense_h_to_4h_bias2) + gelu15: R.Tensor((1, 1, 3072), dtype="float32") = R.nn.gelu(add92) + astype31: R.Tensor((1, 1, 3072), dtype="float16") = R.astype(gelu15, dtype="float16") + permute_dims64: R.Tensor((3072, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_3_mlp_dense_4h_to_h_weight2, axes=None) + matmul64: R.Tensor((1, 1, 768), dtype="float32") = R.matmul(astype31, permute_dims64, out_dtype="float32") + add93: R.Tensor((1, 1, 768), dtype="float32") = R.add(matmul64, gpt_neox_layers_3_mlp_dense_4h_to_h_bias2) + astype32: R.Tensor((1, 1, 768), dtype="float16") = R.astype(add93, dtype="float16") + add94: R.Tensor((1, 1, 768), dtype="float16") = R.add(astype32, add91) + add95: R.Tensor((1, 1, 768), dtype="float16") = R.add(add94, add89) + layer_norm33: R.Tensor((1, 1, 768), dtype="float16") = R.nn.layer_norm(add95, gpt_neox_layers_4_input_layernorm_weight2, gpt_neox_layers_4_input_layernorm_bias2, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims65: R.Tensor((768, 2304), dtype="float16") = R.permute_dims(gpt_neox_layers_4_attention_query_key_value_weight2, axes=None) + matmul65: R.Tensor((1, 1, 2304), dtype="float16") = R.matmul(layer_norm33, permute_dims65, out_dtype="void") + add96: R.Tensor((1, 1, 2304), dtype="float16") = R.add(matmul65, gpt_neox_layers_4_attention_query_key_value_bias2) + reshape64: R.Tensor((1, 1, 36, 64), dtype="float16") = R.reshape(add96, R.shape([1, 1, 36, 64])) + reshape65: R.Tensor((1, 36, 64), dtype="float16") = R.reshape(reshape64, R.shape([1, 36, 64])) + lv17 = R.call_dps_packed("vm.builtin.attention_kv_cache_attention_with_fused_qkv", (paged_kv_cache, R.prim_value(4), R.prim_value(T.float32(1.0)), reshape65), out_sinfo=R.Tensor((1, 12, 64), dtype="float16")) + reshape66: R.Tensor((1, 1, 12, 64), dtype="float16") = R.reshape(lv17, R.shape([1, 1, 12, 64])) + reshape67: R.Tensor((1, 1, 768), dtype="float16") = R.reshape(reshape66, R.shape([1, 1, 768])) + permute_dims66: R.Tensor((768, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_4_attention_dense_weight2, axes=None) + matmul66: R.Tensor((1, 1, 768), dtype="float16") = R.matmul(reshape67, permute_dims66, out_dtype="void") + add97: R.Tensor((1, 1, 768), dtype="float16") = R.add(matmul66, gpt_neox_layers_4_attention_dense_bias2) + layer_norm34: R.Tensor((1, 1, 768), dtype="float16") = R.nn.layer_norm(add95, gpt_neox_layers_4_post_attention_layernorm_weight2, gpt_neox_layers_4_post_attention_layernorm_bias2, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims67: R.Tensor((768, 3072), dtype="float16") = R.permute_dims(gpt_neox_layers_4_mlp_dense_h_to_4h_weight2, axes=None) + matmul67: R.Tensor((1, 1, 3072), dtype="float32") = R.matmul(layer_norm34, permute_dims67, out_dtype="float32") + add98: R.Tensor((1, 1, 3072), dtype="float32") = R.add(matmul67, gpt_neox_layers_4_mlp_dense_h_to_4h_bias2) + gelu16: R.Tensor((1, 1, 3072), dtype="float32") = R.nn.gelu(add98) + astype33: R.Tensor((1, 1, 3072), dtype="float16") = R.astype(gelu16, dtype="float16") + permute_dims68: R.Tensor((3072, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_4_mlp_dense_4h_to_h_weight2, axes=None) + matmul68: R.Tensor((1, 1, 768), dtype="float32") = R.matmul(astype33, permute_dims68, out_dtype="float32") + add99: R.Tensor((1, 1, 768), dtype="float32") = R.add(matmul68, gpt_neox_layers_4_mlp_dense_4h_to_h_bias2) + astype34: R.Tensor((1, 1, 768), dtype="float16") = R.astype(add99, dtype="float16") + add100: R.Tensor((1, 1, 768), dtype="float16") = R.add(astype34, add97) + add101: R.Tensor((1, 1, 768), dtype="float16") = R.add(add100, add95) + layer_norm35: R.Tensor((1, 1, 768), dtype="float16") = R.nn.layer_norm(add101, gpt_neox_layers_5_input_layernorm_weight2, gpt_neox_layers_5_input_layernorm_bias2, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims69: R.Tensor((768, 2304), dtype="float16") = R.permute_dims(gpt_neox_layers_5_attention_query_key_value_weight2, axes=None) + matmul69: R.Tensor((1, 1, 2304), dtype="float16") = R.matmul(layer_norm35, permute_dims69, out_dtype="void") + add102: R.Tensor((1, 1, 2304), dtype="float16") = R.add(matmul69, gpt_neox_layers_5_attention_query_key_value_bias2) + reshape68: R.Tensor((1, 1, 36, 64), dtype="float16") = R.reshape(add102, R.shape([1, 1, 36, 64])) + reshape69: R.Tensor((1, 36, 64), dtype="float16") = R.reshape(reshape68, R.shape([1, 36, 64])) + lv18 = R.call_dps_packed("vm.builtin.attention_kv_cache_attention_with_fused_qkv", (paged_kv_cache, R.prim_value(5), R.prim_value(T.float32(1.0)), reshape69), out_sinfo=R.Tensor((1, 12, 64), dtype="float16")) + reshape70: R.Tensor((1, 1, 12, 64), dtype="float16") = R.reshape(lv18, R.shape([1, 1, 12, 64])) + reshape71: R.Tensor((1, 1, 768), dtype="float16") = R.reshape(reshape70, R.shape([1, 1, 768])) + permute_dims70: R.Tensor((768, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_5_attention_dense_weight2, axes=None) + matmul70: R.Tensor((1, 1, 768), dtype="float16") = R.matmul(reshape71, permute_dims70, out_dtype="void") + add103: R.Tensor((1, 1, 768), dtype="float16") = R.add(matmul70, gpt_neox_layers_5_attention_dense_bias2) + layer_norm36: R.Tensor((1, 1, 768), dtype="float16") = R.nn.layer_norm(add101, gpt_neox_layers_5_post_attention_layernorm_weight2, gpt_neox_layers_5_post_attention_layernorm_bias2, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims71: R.Tensor((768, 3072), dtype="float16") = R.permute_dims(gpt_neox_layers_5_mlp_dense_h_to_4h_weight2, axes=None) + matmul71: R.Tensor((1, 1, 3072), dtype="float32") = R.matmul(layer_norm36, permute_dims71, out_dtype="float32") + add104: R.Tensor((1, 1, 3072), dtype="float32") = R.add(matmul71, gpt_neox_layers_5_mlp_dense_h_to_4h_bias2) + gelu17: R.Tensor((1, 1, 3072), dtype="float32") = R.nn.gelu(add104) + astype35: R.Tensor((1, 1, 3072), dtype="float16") = R.astype(gelu17, dtype="float16") + permute_dims72: R.Tensor((3072, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_5_mlp_dense_4h_to_h_weight2, axes=None) + matmul72: R.Tensor((1, 1, 768), dtype="float32") = R.matmul(astype35, permute_dims72, out_dtype="float32") + add105: R.Tensor((1, 1, 768), dtype="float32") = R.add(matmul72, gpt_neox_layers_5_mlp_dense_4h_to_h_bias2) + astype36: R.Tensor((1, 1, 768), dtype="float16") = R.astype(add105, dtype="float16") + add106: R.Tensor((1, 1, 768), dtype="float16") = R.add(astype36, add103) + add107: R.Tensor((1, 1, 768), dtype="float16") = R.add(add106, add101) + layer_norm37: R.Tensor((1, 1, 768), dtype="float16") = R.nn.layer_norm(add107, gpt_neox_layers_6_input_layernorm_weight2, gpt_neox_layers_6_input_layernorm_bias2, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims73: R.Tensor((768, 2304), dtype="float16") = R.permute_dims(gpt_neox_layers_6_attention_query_key_value_weight2, axes=None) + matmul73: R.Tensor((1, 1, 2304), dtype="float16") = R.matmul(layer_norm37, permute_dims73, out_dtype="void") + add108: R.Tensor((1, 1, 2304), dtype="float16") = R.add(matmul73, gpt_neox_layers_6_attention_query_key_value_bias2) + reshape72: R.Tensor((1, 1, 36, 64), dtype="float16") = R.reshape(add108, R.shape([1, 1, 36, 64])) + reshape73: R.Tensor((1, 36, 64), dtype="float16") = R.reshape(reshape72, R.shape([1, 36, 64])) + lv19 = R.call_dps_packed("vm.builtin.attention_kv_cache_attention_with_fused_qkv", (paged_kv_cache, R.prim_value(6), R.prim_value(T.float32(1.0)), reshape73), out_sinfo=R.Tensor((1, 12, 64), dtype="float16")) + reshape74: R.Tensor((1, 1, 12, 64), dtype="float16") = R.reshape(lv19, R.shape([1, 1, 12, 64])) + reshape75: R.Tensor((1, 1, 768), dtype="float16") = R.reshape(reshape74, R.shape([1, 1, 768])) + permute_dims74: R.Tensor((768, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_6_attention_dense_weight2, axes=None) + matmul74: R.Tensor((1, 1, 768), dtype="float16") = R.matmul(reshape75, permute_dims74, out_dtype="void") + add109: R.Tensor((1, 1, 768), dtype="float16") = R.add(matmul74, gpt_neox_layers_6_attention_dense_bias2) + layer_norm38: R.Tensor((1, 1, 768), dtype="float16") = R.nn.layer_norm(add107, gpt_neox_layers_6_post_attention_layernorm_weight2, gpt_neox_layers_6_post_attention_layernorm_bias2, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims75: R.Tensor((768, 3072), dtype="float16") = R.permute_dims(gpt_neox_layers_6_mlp_dense_h_to_4h_weight2, axes=None) + matmul75: R.Tensor((1, 1, 3072), dtype="float32") = R.matmul(layer_norm38, permute_dims75, out_dtype="float32") + add110: R.Tensor((1, 1, 3072), dtype="float32") = R.add(matmul75, gpt_neox_layers_6_mlp_dense_h_to_4h_bias2) + gelu18: R.Tensor((1, 1, 3072), dtype="float32") = R.nn.gelu(add110) + astype37: R.Tensor((1, 1, 3072), dtype="float16") = R.astype(gelu18, dtype="float16") + permute_dims76: R.Tensor((3072, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_6_mlp_dense_4h_to_h_weight2, axes=None) + matmul76: R.Tensor((1, 1, 768), dtype="float32") = R.matmul(astype37, permute_dims76, out_dtype="float32") + add111: R.Tensor((1, 1, 768), dtype="float32") = R.add(matmul76, gpt_neox_layers_6_mlp_dense_4h_to_h_bias2) + astype38: R.Tensor((1, 1, 768), dtype="float16") = R.astype(add111, dtype="float16") + add112: R.Tensor((1, 1, 768), dtype="float16") = R.add(astype38, add109) + add113: R.Tensor((1, 1, 768), dtype="float16") = R.add(add112, add107) + layer_norm39: R.Tensor((1, 1, 768), dtype="float16") = R.nn.layer_norm(add113, gpt_neox_layers_7_input_layernorm_weight2, gpt_neox_layers_7_input_layernorm_bias2, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims77: R.Tensor((768, 2304), dtype="float16") = R.permute_dims(gpt_neox_layers_7_attention_query_key_value_weight2, axes=None) + matmul77: R.Tensor((1, 1, 2304), dtype="float16") = R.matmul(layer_norm39, permute_dims77, out_dtype="void") + add114: R.Tensor((1, 1, 2304), dtype="float16") = R.add(matmul77, gpt_neox_layers_7_attention_query_key_value_bias2) + reshape76: R.Tensor((1, 1, 36, 64), dtype="float16") = R.reshape(add114, R.shape([1, 1, 36, 64])) + reshape77: R.Tensor((1, 36, 64), dtype="float16") = R.reshape(reshape76, R.shape([1, 36, 64])) + lv20 = R.call_dps_packed("vm.builtin.attention_kv_cache_attention_with_fused_qkv", (paged_kv_cache, R.prim_value(7), R.prim_value(T.float32(1.0)), reshape77), out_sinfo=R.Tensor((1, 12, 64), dtype="float16")) + reshape78: R.Tensor((1, 1, 12, 64), dtype="float16") = R.reshape(lv20, R.shape([1, 1, 12, 64])) + reshape79: R.Tensor((1, 1, 768), dtype="float16") = R.reshape(reshape78, R.shape([1, 1, 768])) + permute_dims78: R.Tensor((768, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_7_attention_dense_weight2, axes=None) + matmul78: R.Tensor((1, 1, 768), dtype="float16") = R.matmul(reshape79, permute_dims78, out_dtype="void") + add115: R.Tensor((1, 1, 768), dtype="float16") = R.add(matmul78, gpt_neox_layers_7_attention_dense_bias2) + layer_norm40: R.Tensor((1, 1, 768), dtype="float16") = R.nn.layer_norm(add113, gpt_neox_layers_7_post_attention_layernorm_weight2, gpt_neox_layers_7_post_attention_layernorm_bias2, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims79: R.Tensor((768, 3072), dtype="float16") = R.permute_dims(gpt_neox_layers_7_mlp_dense_h_to_4h_weight2, axes=None) + matmul79: R.Tensor((1, 1, 3072), dtype="float32") = R.matmul(layer_norm40, permute_dims79, out_dtype="float32") + add116: R.Tensor((1, 1, 3072), dtype="float32") = R.add(matmul79, gpt_neox_layers_7_mlp_dense_h_to_4h_bias2) + gelu19: R.Tensor((1, 1, 3072), dtype="float32") = R.nn.gelu(add116) + astype39: R.Tensor((1, 1, 3072), dtype="float16") = R.astype(gelu19, dtype="float16") + permute_dims80: R.Tensor((3072, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_7_mlp_dense_4h_to_h_weight2, axes=None) + matmul80: R.Tensor((1, 1, 768), dtype="float32") = R.matmul(astype39, permute_dims80, out_dtype="float32") + add117: R.Tensor((1, 1, 768), dtype="float32") = R.add(matmul80, gpt_neox_layers_7_mlp_dense_4h_to_h_bias2) + astype40: R.Tensor((1, 1, 768), dtype="float16") = R.astype(add117, dtype="float16") + add118: R.Tensor((1, 1, 768), dtype="float16") = R.add(astype40, add115) + add119: R.Tensor((1, 1, 768), dtype="float16") = R.add(add118, add113) + layer_norm41: R.Tensor((1, 1, 768), dtype="float16") = R.nn.layer_norm(add119, gpt_neox_layers_8_input_layernorm_weight2, gpt_neox_layers_8_input_layernorm_bias2, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims81: R.Tensor((768, 2304), dtype="float16") = R.permute_dims(gpt_neox_layers_8_attention_query_key_value_weight2, axes=None) + matmul81: R.Tensor((1, 1, 2304), dtype="float16") = R.matmul(layer_norm41, permute_dims81, out_dtype="void") + add120: R.Tensor((1, 1, 2304), dtype="float16") = R.add(matmul81, gpt_neox_layers_8_attention_query_key_value_bias2) + reshape80: R.Tensor((1, 1, 36, 64), dtype="float16") = R.reshape(add120, R.shape([1, 1, 36, 64])) + reshape81: R.Tensor((1, 36, 64), dtype="float16") = R.reshape(reshape80, R.shape([1, 36, 64])) + lv21 = R.call_dps_packed("vm.builtin.attention_kv_cache_attention_with_fused_qkv", (paged_kv_cache, R.prim_value(8), R.prim_value(T.float32(1.0)), reshape81), out_sinfo=R.Tensor((1, 12, 64), dtype="float16")) + reshape82: R.Tensor((1, 1, 12, 64), dtype="float16") = R.reshape(lv21, R.shape([1, 1, 12, 64])) + reshape83: R.Tensor((1, 1, 768), dtype="float16") = R.reshape(reshape82, R.shape([1, 1, 768])) + permute_dims82: R.Tensor((768, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_8_attention_dense_weight2, axes=None) + matmul82: R.Tensor((1, 1, 768), dtype="float16") = R.matmul(reshape83, permute_dims82, out_dtype="void") + add121: R.Tensor((1, 1, 768), dtype="float16") = R.add(matmul82, gpt_neox_layers_8_attention_dense_bias2) + layer_norm42: R.Tensor((1, 1, 768), dtype="float16") = R.nn.layer_norm(add119, gpt_neox_layers_8_post_attention_layernorm_weight2, gpt_neox_layers_8_post_attention_layernorm_bias2, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims83: R.Tensor((768, 3072), dtype="float16") = R.permute_dims(gpt_neox_layers_8_mlp_dense_h_to_4h_weight2, axes=None) + matmul83: R.Tensor((1, 1, 3072), dtype="float32") = R.matmul(layer_norm42, permute_dims83, out_dtype="float32") + add122: R.Tensor((1, 1, 3072), dtype="float32") = R.add(matmul83, gpt_neox_layers_8_mlp_dense_h_to_4h_bias2) + gelu20: R.Tensor((1, 1, 3072), dtype="float32") = R.nn.gelu(add122) + astype41: R.Tensor((1, 1, 3072), dtype="float16") = R.astype(gelu20, dtype="float16") + permute_dims84: R.Tensor((3072, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_8_mlp_dense_4h_to_h_weight2, axes=None) + matmul84: R.Tensor((1, 1, 768), dtype="float32") = R.matmul(astype41, permute_dims84, out_dtype="float32") + add123: R.Tensor((1, 1, 768), dtype="float32") = R.add(matmul84, gpt_neox_layers_8_mlp_dense_4h_to_h_bias2) + astype42: R.Tensor((1, 1, 768), dtype="float16") = R.astype(add123, dtype="float16") + add124: R.Tensor((1, 1, 768), dtype="float16") = R.add(astype42, add121) + add125: R.Tensor((1, 1, 768), dtype="float16") = R.add(add124, add119) + layer_norm43: R.Tensor((1, 1, 768), dtype="float16") = R.nn.layer_norm(add125, gpt_neox_layers_9_input_layernorm_weight2, gpt_neox_layers_9_input_layernorm_bias2, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims85: R.Tensor((768, 2304), dtype="float16") = R.permute_dims(gpt_neox_layers_9_attention_query_key_value_weight2, axes=None) + matmul85: R.Tensor((1, 1, 2304), dtype="float16") = R.matmul(layer_norm43, permute_dims85, out_dtype="void") + add126: R.Tensor((1, 1, 2304), dtype="float16") = R.add(matmul85, gpt_neox_layers_9_attention_query_key_value_bias2) + reshape84: R.Tensor((1, 1, 36, 64), dtype="float16") = R.reshape(add126, R.shape([1, 1, 36, 64])) + reshape85: R.Tensor((1, 36, 64), dtype="float16") = R.reshape(reshape84, R.shape([1, 36, 64])) + lv22 = R.call_dps_packed("vm.builtin.attention_kv_cache_attention_with_fused_qkv", (paged_kv_cache, R.prim_value(9), R.prim_value(T.float32(1.0)), reshape85), out_sinfo=R.Tensor((1, 12, 64), dtype="float16")) + reshape86: R.Tensor((1, 1, 12, 64), dtype="float16") = R.reshape(lv22, R.shape([1, 1, 12, 64])) + reshape87: R.Tensor((1, 1, 768), dtype="float16") = R.reshape(reshape86, R.shape([1, 1, 768])) + permute_dims86: R.Tensor((768, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_9_attention_dense_weight2, axes=None) + matmul86: R.Tensor((1, 1, 768), dtype="float16") = R.matmul(reshape87, permute_dims86, out_dtype="void") + add127: R.Tensor((1, 1, 768), dtype="float16") = R.add(matmul86, gpt_neox_layers_9_attention_dense_bias2) + layer_norm44: R.Tensor((1, 1, 768), dtype="float16") = R.nn.layer_norm(add125, gpt_neox_layers_9_post_attention_layernorm_weight2, gpt_neox_layers_9_post_attention_layernorm_bias2, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims87: R.Tensor((768, 3072), dtype="float16") = R.permute_dims(gpt_neox_layers_9_mlp_dense_h_to_4h_weight2, axes=None) + matmul87: R.Tensor((1, 1, 3072), dtype="float32") = R.matmul(layer_norm44, permute_dims87, out_dtype="float32") + add128: R.Tensor((1, 1, 3072), dtype="float32") = R.add(matmul87, gpt_neox_layers_9_mlp_dense_h_to_4h_bias2) + gelu21: R.Tensor((1, 1, 3072), dtype="float32") = R.nn.gelu(add128) + astype43: R.Tensor((1, 1, 3072), dtype="float16") = R.astype(gelu21, dtype="float16") + permute_dims88: R.Tensor((3072, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_9_mlp_dense_4h_to_h_weight2, axes=None) + matmul88: R.Tensor((1, 1, 768), dtype="float32") = R.matmul(astype43, permute_dims88, out_dtype="float32") + add129: R.Tensor((1, 1, 768), dtype="float32") = R.add(matmul88, gpt_neox_layers_9_mlp_dense_4h_to_h_bias2) + astype44: R.Tensor((1, 1, 768), dtype="float16") = R.astype(add129, dtype="float16") + add130: R.Tensor((1, 1, 768), dtype="float16") = R.add(astype44, add127) + add131: R.Tensor((1, 1, 768), dtype="float16") = R.add(add130, add125) + layer_norm45: R.Tensor((1, 1, 768), dtype="float16") = R.nn.layer_norm(add131, gpt_neox_layers_10_input_layernorm_weight2, gpt_neox_layers_10_input_layernorm_bias2, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims89: R.Tensor((768, 2304), dtype="float16") = R.permute_dims(gpt_neox_layers_10_attention_query_key_value_weight2, axes=None) + matmul89: R.Tensor((1, 1, 2304), dtype="float16") = R.matmul(layer_norm45, permute_dims89, out_dtype="void") + add132: R.Tensor((1, 1, 2304), dtype="float16") = R.add(matmul89, gpt_neox_layers_10_attention_query_key_value_bias2) + reshape88: R.Tensor((1, 1, 36, 64), dtype="float16") = R.reshape(add132, R.shape([1, 1, 36, 64])) + reshape89: R.Tensor((1, 36, 64), dtype="float16") = R.reshape(reshape88, R.shape([1, 36, 64])) + lv23 = R.call_dps_packed("vm.builtin.attention_kv_cache_attention_with_fused_qkv", (paged_kv_cache, R.prim_value(10), R.prim_value(T.float32(1.0)), reshape89), out_sinfo=R.Tensor((1, 12, 64), dtype="float16")) + reshape90: R.Tensor((1, 1, 12, 64), dtype="float16") = R.reshape(lv23, R.shape([1, 1, 12, 64])) + reshape91: R.Tensor((1, 1, 768), dtype="float16") = R.reshape(reshape90, R.shape([1, 1, 768])) + permute_dims90: R.Tensor((768, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_10_attention_dense_weight2, axes=None) + matmul90: R.Tensor((1, 1, 768), dtype="float16") = R.matmul(reshape91, permute_dims90, out_dtype="void") + add133: R.Tensor((1, 1, 768), dtype="float16") = R.add(matmul90, gpt_neox_layers_10_attention_dense_bias2) + layer_norm46: R.Tensor((1, 1, 768), dtype="float16") = R.nn.layer_norm(add131, gpt_neox_layers_10_post_attention_layernorm_weight2, gpt_neox_layers_10_post_attention_layernorm_bias2, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims91: R.Tensor((768, 3072), dtype="float16") = R.permute_dims(gpt_neox_layers_10_mlp_dense_h_to_4h_weight2, axes=None) + matmul91: R.Tensor((1, 1, 3072), dtype="float32") = R.matmul(layer_norm46, permute_dims91, out_dtype="float32") + add134: R.Tensor((1, 1, 3072), dtype="float32") = R.add(matmul91, gpt_neox_layers_10_mlp_dense_h_to_4h_bias2) + gelu22: R.Tensor((1, 1, 3072), dtype="float32") = R.nn.gelu(add134) + astype45: R.Tensor((1, 1, 3072), dtype="float16") = R.astype(gelu22, dtype="float16") + permute_dims92: R.Tensor((3072, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_10_mlp_dense_4h_to_h_weight2, axes=None) + matmul92: R.Tensor((1, 1, 768), dtype="float32") = R.matmul(astype45, permute_dims92, out_dtype="float32") + add135: R.Tensor((1, 1, 768), dtype="float32") = R.add(matmul92, gpt_neox_layers_10_mlp_dense_4h_to_h_bias2) + astype46: R.Tensor((1, 1, 768), dtype="float16") = R.astype(add135, dtype="float16") + add136: R.Tensor((1, 1, 768), dtype="float16") = R.add(astype46, add133) + add137: R.Tensor((1, 1, 768), dtype="float16") = R.add(add136, add131) + layer_norm47: R.Tensor((1, 1, 768), dtype="float16") = R.nn.layer_norm(add137, gpt_neox_layers_11_input_layernorm_weight2, gpt_neox_layers_11_input_layernorm_bias2, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims93: R.Tensor((768, 2304), dtype="float16") = R.permute_dims(gpt_neox_layers_11_attention_query_key_value_weight2, axes=None) + matmul93: R.Tensor((1, 1, 2304), dtype="float16") = R.matmul(layer_norm47, permute_dims93, out_dtype="void") + add138: R.Tensor((1, 1, 2304), dtype="float16") = R.add(matmul93, gpt_neox_layers_11_attention_query_key_value_bias2) + reshape92: R.Tensor((1, 1, 36, 64), dtype="float16") = R.reshape(add138, R.shape([1, 1, 36, 64])) + reshape93: R.Tensor((1, 36, 64), dtype="float16") = R.reshape(reshape92, R.shape([1, 36, 64])) + lv24 = R.call_dps_packed("vm.builtin.attention_kv_cache_attention_with_fused_qkv", (paged_kv_cache, R.prim_value(11), R.prim_value(T.float32(1.0)), reshape93), out_sinfo=R.Tensor((1, 12, 64), dtype="float16")) + reshape94: R.Tensor((1, 1, 12, 64), dtype="float16") = R.reshape(lv24, R.shape([1, 1, 12, 64])) + reshape95: R.Tensor((1, 1, 768), dtype="float16") = R.reshape(reshape94, R.shape([1, 1, 768])) + permute_dims94: R.Tensor((768, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_11_attention_dense_weight2, axes=None) + matmul94: R.Tensor((1, 1, 768), dtype="float16") = R.matmul(reshape95, permute_dims94, out_dtype="void") + add139: R.Tensor((1, 1, 768), dtype="float16") = R.add(matmul94, gpt_neox_layers_11_attention_dense_bias2) + layer_norm48: R.Tensor((1, 1, 768), dtype="float16") = R.nn.layer_norm(add137, gpt_neox_layers_11_post_attention_layernorm_weight2, gpt_neox_layers_11_post_attention_layernorm_bias2, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims95: R.Tensor((768, 3072), dtype="float16") = R.permute_dims(gpt_neox_layers_11_mlp_dense_h_to_4h_weight2, axes=None) + matmul95: R.Tensor((1, 1, 3072), dtype="float32") = R.matmul(layer_norm48, permute_dims95, out_dtype="float32") + add140: R.Tensor((1, 1, 3072), dtype="float32") = R.add(matmul95, gpt_neox_layers_11_mlp_dense_h_to_4h_bias2) + gelu23: R.Tensor((1, 1, 3072), dtype="float32") = R.nn.gelu(add140) + astype47: R.Tensor((1, 1, 3072), dtype="float16") = R.astype(gelu23, dtype="float16") + permute_dims96: R.Tensor((3072, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_11_mlp_dense_4h_to_h_weight2, axes=None) + matmul96: R.Tensor((1, 1, 768), dtype="float32") = R.matmul(astype47, permute_dims96, out_dtype="float32") + add141: R.Tensor((1, 1, 768), dtype="float32") = R.add(matmul96, gpt_neox_layers_11_mlp_dense_4h_to_h_bias2) + astype48: R.Tensor((1, 1, 768), dtype="float16") = R.astype(add141, dtype="float16") + add142: R.Tensor((1, 1, 768), dtype="float16") = R.add(astype48, add139) + add143: R.Tensor((1, 1, 768), dtype="float16") = R.add(add142, add137) + layer_norm49: R.Tensor((1, 1, 768), dtype="float16") = R.nn.layer_norm(add143, gpt_neox_final_layer_norm_weight2, gpt_neox_final_layer_norm_bias2, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims97: R.Tensor((768, vocab_size), dtype="float16") = R.permute_dims(embed_out_weight2, axes=None) + matmul97: R.Tensor((1, 1, vocab_size), dtype="float16") = R.matmul(layer_norm49, permute_dims97, out_dtype="void") + astype49: R.Tensor((1, 1, vocab_size), dtype="float32") = R.astype(matmul97, dtype="float32") + gv2: R.Tuple(R.Tensor((1, 1, vocab_size), dtype="float32"), R.Object) = astype49, paged_kv_cache + R.output(gv2) + return gv2 + + @R.function + def embed(input_ids: R.Tensor(("seq_len",), dtype="int32"), packed_params: R.Tuple(R.Tensor(("vocab_size", 768), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((2304, 768), dtype="float16"), R.Tensor((2304,), dtype="float16"), R.Tensor((768, 768), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((3072, 768), dtype="float16"), R.Tensor((3072,), dtype="float32"), R.Tensor((768, 3072), dtype="float16"), R.Tensor((768,), dtype="float32"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((2304, 768), dtype="float16"), R.Tensor((2304,), dtype="float16"), R.Tensor((768, 768), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((3072, 768), dtype="float16"), R.Tensor((3072,), dtype="float32"), R.Tensor((768, 3072), dtype="float16"), R.Tensor((768,), dtype="float32"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((2304, 768), dtype="float16"), R.Tensor((2304,), dtype="float16"), R.Tensor((768, 768), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((3072, 768), dtype="float16"), R.Tensor((3072,), dtype="float32"), R.Tensor((768, 3072), dtype="float16"), R.Tensor((768,), dtype="float32"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((2304, 768), dtype="float16"), R.Tensor((2304,), dtype="float16"), R.Tensor((768, 768), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((3072, 768), dtype="float16"), R.Tensor((3072,), dtype="float32"), R.Tensor((768, 3072), dtype="float16"), R.Tensor((768,), dtype="float32"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((2304, 768), dtype="float16"), R.Tensor((2304,), dtype="float16"), R.Tensor((768, 768), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((3072, 768), dtype="float16"), R.Tensor((3072,), dtype="float32"), R.Tensor((768, 3072), dtype="float16"), R.Tensor((768,), dtype="float32"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((2304, 768), dtype="float16"), R.Tensor((2304,), dtype="float16"), R.Tensor((768, 768), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((3072, 768), dtype="float16"), R.Tensor((3072,), dtype="float32"), R.Tensor((768, 3072), dtype="float16"), R.Tensor((768,), dtype="float32"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((2304, 768), dtype="float16"), R.Tensor((2304,), dtype="float16"), R.Tensor((768, 768), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((3072, 768), dtype="float16"), R.Tensor((3072,), dtype="float32"), R.Tensor((768, 3072), dtype="float16"), R.Tensor((768,), dtype="float32"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((2304, 768), dtype="float16"), R.Tensor((2304,), dtype="float16"), R.Tensor((768, 768), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((3072, 768), dtype="float16"), R.Tensor((3072,), dtype="float32"), R.Tensor((768, 3072), dtype="float16"), R.Tensor((768,), dtype="float32"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((2304, 768), dtype="float16"), R.Tensor((2304,), dtype="float16"), R.Tensor((768, 768), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((3072, 768), dtype="float16"), R.Tensor((3072,), dtype="float32"), R.Tensor((768, 3072), dtype="float16"), R.Tensor((768,), dtype="float32"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((2304, 768), dtype="float16"), R.Tensor((2304,), dtype="float16"), R.Tensor((768, 768), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((3072, 768), dtype="float16"), R.Tensor((3072,), dtype="float32"), R.Tensor((768, 3072), dtype="float16"), R.Tensor((768,), dtype="float32"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((2304, 768), dtype="float16"), R.Tensor((2304,), dtype="float16"), R.Tensor((768, 768), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((3072, 768), dtype="float16"), R.Tensor((3072,), dtype="float32"), R.Tensor((768, 3072), dtype="float16"), R.Tensor((768,), dtype="float32"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((2304, 768), dtype="float16"), R.Tensor((2304,), dtype="float16"), R.Tensor((768, 768), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((3072, 768), dtype="float16"), R.Tensor((3072,), dtype="float32"), R.Tensor((768, 3072), dtype="float16"), R.Tensor((768,), dtype="float32"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor(("vocab_size", 768), dtype="float16"))) -> R.Tensor(("seq_len", 768), dtype="float16"): + seq_len = T.int64() + vocab_size = T.int64() + R.func_attr({"num_input": 1, "relax.memory_plan_dynamic_func_output": True, "tir_non_negative_var": ["vocab_size"], "tir_var_upper_bound": {"batch_size": 128, "seq_len": 2048, "total_seq_len": 2048}}) + with R.dataflow(): + gpt_neox_embed_in_weight: R.Tensor((vocab_size, 768), dtype="float16") = packed_params[0] + gpt_neox_layers_0_input_layernorm_weight: R.Tensor((768,), dtype="float16") = packed_params[1] + gpt_neox_layers_0_input_layernorm_bias: R.Tensor((768,), dtype="float16") = packed_params[2] + gpt_neox_layers_0_post_attention_layernorm_weight: R.Tensor((768,), dtype="float16") = packed_params[3] + gpt_neox_layers_0_post_attention_layernorm_bias: R.Tensor((768,), dtype="float16") = packed_params[4] + gpt_neox_layers_0_attention_query_key_value_weight: R.Tensor((2304, 768), dtype="float16") = packed_params[5] + gpt_neox_layers_0_attention_query_key_value_bias: R.Tensor((2304,), dtype="float16") = packed_params[6] + gpt_neox_layers_0_attention_dense_weight: R.Tensor((768, 768), dtype="float16") = packed_params[7] + gpt_neox_layers_0_attention_dense_bias: R.Tensor((768,), dtype="float16") = packed_params[8] + gpt_neox_layers_0_mlp_dense_h_to_4h_weight: R.Tensor((3072, 768), dtype="float16") = packed_params[9] + gpt_neox_layers_0_mlp_dense_h_to_4h_bias: R.Tensor((3072,), dtype="float32") = packed_params[10] + gpt_neox_layers_0_mlp_dense_4h_to_h_weight: R.Tensor((768, 3072), dtype="float16") = packed_params[11] + gpt_neox_layers_0_mlp_dense_4h_to_h_bias: R.Tensor((768,), dtype="float32") = packed_params[12] + gpt_neox_layers_1_input_layernorm_weight: R.Tensor((768,), dtype="float16") = packed_params[13] + gpt_neox_layers_1_input_layernorm_bias: R.Tensor((768,), dtype="float16") = packed_params[14] + gpt_neox_layers_1_post_attention_layernorm_weight: R.Tensor((768,), dtype="float16") = packed_params[15] + gpt_neox_layers_1_post_attention_layernorm_bias: R.Tensor((768,), dtype="float16") = packed_params[16] + gpt_neox_layers_1_attention_query_key_value_weight: R.Tensor((2304, 768), dtype="float16") = packed_params[17] + gpt_neox_layers_1_attention_query_key_value_bias: R.Tensor((2304,), dtype="float16") = packed_params[18] + gpt_neox_layers_1_attention_dense_weight: R.Tensor((768, 768), dtype="float16") = packed_params[19] + gpt_neox_layers_1_attention_dense_bias: R.Tensor((768,), dtype="float16") = packed_params[20] + gpt_neox_layers_1_mlp_dense_h_to_4h_weight: R.Tensor((3072, 768), dtype="float16") = packed_params[21] + gpt_neox_layers_1_mlp_dense_h_to_4h_bias: R.Tensor((3072,), dtype="float32") = packed_params[22] + gpt_neox_layers_1_mlp_dense_4h_to_h_weight: R.Tensor((768, 3072), dtype="float16") = packed_params[23] + gpt_neox_layers_1_mlp_dense_4h_to_h_bias: R.Tensor((768,), dtype="float32") = packed_params[24] + gpt_neox_layers_2_input_layernorm_weight: R.Tensor((768,), dtype="float16") = packed_params[25] + gpt_neox_layers_2_input_layernorm_bias: R.Tensor((768,), dtype="float16") = packed_params[26] + gpt_neox_layers_2_post_attention_layernorm_weight: R.Tensor((768,), dtype="float16") = packed_params[27] + gpt_neox_layers_2_post_attention_layernorm_bias: R.Tensor((768,), dtype="float16") = packed_params[28] + gpt_neox_layers_2_attention_query_key_value_weight: R.Tensor((2304, 768), dtype="float16") = packed_params[29] + gpt_neox_layers_2_attention_query_key_value_bias: R.Tensor((2304,), dtype="float16") = packed_params[30] + gpt_neox_layers_2_attention_dense_weight: R.Tensor((768, 768), dtype="float16") = packed_params[31] + gpt_neox_layers_2_attention_dense_bias: R.Tensor((768,), dtype="float16") = packed_params[32] + gpt_neox_layers_2_mlp_dense_h_to_4h_weight: R.Tensor((3072, 768), dtype="float16") = packed_params[33] + gpt_neox_layers_2_mlp_dense_h_to_4h_bias: R.Tensor((3072,), dtype="float32") = packed_params[34] + gpt_neox_layers_2_mlp_dense_4h_to_h_weight: R.Tensor((768, 3072), dtype="float16") = packed_params[35] + gpt_neox_layers_2_mlp_dense_4h_to_h_bias: R.Tensor((768,), dtype="float32") = packed_params[36] + gpt_neox_layers_3_input_layernorm_weight: R.Tensor((768,), dtype="float16") = packed_params[37] + gpt_neox_layers_3_input_layernorm_bias: R.Tensor((768,), dtype="float16") = packed_params[38] + gpt_neox_layers_3_post_attention_layernorm_weight: R.Tensor((768,), dtype="float16") = packed_params[39] + gpt_neox_layers_3_post_attention_layernorm_bias: R.Tensor((768,), dtype="float16") = packed_params[40] + gpt_neox_layers_3_attention_query_key_value_weight: R.Tensor((2304, 768), dtype="float16") = packed_params[41] + gpt_neox_layers_3_attention_query_key_value_bias: R.Tensor((2304,), dtype="float16") = packed_params[42] + gpt_neox_layers_3_attention_dense_weight: R.Tensor((768, 768), dtype="float16") = packed_params[43] + gpt_neox_layers_3_attention_dense_bias: R.Tensor((768,), dtype="float16") = packed_params[44] + gpt_neox_layers_3_mlp_dense_h_to_4h_weight: R.Tensor((3072, 768), dtype="float16") = packed_params[45] + gpt_neox_layers_3_mlp_dense_h_to_4h_bias: R.Tensor((3072,), dtype="float32") = packed_params[46] + gpt_neox_layers_3_mlp_dense_4h_to_h_weight: R.Tensor((768, 3072), dtype="float16") = packed_params[47] + gpt_neox_layers_3_mlp_dense_4h_to_h_bias: R.Tensor((768,), dtype="float32") = packed_params[48] + gpt_neox_layers_4_input_layernorm_weight: R.Tensor((768,), dtype="float16") = packed_params[49] + gpt_neox_layers_4_input_layernorm_bias: R.Tensor((768,), dtype="float16") = packed_params[50] + gpt_neox_layers_4_post_attention_layernorm_weight: R.Tensor((768,), dtype="float16") = packed_params[51] + gpt_neox_layers_4_post_attention_layernorm_bias: R.Tensor((768,), dtype="float16") = packed_params[52] + gpt_neox_layers_4_attention_query_key_value_weight: R.Tensor((2304, 768), dtype="float16") = packed_params[53] + gpt_neox_layers_4_attention_query_key_value_bias: R.Tensor((2304,), dtype="float16") = packed_params[54] + gpt_neox_layers_4_attention_dense_weight: R.Tensor((768, 768), dtype="float16") = packed_params[55] + gpt_neox_layers_4_attention_dense_bias: R.Tensor((768,), dtype="float16") = packed_params[56] + gpt_neox_layers_4_mlp_dense_h_to_4h_weight: R.Tensor((3072, 768), dtype="float16") = packed_params[57] + gpt_neox_layers_4_mlp_dense_h_to_4h_bias: R.Tensor((3072,), dtype="float32") = packed_params[58] + gpt_neox_layers_4_mlp_dense_4h_to_h_weight: R.Tensor((768, 3072), dtype="float16") = packed_params[59] + gpt_neox_layers_4_mlp_dense_4h_to_h_bias: R.Tensor((768,), dtype="float32") = packed_params[60] + gpt_neox_layers_5_input_layernorm_weight: R.Tensor((768,), dtype="float16") = packed_params[61] + gpt_neox_layers_5_input_layernorm_bias: R.Tensor((768,), dtype="float16") = packed_params[62] + gpt_neox_layers_5_post_attention_layernorm_weight: R.Tensor((768,), dtype="float16") = packed_params[63] + gpt_neox_layers_5_post_attention_layernorm_bias: R.Tensor((768,), dtype="float16") = packed_params[64] + gpt_neox_layers_5_attention_query_key_value_weight: R.Tensor((2304, 768), dtype="float16") = packed_params[65] + gpt_neox_layers_5_attention_query_key_value_bias: R.Tensor((2304,), dtype="float16") = packed_params[66] + gpt_neox_layers_5_attention_dense_weight: R.Tensor((768, 768), dtype="float16") = packed_params[67] + gpt_neox_layers_5_attention_dense_bias: R.Tensor((768,), dtype="float16") = packed_params[68] + gpt_neox_layers_5_mlp_dense_h_to_4h_weight: R.Tensor((3072, 768), dtype="float16") = packed_params[69] + gpt_neox_layers_5_mlp_dense_h_to_4h_bias: R.Tensor((3072,), dtype="float32") = packed_params[70] + gpt_neox_layers_5_mlp_dense_4h_to_h_weight: R.Tensor((768, 3072), dtype="float16") = packed_params[71] + gpt_neox_layers_5_mlp_dense_4h_to_h_bias: R.Tensor((768,), dtype="float32") = packed_params[72] + gpt_neox_layers_6_input_layernorm_weight: R.Tensor((768,), dtype="float16") = packed_params[73] + gpt_neox_layers_6_input_layernorm_bias: R.Tensor((768,), dtype="float16") = packed_params[74] + gpt_neox_layers_6_post_attention_layernorm_weight: R.Tensor((768,), dtype="float16") = packed_params[75] + gpt_neox_layers_6_post_attention_layernorm_bias: R.Tensor((768,), dtype="float16") = packed_params[76] + gpt_neox_layers_6_attention_query_key_value_weight: R.Tensor((2304, 768), dtype="float16") = packed_params[77] + gpt_neox_layers_6_attention_query_key_value_bias: R.Tensor((2304,), dtype="float16") = packed_params[78] + gpt_neox_layers_6_attention_dense_weight: R.Tensor((768, 768), dtype="float16") = packed_params[79] + gpt_neox_layers_6_attention_dense_bias: R.Tensor((768,), dtype="float16") = packed_params[80] + gpt_neox_layers_6_mlp_dense_h_to_4h_weight: R.Tensor((3072, 768), dtype="float16") = packed_params[81] + gpt_neox_layers_6_mlp_dense_h_to_4h_bias: R.Tensor((3072,), dtype="float32") = packed_params[82] + gpt_neox_layers_6_mlp_dense_4h_to_h_weight: R.Tensor((768, 3072), dtype="float16") = packed_params[83] + gpt_neox_layers_6_mlp_dense_4h_to_h_bias: R.Tensor((768,), dtype="float32") = packed_params[84] + gpt_neox_layers_7_input_layernorm_weight: R.Tensor((768,), dtype="float16") = packed_params[85] + gpt_neox_layers_7_input_layernorm_bias: R.Tensor((768,), dtype="float16") = packed_params[86] + gpt_neox_layers_7_post_attention_layernorm_weight: R.Tensor((768,), dtype="float16") = packed_params[87] + gpt_neox_layers_7_post_attention_layernorm_bias: R.Tensor((768,), dtype="float16") = packed_params[88] + gpt_neox_layers_7_attention_query_key_value_weight: R.Tensor((2304, 768), dtype="float16") = packed_params[89] + gpt_neox_layers_7_attention_query_key_value_bias: R.Tensor((2304,), dtype="float16") = packed_params[90] + gpt_neox_layers_7_attention_dense_weight: R.Tensor((768, 768), dtype="float16") = packed_params[91] + gpt_neox_layers_7_attention_dense_bias: R.Tensor((768,), dtype="float16") = packed_params[92] + gpt_neox_layers_7_mlp_dense_h_to_4h_weight: R.Tensor((3072, 768), dtype="float16") = packed_params[93] + gpt_neox_layers_7_mlp_dense_h_to_4h_bias: R.Tensor((3072,), dtype="float32") = packed_params[94] + gpt_neox_layers_7_mlp_dense_4h_to_h_weight: R.Tensor((768, 3072), dtype="float16") = packed_params[95] + gpt_neox_layers_7_mlp_dense_4h_to_h_bias: R.Tensor((768,), dtype="float32") = packed_params[96] + gpt_neox_layers_8_input_layernorm_weight: R.Tensor((768,), dtype="float16") = packed_params[97] + gpt_neox_layers_8_input_layernorm_bias: R.Tensor((768,), dtype="float16") = packed_params[98] + gpt_neox_layers_8_post_attention_layernorm_weight: R.Tensor((768,), dtype="float16") = packed_params[99] + gpt_neox_layers_8_post_attention_layernorm_bias: R.Tensor((768,), dtype="float16") = packed_params[100] + gpt_neox_layers_8_attention_query_key_value_weight: R.Tensor((2304, 768), dtype="float16") = packed_params[101] + gpt_neox_layers_8_attention_query_key_value_bias: R.Tensor((2304,), dtype="float16") = packed_params[102] + gpt_neox_layers_8_attention_dense_weight: R.Tensor((768, 768), dtype="float16") = packed_params[103] + gpt_neox_layers_8_attention_dense_bias: R.Tensor((768,), dtype="float16") = packed_params[104] + gpt_neox_layers_8_mlp_dense_h_to_4h_weight: R.Tensor((3072, 768), dtype="float16") = packed_params[105] + gpt_neox_layers_8_mlp_dense_h_to_4h_bias: R.Tensor((3072,), dtype="float32") = packed_params[106] + gpt_neox_layers_8_mlp_dense_4h_to_h_weight: R.Tensor((768, 3072), dtype="float16") = packed_params[107] + gpt_neox_layers_8_mlp_dense_4h_to_h_bias: R.Tensor((768,), dtype="float32") = packed_params[108] + gpt_neox_layers_9_input_layernorm_weight: R.Tensor((768,), dtype="float16") = packed_params[109] + gpt_neox_layers_9_input_layernorm_bias: R.Tensor((768,), dtype="float16") = packed_params[110] + gpt_neox_layers_9_post_attention_layernorm_weight: R.Tensor((768,), dtype="float16") = packed_params[111] + gpt_neox_layers_9_post_attention_layernorm_bias: R.Tensor((768,), dtype="float16") = packed_params[112] + gpt_neox_layers_9_attention_query_key_value_weight: R.Tensor((2304, 768), dtype="float16") = packed_params[113] + gpt_neox_layers_9_attention_query_key_value_bias: R.Tensor((2304,), dtype="float16") = packed_params[114] + gpt_neox_layers_9_attention_dense_weight: R.Tensor((768, 768), dtype="float16") = packed_params[115] + gpt_neox_layers_9_attention_dense_bias: R.Tensor((768,), dtype="float16") = packed_params[116] + gpt_neox_layers_9_mlp_dense_h_to_4h_weight: R.Tensor((3072, 768), dtype="float16") = packed_params[117] + gpt_neox_layers_9_mlp_dense_h_to_4h_bias: R.Tensor((3072,), dtype="float32") = packed_params[118] + gpt_neox_layers_9_mlp_dense_4h_to_h_weight: R.Tensor((768, 3072), dtype="float16") = packed_params[119] + gpt_neox_layers_9_mlp_dense_4h_to_h_bias: R.Tensor((768,), dtype="float32") = packed_params[120] + gpt_neox_layers_10_input_layernorm_weight: R.Tensor((768,), dtype="float16") = packed_params[121] + gpt_neox_layers_10_input_layernorm_bias: R.Tensor((768,), dtype="float16") = packed_params[122] + gpt_neox_layers_10_post_attention_layernorm_weight: R.Tensor((768,), dtype="float16") = packed_params[123] + gpt_neox_layers_10_post_attention_layernorm_bias: R.Tensor((768,), dtype="float16") = packed_params[124] + gpt_neox_layers_10_attention_query_key_value_weight: R.Tensor((2304, 768), dtype="float16") = packed_params[125] + gpt_neox_layers_10_attention_query_key_value_bias: R.Tensor((2304,), dtype="float16") = packed_params[126] + gpt_neox_layers_10_attention_dense_weight: R.Tensor((768, 768), dtype="float16") = packed_params[127] + gpt_neox_layers_10_attention_dense_bias: R.Tensor((768,), dtype="float16") = packed_params[128] + gpt_neox_layers_10_mlp_dense_h_to_4h_weight: R.Tensor((3072, 768), dtype="float16") = packed_params[129] + gpt_neox_layers_10_mlp_dense_h_to_4h_bias: R.Tensor((3072,), dtype="float32") = packed_params[130] + gpt_neox_layers_10_mlp_dense_4h_to_h_weight: R.Tensor((768, 3072), dtype="float16") = packed_params[131] + gpt_neox_layers_10_mlp_dense_4h_to_h_bias: R.Tensor((768,), dtype="float32") = packed_params[132] + gpt_neox_layers_11_input_layernorm_weight: R.Tensor((768,), dtype="float16") = packed_params[133] + gpt_neox_layers_11_input_layernorm_bias: R.Tensor((768,), dtype="float16") = packed_params[134] + gpt_neox_layers_11_post_attention_layernorm_weight: R.Tensor((768,), dtype="float16") = packed_params[135] + gpt_neox_layers_11_post_attention_layernorm_bias: R.Tensor((768,), dtype="float16") = packed_params[136] + gpt_neox_layers_11_attention_query_key_value_weight: R.Tensor((2304, 768), dtype="float16") = packed_params[137] + gpt_neox_layers_11_attention_query_key_value_bias: R.Tensor((2304,), dtype="float16") = packed_params[138] + gpt_neox_layers_11_attention_dense_weight: R.Tensor((768, 768), dtype="float16") = packed_params[139] + gpt_neox_layers_11_attention_dense_bias: R.Tensor((768,), dtype="float16") = packed_params[140] + gpt_neox_layers_11_mlp_dense_h_to_4h_weight: R.Tensor((3072, 768), dtype="float16") = packed_params[141] + gpt_neox_layers_11_mlp_dense_h_to_4h_bias: R.Tensor((3072,), dtype="float32") = packed_params[142] + gpt_neox_layers_11_mlp_dense_4h_to_h_weight: R.Tensor((768, 3072), dtype="float16") = packed_params[143] + gpt_neox_layers_11_mlp_dense_4h_to_h_bias: R.Tensor((768,), dtype="float32") = packed_params[144] + gpt_neox_final_layer_norm_weight: R.Tensor((768,), dtype="float16") = packed_params[145] + gpt_neox_final_layer_norm_bias: R.Tensor((768,), dtype="float16") = packed_params[146] + embed_out_weight: R.Tensor((vocab_size, 768), dtype="float16") = packed_params[147] + take: R.Tensor((seq_len, 768), dtype="float16") = R.take(gpt_neox_embed_in_weight, input_ids, axis=0) + gv: R.Tensor((seq_len, 768), dtype="float16") = take + R.output(gv) + return gv + + @R.function + def prefill(input_embed: R.Tensor((1, "seq_len", 768), dtype="float16"), paged_kv_cache: R.Object, packed_params: R.Tuple(R.Tensor(("vocab_size", 768), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((2304, 768), dtype="float16"), R.Tensor((2304,), dtype="float16"), R.Tensor((768, 768), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((3072, 768), dtype="float16"), R.Tensor((3072,), dtype="float32"), R.Tensor((768, 3072), dtype="float16"), R.Tensor((768,), dtype="float32"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((2304, 768), dtype="float16"), R.Tensor((2304,), dtype="float16"), R.Tensor((768, 768), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((3072, 768), dtype="float16"), R.Tensor((3072,), dtype="float32"), R.Tensor((768, 3072), dtype="float16"), R.Tensor((768,), dtype="float32"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((2304, 768), dtype="float16"), R.Tensor((2304,), dtype="float16"), R.Tensor((768, 768), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((3072, 768), dtype="float16"), R.Tensor((3072,), dtype="float32"), R.Tensor((768, 3072), dtype="float16"), R.Tensor((768,), dtype="float32"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((2304, 768), dtype="float16"), R.Tensor((2304,), dtype="float16"), R.Tensor((768, 768), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((3072, 768), dtype="float16"), R.Tensor((3072,), dtype="float32"), R.Tensor((768, 3072), dtype="float16"), R.Tensor((768,), dtype="float32"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((2304, 768), dtype="float16"), R.Tensor((2304,), dtype="float16"), R.Tensor((768, 768), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((3072, 768), dtype="float16"), R.Tensor((3072,), dtype="float32"), R.Tensor((768, 3072), dtype="float16"), R.Tensor((768,), dtype="float32"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((2304, 768), dtype="float16"), R.Tensor((2304,), dtype="float16"), R.Tensor((768, 768), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((3072, 768), dtype="float16"), R.Tensor((3072,), dtype="float32"), R.Tensor((768, 3072), dtype="float16"), R.Tensor((768,), dtype="float32"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((2304, 768), dtype="float16"), R.Tensor((2304,), dtype="float16"), R.Tensor((768, 768), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((3072, 768), dtype="float16"), R.Tensor((3072,), dtype="float32"), R.Tensor((768, 3072), dtype="float16"), R.Tensor((768,), dtype="float32"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((2304, 768), dtype="float16"), R.Tensor((2304,), dtype="float16"), R.Tensor((768, 768), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((3072, 768), dtype="float16"), R.Tensor((3072,), dtype="float32"), R.Tensor((768, 3072), dtype="float16"), R.Tensor((768,), dtype="float32"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((2304, 768), dtype="float16"), R.Tensor((2304,), dtype="float16"), R.Tensor((768, 768), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((3072, 768), dtype="float16"), R.Tensor((3072,), dtype="float32"), R.Tensor((768, 3072), dtype="float16"), R.Tensor((768,), dtype="float32"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((2304, 768), dtype="float16"), R.Tensor((2304,), dtype="float16"), R.Tensor((768, 768), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((3072, 768), dtype="float16"), R.Tensor((3072,), dtype="float32"), R.Tensor((768, 3072), dtype="float16"), R.Tensor((768,), dtype="float32"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((2304, 768), dtype="float16"), R.Tensor((2304,), dtype="float16"), R.Tensor((768, 768), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((3072, 768), dtype="float16"), R.Tensor((3072,), dtype="float32"), R.Tensor((768, 3072), dtype="float16"), R.Tensor((768,), dtype="float32"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((2304, 768), dtype="float16"), R.Tensor((2304,), dtype="float16"), R.Tensor((768, 768), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor((3072, 768), dtype="float16"), R.Tensor((3072,), dtype="float32"), R.Tensor((768, 3072), dtype="float16"), R.Tensor((768,), dtype="float32"), R.Tensor((768,), dtype="float16"), R.Tensor((768,), dtype="float16"), R.Tensor(("vocab_size", 768), dtype="float16"))) -> R.Tuple(R.Tensor((1, 1, "vocab_size"), dtype="float32"), R.Object): + vocab_size = T.int64() + seq_len = T.int64() + R.func_attr({"num_input": 2, "pipeline_parallel_stages": 1, "relax.memory_plan_dynamic_func_output": True, "tir_non_negative_var": ["vocab_size"], "tir_var_upper_bound": {"batch_size": 128, "seq_len": 2048, "total_seq_len": 2048}}) + cls = Module + with R.dataflow(): + gpt_neox_embed_in_weight1: R.Tensor((vocab_size, 768), dtype="float16") = packed_params[0] + gpt_neox_layers_0_input_layernorm_weight1: R.Tensor((768,), dtype="float16") = packed_params[1] + gpt_neox_layers_0_input_layernorm_bias1: R.Tensor((768,), dtype="float16") = packed_params[2] + gpt_neox_layers_0_post_attention_layernorm_weight1: R.Tensor((768,), dtype="float16") = packed_params[3] + gpt_neox_layers_0_post_attention_layernorm_bias1: R.Tensor((768,), dtype="float16") = packed_params[4] + gpt_neox_layers_0_attention_query_key_value_weight1: R.Tensor((2304, 768), dtype="float16") = packed_params[5] + gpt_neox_layers_0_attention_query_key_value_bias1: R.Tensor((2304,), dtype="float16") = packed_params[6] + gpt_neox_layers_0_attention_dense_weight1: R.Tensor((768, 768), dtype="float16") = packed_params[7] + gpt_neox_layers_0_attention_dense_bias1: R.Tensor((768,), dtype="float16") = packed_params[8] + gpt_neox_layers_0_mlp_dense_h_to_4h_weight1: R.Tensor((3072, 768), dtype="float16") = packed_params[9] + gpt_neox_layers_0_mlp_dense_h_to_4h_bias1: R.Tensor((3072,), dtype="float32") = packed_params[10] + gpt_neox_layers_0_mlp_dense_4h_to_h_weight1: R.Tensor((768, 3072), dtype="float16") = packed_params[11] + gpt_neox_layers_0_mlp_dense_4h_to_h_bias1: R.Tensor((768,), dtype="float32") = packed_params[12] + gpt_neox_layers_1_input_layernorm_weight1: R.Tensor((768,), dtype="float16") = packed_params[13] + gpt_neox_layers_1_input_layernorm_bias1: R.Tensor((768,), dtype="float16") = packed_params[14] + gpt_neox_layers_1_post_attention_layernorm_weight1: R.Tensor((768,), dtype="float16") = packed_params[15] + gpt_neox_layers_1_post_attention_layernorm_bias1: R.Tensor((768,), dtype="float16") = packed_params[16] + gpt_neox_layers_1_attention_query_key_value_weight1: R.Tensor((2304, 768), dtype="float16") = packed_params[17] + gpt_neox_layers_1_attention_query_key_value_bias1: R.Tensor((2304,), dtype="float16") = packed_params[18] + gpt_neox_layers_1_attention_dense_weight1: R.Tensor((768, 768), dtype="float16") = packed_params[19] + gpt_neox_layers_1_attention_dense_bias1: R.Tensor((768,), dtype="float16") = packed_params[20] + gpt_neox_layers_1_mlp_dense_h_to_4h_weight1: R.Tensor((3072, 768), dtype="float16") = packed_params[21] + gpt_neox_layers_1_mlp_dense_h_to_4h_bias1: R.Tensor((3072,), dtype="float32") = packed_params[22] + gpt_neox_layers_1_mlp_dense_4h_to_h_weight1: R.Tensor((768, 3072), dtype="float16") = packed_params[23] + gpt_neox_layers_1_mlp_dense_4h_to_h_bias1: R.Tensor((768,), dtype="float32") = packed_params[24] + gpt_neox_layers_2_input_layernorm_weight1: R.Tensor((768,), dtype="float16") = packed_params[25] + gpt_neox_layers_2_input_layernorm_bias1: R.Tensor((768,), dtype="float16") = packed_params[26] + gpt_neox_layers_2_post_attention_layernorm_weight1: R.Tensor((768,), dtype="float16") = packed_params[27] + gpt_neox_layers_2_post_attention_layernorm_bias1: R.Tensor((768,), dtype="float16") = packed_params[28] + gpt_neox_layers_2_attention_query_key_value_weight1: R.Tensor((2304, 768), dtype="float16") = packed_params[29] + gpt_neox_layers_2_attention_query_key_value_bias1: R.Tensor((2304,), dtype="float16") = packed_params[30] + gpt_neox_layers_2_attention_dense_weight1: R.Tensor((768, 768), dtype="float16") = packed_params[31] + gpt_neox_layers_2_attention_dense_bias1: R.Tensor((768,), dtype="float16") = packed_params[32] + gpt_neox_layers_2_mlp_dense_h_to_4h_weight1: R.Tensor((3072, 768), dtype="float16") = packed_params[33] + gpt_neox_layers_2_mlp_dense_h_to_4h_bias1: R.Tensor((3072,), dtype="float32") = packed_params[34] + gpt_neox_layers_2_mlp_dense_4h_to_h_weight1: R.Tensor((768, 3072), dtype="float16") = packed_params[35] + gpt_neox_layers_2_mlp_dense_4h_to_h_bias1: R.Tensor((768,), dtype="float32") = packed_params[36] + gpt_neox_layers_3_input_layernorm_weight1: R.Tensor((768,), dtype="float16") = packed_params[37] + gpt_neox_layers_3_input_layernorm_bias1: R.Tensor((768,), dtype="float16") = packed_params[38] + gpt_neox_layers_3_post_attention_layernorm_weight1: R.Tensor((768,), dtype="float16") = packed_params[39] + gpt_neox_layers_3_post_attention_layernorm_bias1: R.Tensor((768,), dtype="float16") = packed_params[40] + gpt_neox_layers_3_attention_query_key_value_weight1: R.Tensor((2304, 768), dtype="float16") = packed_params[41] + gpt_neox_layers_3_attention_query_key_value_bias1: R.Tensor((2304,), dtype="float16") = packed_params[42] + gpt_neox_layers_3_attention_dense_weight1: R.Tensor((768, 768), dtype="float16") = packed_params[43] + gpt_neox_layers_3_attention_dense_bias1: R.Tensor((768,), dtype="float16") = packed_params[44] + gpt_neox_layers_3_mlp_dense_h_to_4h_weight1: R.Tensor((3072, 768), dtype="float16") = packed_params[45] + gpt_neox_layers_3_mlp_dense_h_to_4h_bias1: R.Tensor((3072,), dtype="float32") = packed_params[46] + gpt_neox_layers_3_mlp_dense_4h_to_h_weight1: R.Tensor((768, 3072), dtype="float16") = packed_params[47] + gpt_neox_layers_3_mlp_dense_4h_to_h_bias1: R.Tensor((768,), dtype="float32") = packed_params[48] + gpt_neox_layers_4_input_layernorm_weight1: R.Tensor((768,), dtype="float16") = packed_params[49] + gpt_neox_layers_4_input_layernorm_bias1: R.Tensor((768,), dtype="float16") = packed_params[50] + gpt_neox_layers_4_post_attention_layernorm_weight1: R.Tensor((768,), dtype="float16") = packed_params[51] + gpt_neox_layers_4_post_attention_layernorm_bias1: R.Tensor((768,), dtype="float16") = packed_params[52] + gpt_neox_layers_4_attention_query_key_value_weight1: R.Tensor((2304, 768), dtype="float16") = packed_params[53] + gpt_neox_layers_4_attention_query_key_value_bias1: R.Tensor((2304,), dtype="float16") = packed_params[54] + gpt_neox_layers_4_attention_dense_weight1: R.Tensor((768, 768), dtype="float16") = packed_params[55] + gpt_neox_layers_4_attention_dense_bias1: R.Tensor((768,), dtype="float16") = packed_params[56] + gpt_neox_layers_4_mlp_dense_h_to_4h_weight1: R.Tensor((3072, 768), dtype="float16") = packed_params[57] + gpt_neox_layers_4_mlp_dense_h_to_4h_bias1: R.Tensor((3072,), dtype="float32") = packed_params[58] + gpt_neox_layers_4_mlp_dense_4h_to_h_weight1: R.Tensor((768, 3072), dtype="float16") = packed_params[59] + gpt_neox_layers_4_mlp_dense_4h_to_h_bias1: R.Tensor((768,), dtype="float32") = packed_params[60] + gpt_neox_layers_5_input_layernorm_weight1: R.Tensor((768,), dtype="float16") = packed_params[61] + gpt_neox_layers_5_input_layernorm_bias1: R.Tensor((768,), dtype="float16") = packed_params[62] + gpt_neox_layers_5_post_attention_layernorm_weight1: R.Tensor((768,), dtype="float16") = packed_params[63] + gpt_neox_layers_5_post_attention_layernorm_bias1: R.Tensor((768,), dtype="float16") = packed_params[64] + gpt_neox_layers_5_attention_query_key_value_weight1: R.Tensor((2304, 768), dtype="float16") = packed_params[65] + gpt_neox_layers_5_attention_query_key_value_bias1: R.Tensor((2304,), dtype="float16") = packed_params[66] + gpt_neox_layers_5_attention_dense_weight1: R.Tensor((768, 768), dtype="float16") = packed_params[67] + gpt_neox_layers_5_attention_dense_bias1: R.Tensor((768,), dtype="float16") = packed_params[68] + gpt_neox_layers_5_mlp_dense_h_to_4h_weight1: R.Tensor((3072, 768), dtype="float16") = packed_params[69] + gpt_neox_layers_5_mlp_dense_h_to_4h_bias1: R.Tensor((3072,), dtype="float32") = packed_params[70] + gpt_neox_layers_5_mlp_dense_4h_to_h_weight1: R.Tensor((768, 3072), dtype="float16") = packed_params[71] + gpt_neox_layers_5_mlp_dense_4h_to_h_bias1: R.Tensor((768,), dtype="float32") = packed_params[72] + gpt_neox_layers_6_input_layernorm_weight1: R.Tensor((768,), dtype="float16") = packed_params[73] + gpt_neox_layers_6_input_layernorm_bias1: R.Tensor((768,), dtype="float16") = packed_params[74] + gpt_neox_layers_6_post_attention_layernorm_weight1: R.Tensor((768,), dtype="float16") = packed_params[75] + gpt_neox_layers_6_post_attention_layernorm_bias1: R.Tensor((768,), dtype="float16") = packed_params[76] + gpt_neox_layers_6_attention_query_key_value_weight1: R.Tensor((2304, 768), dtype="float16") = packed_params[77] + gpt_neox_layers_6_attention_query_key_value_bias1: R.Tensor((2304,), dtype="float16") = packed_params[78] + gpt_neox_layers_6_attention_dense_weight1: R.Tensor((768, 768), dtype="float16") = packed_params[79] + gpt_neox_layers_6_attention_dense_bias1: R.Tensor((768,), dtype="float16") = packed_params[80] + gpt_neox_layers_6_mlp_dense_h_to_4h_weight1: R.Tensor((3072, 768), dtype="float16") = packed_params[81] + gpt_neox_layers_6_mlp_dense_h_to_4h_bias1: R.Tensor((3072,), dtype="float32") = packed_params[82] + gpt_neox_layers_6_mlp_dense_4h_to_h_weight1: R.Tensor((768, 3072), dtype="float16") = packed_params[83] + gpt_neox_layers_6_mlp_dense_4h_to_h_bias1: R.Tensor((768,), dtype="float32") = packed_params[84] + gpt_neox_layers_7_input_layernorm_weight1: R.Tensor((768,), dtype="float16") = packed_params[85] + gpt_neox_layers_7_input_layernorm_bias1: R.Tensor((768,), dtype="float16") = packed_params[86] + gpt_neox_layers_7_post_attention_layernorm_weight1: R.Tensor((768,), dtype="float16") = packed_params[87] + gpt_neox_layers_7_post_attention_layernorm_bias1: R.Tensor((768,), dtype="float16") = packed_params[88] + gpt_neox_layers_7_attention_query_key_value_weight1: R.Tensor((2304, 768), dtype="float16") = packed_params[89] + gpt_neox_layers_7_attention_query_key_value_bias1: R.Tensor((2304,), dtype="float16") = packed_params[90] + gpt_neox_layers_7_attention_dense_weight1: R.Tensor((768, 768), dtype="float16") = packed_params[91] + gpt_neox_layers_7_attention_dense_bias1: R.Tensor((768,), dtype="float16") = packed_params[92] + gpt_neox_layers_7_mlp_dense_h_to_4h_weight1: R.Tensor((3072, 768), dtype="float16") = packed_params[93] + gpt_neox_layers_7_mlp_dense_h_to_4h_bias1: R.Tensor((3072,), dtype="float32") = packed_params[94] + gpt_neox_layers_7_mlp_dense_4h_to_h_weight1: R.Tensor((768, 3072), dtype="float16") = packed_params[95] + gpt_neox_layers_7_mlp_dense_4h_to_h_bias1: R.Tensor((768,), dtype="float32") = packed_params[96] + gpt_neox_layers_8_input_layernorm_weight1: R.Tensor((768,), dtype="float16") = packed_params[97] + gpt_neox_layers_8_input_layernorm_bias1: R.Tensor((768,), dtype="float16") = packed_params[98] + gpt_neox_layers_8_post_attention_layernorm_weight1: R.Tensor((768,), dtype="float16") = packed_params[99] + gpt_neox_layers_8_post_attention_layernorm_bias1: R.Tensor((768,), dtype="float16") = packed_params[100] + gpt_neox_layers_8_attention_query_key_value_weight1: R.Tensor((2304, 768), dtype="float16") = packed_params[101] + gpt_neox_layers_8_attention_query_key_value_bias1: R.Tensor((2304,), dtype="float16") = packed_params[102] + gpt_neox_layers_8_attention_dense_weight1: R.Tensor((768, 768), dtype="float16") = packed_params[103] + gpt_neox_layers_8_attention_dense_bias1: R.Tensor((768,), dtype="float16") = packed_params[104] + gpt_neox_layers_8_mlp_dense_h_to_4h_weight1: R.Tensor((3072, 768), dtype="float16") = packed_params[105] + gpt_neox_layers_8_mlp_dense_h_to_4h_bias1: R.Tensor((3072,), dtype="float32") = packed_params[106] + gpt_neox_layers_8_mlp_dense_4h_to_h_weight1: R.Tensor((768, 3072), dtype="float16") = packed_params[107] + gpt_neox_layers_8_mlp_dense_4h_to_h_bias1: R.Tensor((768,), dtype="float32") = packed_params[108] + gpt_neox_layers_9_input_layernorm_weight1: R.Tensor((768,), dtype="float16") = packed_params[109] + gpt_neox_layers_9_input_layernorm_bias1: R.Tensor((768,), dtype="float16") = packed_params[110] + gpt_neox_layers_9_post_attention_layernorm_weight1: R.Tensor((768,), dtype="float16") = packed_params[111] + gpt_neox_layers_9_post_attention_layernorm_bias1: R.Tensor((768,), dtype="float16") = packed_params[112] + gpt_neox_layers_9_attention_query_key_value_weight1: R.Tensor((2304, 768), dtype="float16") = packed_params[113] + gpt_neox_layers_9_attention_query_key_value_bias1: R.Tensor((2304,), dtype="float16") = packed_params[114] + gpt_neox_layers_9_attention_dense_weight1: R.Tensor((768, 768), dtype="float16") = packed_params[115] + gpt_neox_layers_9_attention_dense_bias1: R.Tensor((768,), dtype="float16") = packed_params[116] + gpt_neox_layers_9_mlp_dense_h_to_4h_weight1: R.Tensor((3072, 768), dtype="float16") = packed_params[117] + gpt_neox_layers_9_mlp_dense_h_to_4h_bias1: R.Tensor((3072,), dtype="float32") = packed_params[118] + gpt_neox_layers_9_mlp_dense_4h_to_h_weight1: R.Tensor((768, 3072), dtype="float16") = packed_params[119] + gpt_neox_layers_9_mlp_dense_4h_to_h_bias1: R.Tensor((768,), dtype="float32") = packed_params[120] + gpt_neox_layers_10_input_layernorm_weight1: R.Tensor((768,), dtype="float16") = packed_params[121] + gpt_neox_layers_10_input_layernorm_bias1: R.Tensor((768,), dtype="float16") = packed_params[122] + gpt_neox_layers_10_post_attention_layernorm_weight1: R.Tensor((768,), dtype="float16") = packed_params[123] + gpt_neox_layers_10_post_attention_layernorm_bias1: R.Tensor((768,), dtype="float16") = packed_params[124] + gpt_neox_layers_10_attention_query_key_value_weight1: R.Tensor((2304, 768), dtype="float16") = packed_params[125] + gpt_neox_layers_10_attention_query_key_value_bias1: R.Tensor((2304,), dtype="float16") = packed_params[126] + gpt_neox_layers_10_attention_dense_weight1: R.Tensor((768, 768), dtype="float16") = packed_params[127] + gpt_neox_layers_10_attention_dense_bias1: R.Tensor((768,), dtype="float16") = packed_params[128] + gpt_neox_layers_10_mlp_dense_h_to_4h_weight1: R.Tensor((3072, 768), dtype="float16") = packed_params[129] + gpt_neox_layers_10_mlp_dense_h_to_4h_bias1: R.Tensor((3072,), dtype="float32") = packed_params[130] + gpt_neox_layers_10_mlp_dense_4h_to_h_weight1: R.Tensor((768, 3072), dtype="float16") = packed_params[131] + gpt_neox_layers_10_mlp_dense_4h_to_h_bias1: R.Tensor((768,), dtype="float32") = packed_params[132] + gpt_neox_layers_11_input_layernorm_weight1: R.Tensor((768,), dtype="float16") = packed_params[133] + gpt_neox_layers_11_input_layernorm_bias1: R.Tensor((768,), dtype="float16") = packed_params[134] + gpt_neox_layers_11_post_attention_layernorm_weight1: R.Tensor((768,), dtype="float16") = packed_params[135] + gpt_neox_layers_11_post_attention_layernorm_bias1: R.Tensor((768,), dtype="float16") = packed_params[136] + gpt_neox_layers_11_attention_query_key_value_weight1: R.Tensor((2304, 768), dtype="float16") = packed_params[137] + gpt_neox_layers_11_attention_query_key_value_bias1: R.Tensor((2304,), dtype="float16") = packed_params[138] + gpt_neox_layers_11_attention_dense_weight1: R.Tensor((768, 768), dtype="float16") = packed_params[139] + gpt_neox_layers_11_attention_dense_bias1: R.Tensor((768,), dtype="float16") = packed_params[140] + gpt_neox_layers_11_mlp_dense_h_to_4h_weight1: R.Tensor((3072, 768), dtype="float16") = packed_params[141] + gpt_neox_layers_11_mlp_dense_h_to_4h_bias1: R.Tensor((3072,), dtype="float32") = packed_params[142] + gpt_neox_layers_11_mlp_dense_4h_to_h_weight1: R.Tensor((768, 3072), dtype="float16") = packed_params[143] + gpt_neox_layers_11_mlp_dense_4h_to_h_bias1: R.Tensor((768,), dtype="float32") = packed_params[144] + gpt_neox_final_layer_norm_weight1: R.Tensor((768,), dtype="float16") = packed_params[145] + gpt_neox_final_layer_norm_bias1: R.Tensor((768,), dtype="float16") = packed_params[146] + embed_out_weight1: R.Tensor((vocab_size, 768), dtype="float16") = packed_params[147] + layer_norm: R.Tensor((1, seq_len, 768), dtype="float16") = R.nn.layer_norm(input_embed, gpt_neox_layers_0_input_layernorm_weight1, gpt_neox_layers_0_input_layernorm_bias1, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims: R.Tensor((768, 2304), dtype="float16") = R.permute_dims(gpt_neox_layers_0_attention_query_key_value_weight1, axes=None) + matmul: R.Tensor((1, seq_len, 2304), dtype="float16") = R.matmul(layer_norm, permute_dims, out_dtype="void") + add: R.Tensor((1, seq_len, 2304), dtype="float16") = R.add(matmul, gpt_neox_layers_0_attention_query_key_value_bias1) + reshape: R.Tensor((1, seq_len, 36, 64), dtype="float16") = R.reshape(add, R.shape([1, seq_len, 36, 64])) + reshape1: R.Tensor((seq_len, 36, 64), dtype="float16") = R.reshape(reshape, R.shape([seq_len, 36, 64])) + lv = R.call_dps_packed("vm.builtin.attention_kv_cache_attention_with_fused_qkv", (paged_kv_cache, R.prim_value(0), R.prim_value(T.float32(1.0)), reshape1), out_sinfo=R.Tensor((seq_len, 12, 64), dtype="float16")) + reshape2: R.Tensor((1, seq_len, 12, 64), dtype="float16") = R.reshape(lv, R.shape([1, seq_len, 12, 64])) + reshape3: R.Tensor((1, seq_len, 768), dtype="float16") = R.reshape(reshape2, R.shape([1, seq_len, 768])) + permute_dims1: R.Tensor((768, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_0_attention_dense_weight1, axes=None) + matmul1: R.Tensor((1, seq_len, 768), dtype="float16") = R.matmul(reshape3, permute_dims1, out_dtype="void") + add1: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(matmul1, gpt_neox_layers_0_attention_dense_bias1) + layer_norm1: R.Tensor((1, seq_len, 768), dtype="float16") = R.nn.layer_norm(input_embed, gpt_neox_layers_0_post_attention_layernorm_weight1, gpt_neox_layers_0_post_attention_layernorm_bias1, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims2: R.Tensor((768, 3072), dtype="float16") = R.permute_dims(gpt_neox_layers_0_mlp_dense_h_to_4h_weight1, axes=None) + matmul2: R.Tensor((1, seq_len, 3072), dtype="float32") = R.matmul(layer_norm1, permute_dims2, out_dtype="float32") + add2: R.Tensor((1, seq_len, 3072), dtype="float32") = R.add(matmul2, gpt_neox_layers_0_mlp_dense_h_to_4h_bias1) + gelu: R.Tensor((1, seq_len, 3072), dtype="float32") = R.nn.gelu(add2) + astype: R.Tensor((1, seq_len, 3072), dtype="float16") = R.astype(gelu, dtype="float16") + permute_dims3: R.Tensor((3072, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_0_mlp_dense_4h_to_h_weight1, axes=None) + matmul3: R.Tensor((1, seq_len, 768), dtype="float32") = R.matmul(astype, permute_dims3, out_dtype="float32") + add3: R.Tensor((1, seq_len, 768), dtype="float32") = R.add(matmul3, gpt_neox_layers_0_mlp_dense_4h_to_h_bias1) + astype1: R.Tensor((1, seq_len, 768), dtype="float16") = R.astype(add3, dtype="float16") + add4: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(astype1, add1) + add5: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(add4, input_embed) + layer_norm2: R.Tensor((1, seq_len, 768), dtype="float16") = R.nn.layer_norm(add5, gpt_neox_layers_1_input_layernorm_weight1, gpt_neox_layers_1_input_layernorm_bias1, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims4: R.Tensor((768, 2304), dtype="float16") = R.permute_dims(gpt_neox_layers_1_attention_query_key_value_weight1, axes=None) + matmul4: R.Tensor((1, seq_len, 2304), dtype="float16") = R.matmul(layer_norm2, permute_dims4, out_dtype="void") + add6: R.Tensor((1, seq_len, 2304), dtype="float16") = R.add(matmul4, gpt_neox_layers_1_attention_query_key_value_bias1) + reshape4: R.Tensor((1, seq_len, 36, 64), dtype="float16") = R.reshape(add6, R.shape([1, seq_len, 36, 64])) + reshape5: R.Tensor((seq_len, 36, 64), dtype="float16") = R.reshape(reshape4, R.shape([seq_len, 36, 64])) + lv1 = R.call_dps_packed("vm.builtin.attention_kv_cache_attention_with_fused_qkv", (paged_kv_cache, R.prim_value(1), R.prim_value(T.float32(1.0)), reshape5), out_sinfo=R.Tensor((seq_len, 12, 64), dtype="float16")) + reshape6: R.Tensor((1, seq_len, 12, 64), dtype="float16") = R.reshape(lv1, R.shape([1, seq_len, 12, 64])) + reshape7: R.Tensor((1, seq_len, 768), dtype="float16") = R.reshape(reshape6, R.shape([1, seq_len, 768])) + permute_dims5: R.Tensor((768, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_1_attention_dense_weight1, axes=None) + matmul5: R.Tensor((1, seq_len, 768), dtype="float16") = R.matmul(reshape7, permute_dims5, out_dtype="void") + add7: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(matmul5, gpt_neox_layers_1_attention_dense_bias1) + layer_norm3: R.Tensor((1, seq_len, 768), dtype="float16") = R.nn.layer_norm(add5, gpt_neox_layers_1_post_attention_layernorm_weight1, gpt_neox_layers_1_post_attention_layernorm_bias1, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims6: R.Tensor((768, 3072), dtype="float16") = R.permute_dims(gpt_neox_layers_1_mlp_dense_h_to_4h_weight1, axes=None) + matmul6: R.Tensor((1, seq_len, 3072), dtype="float32") = R.matmul(layer_norm3, permute_dims6, out_dtype="float32") + add8: R.Tensor((1, seq_len, 3072), dtype="float32") = R.add(matmul6, gpt_neox_layers_1_mlp_dense_h_to_4h_bias1) + gelu1: R.Tensor((1, seq_len, 3072), dtype="float32") = R.nn.gelu(add8) + astype2: R.Tensor((1, seq_len, 3072), dtype="float16") = R.astype(gelu1, dtype="float16") + permute_dims7: R.Tensor((3072, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_1_mlp_dense_4h_to_h_weight1, axes=None) + matmul7: R.Tensor((1, seq_len, 768), dtype="float32") = R.matmul(astype2, permute_dims7, out_dtype="float32") + add9: R.Tensor((1, seq_len, 768), dtype="float32") = R.add(matmul7, gpt_neox_layers_1_mlp_dense_4h_to_h_bias1) + astype3: R.Tensor((1, seq_len, 768), dtype="float16") = R.astype(add9, dtype="float16") + add10: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(astype3, add7) + add11: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(add10, add5) + layer_norm4: R.Tensor((1, seq_len, 768), dtype="float16") = R.nn.layer_norm(add11, gpt_neox_layers_2_input_layernorm_weight1, gpt_neox_layers_2_input_layernorm_bias1, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims8: R.Tensor((768, 2304), dtype="float16") = R.permute_dims(gpt_neox_layers_2_attention_query_key_value_weight1, axes=None) + matmul8: R.Tensor((1, seq_len, 2304), dtype="float16") = R.matmul(layer_norm4, permute_dims8, out_dtype="void") + add12: R.Tensor((1, seq_len, 2304), dtype="float16") = R.add(matmul8, gpt_neox_layers_2_attention_query_key_value_bias1) + reshape8: R.Tensor((1, seq_len, 36, 64), dtype="float16") = R.reshape(add12, R.shape([1, seq_len, 36, 64])) + reshape9: R.Tensor((seq_len, 36, 64), dtype="float16") = R.reshape(reshape8, R.shape([seq_len, 36, 64])) + lv2 = R.call_dps_packed("vm.builtin.attention_kv_cache_attention_with_fused_qkv", (paged_kv_cache, R.prim_value(2), R.prim_value(T.float32(1.0)), reshape9), out_sinfo=R.Tensor((seq_len, 12, 64), dtype="float16")) + reshape10: R.Tensor((1, seq_len, 12, 64), dtype="float16") = R.reshape(lv2, R.shape([1, seq_len, 12, 64])) + reshape11: R.Tensor((1, seq_len, 768), dtype="float16") = R.reshape(reshape10, R.shape([1, seq_len, 768])) + permute_dims9: R.Tensor((768, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_2_attention_dense_weight1, axes=None) + matmul9: R.Tensor((1, seq_len, 768), dtype="float16") = R.matmul(reshape11, permute_dims9, out_dtype="void") + add13: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(matmul9, gpt_neox_layers_2_attention_dense_bias1) + layer_norm5: R.Tensor((1, seq_len, 768), dtype="float16") = R.nn.layer_norm(add11, gpt_neox_layers_2_post_attention_layernorm_weight1, gpt_neox_layers_2_post_attention_layernorm_bias1, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims10: R.Tensor((768, 3072), dtype="float16") = R.permute_dims(gpt_neox_layers_2_mlp_dense_h_to_4h_weight1, axes=None) + matmul10: R.Tensor((1, seq_len, 3072), dtype="float32") = R.matmul(layer_norm5, permute_dims10, out_dtype="float32") + add14: R.Tensor((1, seq_len, 3072), dtype="float32") = R.add(matmul10, gpt_neox_layers_2_mlp_dense_h_to_4h_bias1) + gelu2: R.Tensor((1, seq_len, 3072), dtype="float32") = R.nn.gelu(add14) + astype4: R.Tensor((1, seq_len, 3072), dtype="float16") = R.astype(gelu2, dtype="float16") + permute_dims11: R.Tensor((3072, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_2_mlp_dense_4h_to_h_weight1, axes=None) + matmul11: R.Tensor((1, seq_len, 768), dtype="float32") = R.matmul(astype4, permute_dims11, out_dtype="float32") + add15: R.Tensor((1, seq_len, 768), dtype="float32") = R.add(matmul11, gpt_neox_layers_2_mlp_dense_4h_to_h_bias1) + astype5: R.Tensor((1, seq_len, 768), dtype="float16") = R.astype(add15, dtype="float16") + add16: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(astype5, add13) + add17: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(add16, add11) + layer_norm6: R.Tensor((1, seq_len, 768), dtype="float16") = R.nn.layer_norm(add17, gpt_neox_layers_3_input_layernorm_weight1, gpt_neox_layers_3_input_layernorm_bias1, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims12: R.Tensor((768, 2304), dtype="float16") = R.permute_dims(gpt_neox_layers_3_attention_query_key_value_weight1, axes=None) + matmul12: R.Tensor((1, seq_len, 2304), dtype="float16") = R.matmul(layer_norm6, permute_dims12, out_dtype="void") + add18: R.Tensor((1, seq_len, 2304), dtype="float16") = R.add(matmul12, gpt_neox_layers_3_attention_query_key_value_bias1) + reshape12: R.Tensor((1, seq_len, 36, 64), dtype="float16") = R.reshape(add18, R.shape([1, seq_len, 36, 64])) + reshape13: R.Tensor((seq_len, 36, 64), dtype="float16") = R.reshape(reshape12, R.shape([seq_len, 36, 64])) + lv3 = R.call_dps_packed("vm.builtin.attention_kv_cache_attention_with_fused_qkv", (paged_kv_cache, R.prim_value(3), R.prim_value(T.float32(1.0)), reshape13), out_sinfo=R.Tensor((seq_len, 12, 64), dtype="float16")) + reshape14: R.Tensor((1, seq_len, 12, 64), dtype="float16") = R.reshape(lv3, R.shape([1, seq_len, 12, 64])) + reshape15: R.Tensor((1, seq_len, 768), dtype="float16") = R.reshape(reshape14, R.shape([1, seq_len, 768])) + permute_dims13: R.Tensor((768, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_3_attention_dense_weight1, axes=None) + matmul13: R.Tensor((1, seq_len, 768), dtype="float16") = R.matmul(reshape15, permute_dims13, out_dtype="void") + add19: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(matmul13, gpt_neox_layers_3_attention_dense_bias1) + layer_norm7: R.Tensor((1, seq_len, 768), dtype="float16") = R.nn.layer_norm(add17, gpt_neox_layers_3_post_attention_layernorm_weight1, gpt_neox_layers_3_post_attention_layernorm_bias1, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims14: R.Tensor((768, 3072), dtype="float16") = R.permute_dims(gpt_neox_layers_3_mlp_dense_h_to_4h_weight1, axes=None) + matmul14: R.Tensor((1, seq_len, 3072), dtype="float32") = R.matmul(layer_norm7, permute_dims14, out_dtype="float32") + add20: R.Tensor((1, seq_len, 3072), dtype="float32") = R.add(matmul14, gpt_neox_layers_3_mlp_dense_h_to_4h_bias1) + gelu3: R.Tensor((1, seq_len, 3072), dtype="float32") = R.nn.gelu(add20) + astype6: R.Tensor((1, seq_len, 3072), dtype="float16") = R.astype(gelu3, dtype="float16") + permute_dims15: R.Tensor((3072, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_3_mlp_dense_4h_to_h_weight1, axes=None) + matmul15: R.Tensor((1, seq_len, 768), dtype="float32") = R.matmul(astype6, permute_dims15, out_dtype="float32") + add21: R.Tensor((1, seq_len, 768), dtype="float32") = R.add(matmul15, gpt_neox_layers_3_mlp_dense_4h_to_h_bias1) + astype7: R.Tensor((1, seq_len, 768), dtype="float16") = R.astype(add21, dtype="float16") + add22: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(astype7, add19) + add23: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(add22, add17) + layer_norm8: R.Tensor((1, seq_len, 768), dtype="float16") = R.nn.layer_norm(add23, gpt_neox_layers_4_input_layernorm_weight1, gpt_neox_layers_4_input_layernorm_bias1, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims16: R.Tensor((768, 2304), dtype="float16") = R.permute_dims(gpt_neox_layers_4_attention_query_key_value_weight1, axes=None) + matmul16: R.Tensor((1, seq_len, 2304), dtype="float16") = R.matmul(layer_norm8, permute_dims16, out_dtype="void") + add24: R.Tensor((1, seq_len, 2304), dtype="float16") = R.add(matmul16, gpt_neox_layers_4_attention_query_key_value_bias1) + reshape16: R.Tensor((1, seq_len, 36, 64), dtype="float16") = R.reshape(add24, R.shape([1, seq_len, 36, 64])) + reshape17: R.Tensor((seq_len, 36, 64), dtype="float16") = R.reshape(reshape16, R.shape([seq_len, 36, 64])) + lv4 = R.call_dps_packed("vm.builtin.attention_kv_cache_attention_with_fused_qkv", (paged_kv_cache, R.prim_value(4), R.prim_value(T.float32(1.0)), reshape17), out_sinfo=R.Tensor((seq_len, 12, 64), dtype="float16")) + reshape18: R.Tensor((1, seq_len, 12, 64), dtype="float16") = R.reshape(lv4, R.shape([1, seq_len, 12, 64])) + reshape19: R.Tensor((1, seq_len, 768), dtype="float16") = R.reshape(reshape18, R.shape([1, seq_len, 768])) + permute_dims17: R.Tensor((768, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_4_attention_dense_weight1, axes=None) + matmul17: R.Tensor((1, seq_len, 768), dtype="float16") = R.matmul(reshape19, permute_dims17, out_dtype="void") + add25: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(matmul17, gpt_neox_layers_4_attention_dense_bias1) + layer_norm9: R.Tensor((1, seq_len, 768), dtype="float16") = R.nn.layer_norm(add23, gpt_neox_layers_4_post_attention_layernorm_weight1, gpt_neox_layers_4_post_attention_layernorm_bias1, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims18: R.Tensor((768, 3072), dtype="float16") = R.permute_dims(gpt_neox_layers_4_mlp_dense_h_to_4h_weight1, axes=None) + matmul18: R.Tensor((1, seq_len, 3072), dtype="float32") = R.matmul(layer_norm9, permute_dims18, out_dtype="float32") + add26: R.Tensor((1, seq_len, 3072), dtype="float32") = R.add(matmul18, gpt_neox_layers_4_mlp_dense_h_to_4h_bias1) + gelu4: R.Tensor((1, seq_len, 3072), dtype="float32") = R.nn.gelu(add26) + astype8: R.Tensor((1, seq_len, 3072), dtype="float16") = R.astype(gelu4, dtype="float16") + permute_dims19: R.Tensor((3072, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_4_mlp_dense_4h_to_h_weight1, axes=None) + matmul19: R.Tensor((1, seq_len, 768), dtype="float32") = R.matmul(astype8, permute_dims19, out_dtype="float32") + add27: R.Tensor((1, seq_len, 768), dtype="float32") = R.add(matmul19, gpt_neox_layers_4_mlp_dense_4h_to_h_bias1) + astype9: R.Tensor((1, seq_len, 768), dtype="float16") = R.astype(add27, dtype="float16") + add28: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(astype9, add25) + add29: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(add28, add23) + layer_norm10: R.Tensor((1, seq_len, 768), dtype="float16") = R.nn.layer_norm(add29, gpt_neox_layers_5_input_layernorm_weight1, gpt_neox_layers_5_input_layernorm_bias1, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims20: R.Tensor((768, 2304), dtype="float16") = R.permute_dims(gpt_neox_layers_5_attention_query_key_value_weight1, axes=None) + matmul20: R.Tensor((1, seq_len, 2304), dtype="float16") = R.matmul(layer_norm10, permute_dims20, out_dtype="void") + add30: R.Tensor((1, seq_len, 2304), dtype="float16") = R.add(matmul20, gpt_neox_layers_5_attention_query_key_value_bias1) + reshape20: R.Tensor((1, seq_len, 36, 64), dtype="float16") = R.reshape(add30, R.shape([1, seq_len, 36, 64])) + reshape21: R.Tensor((seq_len, 36, 64), dtype="float16") = R.reshape(reshape20, R.shape([seq_len, 36, 64])) + lv5 = R.call_dps_packed("vm.builtin.attention_kv_cache_attention_with_fused_qkv", (paged_kv_cache, R.prim_value(5), R.prim_value(T.float32(1.0)), reshape21), out_sinfo=R.Tensor((seq_len, 12, 64), dtype="float16")) + reshape22: R.Tensor((1, seq_len, 12, 64), dtype="float16") = R.reshape(lv5, R.shape([1, seq_len, 12, 64])) + reshape23: R.Tensor((1, seq_len, 768), dtype="float16") = R.reshape(reshape22, R.shape([1, seq_len, 768])) + permute_dims21: R.Tensor((768, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_5_attention_dense_weight1, axes=None) + matmul21: R.Tensor((1, seq_len, 768), dtype="float16") = R.matmul(reshape23, permute_dims21, out_dtype="void") + add31: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(matmul21, gpt_neox_layers_5_attention_dense_bias1) + layer_norm11: R.Tensor((1, seq_len, 768), dtype="float16") = R.nn.layer_norm(add29, gpt_neox_layers_5_post_attention_layernorm_weight1, gpt_neox_layers_5_post_attention_layernorm_bias1, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims22: R.Tensor((768, 3072), dtype="float16") = R.permute_dims(gpt_neox_layers_5_mlp_dense_h_to_4h_weight1, axes=None) + matmul22: R.Tensor((1, seq_len, 3072), dtype="float32") = R.matmul(layer_norm11, permute_dims22, out_dtype="float32") + add32: R.Tensor((1, seq_len, 3072), dtype="float32") = R.add(matmul22, gpt_neox_layers_5_mlp_dense_h_to_4h_bias1) + gelu5: R.Tensor((1, seq_len, 3072), dtype="float32") = R.nn.gelu(add32) + astype10: R.Tensor((1, seq_len, 3072), dtype="float16") = R.astype(gelu5, dtype="float16") + permute_dims23: R.Tensor((3072, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_5_mlp_dense_4h_to_h_weight1, axes=None) + matmul23: R.Tensor((1, seq_len, 768), dtype="float32") = R.matmul(astype10, permute_dims23, out_dtype="float32") + add33: R.Tensor((1, seq_len, 768), dtype="float32") = R.add(matmul23, gpt_neox_layers_5_mlp_dense_4h_to_h_bias1) + astype11: R.Tensor((1, seq_len, 768), dtype="float16") = R.astype(add33, dtype="float16") + add34: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(astype11, add31) + add35: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(add34, add29) + layer_norm12: R.Tensor((1, seq_len, 768), dtype="float16") = R.nn.layer_norm(add35, gpt_neox_layers_6_input_layernorm_weight1, gpt_neox_layers_6_input_layernorm_bias1, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims24: R.Tensor((768, 2304), dtype="float16") = R.permute_dims(gpt_neox_layers_6_attention_query_key_value_weight1, axes=None) + matmul24: R.Tensor((1, seq_len, 2304), dtype="float16") = R.matmul(layer_norm12, permute_dims24, out_dtype="void") + add36: R.Tensor((1, seq_len, 2304), dtype="float16") = R.add(matmul24, gpt_neox_layers_6_attention_query_key_value_bias1) + reshape24: R.Tensor((1, seq_len, 36, 64), dtype="float16") = R.reshape(add36, R.shape([1, seq_len, 36, 64])) + reshape25: R.Tensor((seq_len, 36, 64), dtype="float16") = R.reshape(reshape24, R.shape([seq_len, 36, 64])) + lv6 = R.call_dps_packed("vm.builtin.attention_kv_cache_attention_with_fused_qkv", (paged_kv_cache, R.prim_value(6), R.prim_value(T.float32(1.0)), reshape25), out_sinfo=R.Tensor((seq_len, 12, 64), dtype="float16")) + reshape26: R.Tensor((1, seq_len, 12, 64), dtype="float16") = R.reshape(lv6, R.shape([1, seq_len, 12, 64])) + reshape27: R.Tensor((1, seq_len, 768), dtype="float16") = R.reshape(reshape26, R.shape([1, seq_len, 768])) + permute_dims25: R.Tensor((768, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_6_attention_dense_weight1, axes=None) + matmul25: R.Tensor((1, seq_len, 768), dtype="float16") = R.matmul(reshape27, permute_dims25, out_dtype="void") + add37: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(matmul25, gpt_neox_layers_6_attention_dense_bias1) + layer_norm13: R.Tensor((1, seq_len, 768), dtype="float16") = R.nn.layer_norm(add35, gpt_neox_layers_6_post_attention_layernorm_weight1, gpt_neox_layers_6_post_attention_layernorm_bias1, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims26: R.Tensor((768, 3072), dtype="float16") = R.permute_dims(gpt_neox_layers_6_mlp_dense_h_to_4h_weight1, axes=None) + matmul26: R.Tensor((1, seq_len, 3072), dtype="float32") = R.matmul(layer_norm13, permute_dims26, out_dtype="float32") + add38: R.Tensor((1, seq_len, 3072), dtype="float32") = R.add(matmul26, gpt_neox_layers_6_mlp_dense_h_to_4h_bias1) + gelu6: R.Tensor((1, seq_len, 3072), dtype="float32") = R.nn.gelu(add38) + astype12: R.Tensor((1, seq_len, 3072), dtype="float16") = R.astype(gelu6, dtype="float16") + permute_dims27: R.Tensor((3072, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_6_mlp_dense_4h_to_h_weight1, axes=None) + matmul27: R.Tensor((1, seq_len, 768), dtype="float32") = R.matmul(astype12, permute_dims27, out_dtype="float32") + add39: R.Tensor((1, seq_len, 768), dtype="float32") = R.add(matmul27, gpt_neox_layers_6_mlp_dense_4h_to_h_bias1) + astype13: R.Tensor((1, seq_len, 768), dtype="float16") = R.astype(add39, dtype="float16") + add40: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(astype13, add37) + add41: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(add40, add35) + layer_norm14: R.Tensor((1, seq_len, 768), dtype="float16") = R.nn.layer_norm(add41, gpt_neox_layers_7_input_layernorm_weight1, gpt_neox_layers_7_input_layernorm_bias1, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims28: R.Tensor((768, 2304), dtype="float16") = R.permute_dims(gpt_neox_layers_7_attention_query_key_value_weight1, axes=None) + matmul28: R.Tensor((1, seq_len, 2304), dtype="float16") = R.matmul(layer_norm14, permute_dims28, out_dtype="void") + add42: R.Tensor((1, seq_len, 2304), dtype="float16") = R.add(matmul28, gpt_neox_layers_7_attention_query_key_value_bias1) + reshape28: R.Tensor((1, seq_len, 36, 64), dtype="float16") = R.reshape(add42, R.shape([1, seq_len, 36, 64])) + reshape29: R.Tensor((seq_len, 36, 64), dtype="float16") = R.reshape(reshape28, R.shape([seq_len, 36, 64])) + lv7 = R.call_dps_packed("vm.builtin.attention_kv_cache_attention_with_fused_qkv", (paged_kv_cache, R.prim_value(7), R.prim_value(T.float32(1.0)), reshape29), out_sinfo=R.Tensor((seq_len, 12, 64), dtype="float16")) + reshape30: R.Tensor((1, seq_len, 12, 64), dtype="float16") = R.reshape(lv7, R.shape([1, seq_len, 12, 64])) + reshape31: R.Tensor((1, seq_len, 768), dtype="float16") = R.reshape(reshape30, R.shape([1, seq_len, 768])) + permute_dims29: R.Tensor((768, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_7_attention_dense_weight1, axes=None) + matmul29: R.Tensor((1, seq_len, 768), dtype="float16") = R.matmul(reshape31, permute_dims29, out_dtype="void") + add43: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(matmul29, gpt_neox_layers_7_attention_dense_bias1) + layer_norm15: R.Tensor((1, seq_len, 768), dtype="float16") = R.nn.layer_norm(add41, gpt_neox_layers_7_post_attention_layernorm_weight1, gpt_neox_layers_7_post_attention_layernorm_bias1, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims30: R.Tensor((768, 3072), dtype="float16") = R.permute_dims(gpt_neox_layers_7_mlp_dense_h_to_4h_weight1, axes=None) + matmul30: R.Tensor((1, seq_len, 3072), dtype="float32") = R.matmul(layer_norm15, permute_dims30, out_dtype="float32") + add44: R.Tensor((1, seq_len, 3072), dtype="float32") = R.add(matmul30, gpt_neox_layers_7_mlp_dense_h_to_4h_bias1) + gelu7: R.Tensor((1, seq_len, 3072), dtype="float32") = R.nn.gelu(add44) + astype14: R.Tensor((1, seq_len, 3072), dtype="float16") = R.astype(gelu7, dtype="float16") + permute_dims31: R.Tensor((3072, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_7_mlp_dense_4h_to_h_weight1, axes=None) + matmul31: R.Tensor((1, seq_len, 768), dtype="float32") = R.matmul(astype14, permute_dims31, out_dtype="float32") + add45: R.Tensor((1, seq_len, 768), dtype="float32") = R.add(matmul31, gpt_neox_layers_7_mlp_dense_4h_to_h_bias1) + astype15: R.Tensor((1, seq_len, 768), dtype="float16") = R.astype(add45, dtype="float16") + add46: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(astype15, add43) + add47: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(add46, add41) + layer_norm16: R.Tensor((1, seq_len, 768), dtype="float16") = R.nn.layer_norm(add47, gpt_neox_layers_8_input_layernorm_weight1, gpt_neox_layers_8_input_layernorm_bias1, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims32: R.Tensor((768, 2304), dtype="float16") = R.permute_dims(gpt_neox_layers_8_attention_query_key_value_weight1, axes=None) + matmul32: R.Tensor((1, seq_len, 2304), dtype="float16") = R.matmul(layer_norm16, permute_dims32, out_dtype="void") + add48: R.Tensor((1, seq_len, 2304), dtype="float16") = R.add(matmul32, gpt_neox_layers_8_attention_query_key_value_bias1) + reshape32: R.Tensor((1, seq_len, 36, 64), dtype="float16") = R.reshape(add48, R.shape([1, seq_len, 36, 64])) + reshape33: R.Tensor((seq_len, 36, 64), dtype="float16") = R.reshape(reshape32, R.shape([seq_len, 36, 64])) + lv8 = R.call_dps_packed("vm.builtin.attention_kv_cache_attention_with_fused_qkv", (paged_kv_cache, R.prim_value(8), R.prim_value(T.float32(1.0)), reshape33), out_sinfo=R.Tensor((seq_len, 12, 64), dtype="float16")) + reshape34: R.Tensor((1, seq_len, 12, 64), dtype="float16") = R.reshape(lv8, R.shape([1, seq_len, 12, 64])) + reshape35: R.Tensor((1, seq_len, 768), dtype="float16") = R.reshape(reshape34, R.shape([1, seq_len, 768])) + permute_dims33: R.Tensor((768, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_8_attention_dense_weight1, axes=None) + matmul33: R.Tensor((1, seq_len, 768), dtype="float16") = R.matmul(reshape35, permute_dims33, out_dtype="void") + add49: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(matmul33, gpt_neox_layers_8_attention_dense_bias1) + layer_norm17: R.Tensor((1, seq_len, 768), dtype="float16") = R.nn.layer_norm(add47, gpt_neox_layers_8_post_attention_layernorm_weight1, gpt_neox_layers_8_post_attention_layernorm_bias1, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims34: R.Tensor((768, 3072), dtype="float16") = R.permute_dims(gpt_neox_layers_8_mlp_dense_h_to_4h_weight1, axes=None) + matmul34: R.Tensor((1, seq_len, 3072), dtype="float32") = R.matmul(layer_norm17, permute_dims34, out_dtype="float32") + add50: R.Tensor((1, seq_len, 3072), dtype="float32") = R.add(matmul34, gpt_neox_layers_8_mlp_dense_h_to_4h_bias1) + gelu8: R.Tensor((1, seq_len, 3072), dtype="float32") = R.nn.gelu(add50) + astype16: R.Tensor((1, seq_len, 3072), dtype="float16") = R.astype(gelu8, dtype="float16") + permute_dims35: R.Tensor((3072, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_8_mlp_dense_4h_to_h_weight1, axes=None) + matmul35: R.Tensor((1, seq_len, 768), dtype="float32") = R.matmul(astype16, permute_dims35, out_dtype="float32") + add51: R.Tensor((1, seq_len, 768), dtype="float32") = R.add(matmul35, gpt_neox_layers_8_mlp_dense_4h_to_h_bias1) + astype17: R.Tensor((1, seq_len, 768), dtype="float16") = R.astype(add51, dtype="float16") + add52: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(astype17, add49) + add53: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(add52, add47) + layer_norm18: R.Tensor((1, seq_len, 768), dtype="float16") = R.nn.layer_norm(add53, gpt_neox_layers_9_input_layernorm_weight1, gpt_neox_layers_9_input_layernorm_bias1, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims36: R.Tensor((768, 2304), dtype="float16") = R.permute_dims(gpt_neox_layers_9_attention_query_key_value_weight1, axes=None) + matmul36: R.Tensor((1, seq_len, 2304), dtype="float16") = R.matmul(layer_norm18, permute_dims36, out_dtype="void") + add54: R.Tensor((1, seq_len, 2304), dtype="float16") = R.add(matmul36, gpt_neox_layers_9_attention_query_key_value_bias1) + reshape36: R.Tensor((1, seq_len, 36, 64), dtype="float16") = R.reshape(add54, R.shape([1, seq_len, 36, 64])) + reshape37: R.Tensor((seq_len, 36, 64), dtype="float16") = R.reshape(reshape36, R.shape([seq_len, 36, 64])) + lv9 = R.call_dps_packed("vm.builtin.attention_kv_cache_attention_with_fused_qkv", (paged_kv_cache, R.prim_value(9), R.prim_value(T.float32(1.0)), reshape37), out_sinfo=R.Tensor((seq_len, 12, 64), dtype="float16")) + reshape38: R.Tensor((1, seq_len, 12, 64), dtype="float16") = R.reshape(lv9, R.shape([1, seq_len, 12, 64])) + reshape39: R.Tensor((1, seq_len, 768), dtype="float16") = R.reshape(reshape38, R.shape([1, seq_len, 768])) + permute_dims37: R.Tensor((768, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_9_attention_dense_weight1, axes=None) + matmul37: R.Tensor((1, seq_len, 768), dtype="float16") = R.matmul(reshape39, permute_dims37, out_dtype="void") + add55: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(matmul37, gpt_neox_layers_9_attention_dense_bias1) + layer_norm19: R.Tensor((1, seq_len, 768), dtype="float16") = R.nn.layer_norm(add53, gpt_neox_layers_9_post_attention_layernorm_weight1, gpt_neox_layers_9_post_attention_layernorm_bias1, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims38: R.Tensor((768, 3072), dtype="float16") = R.permute_dims(gpt_neox_layers_9_mlp_dense_h_to_4h_weight1, axes=None) + matmul38: R.Tensor((1, seq_len, 3072), dtype="float32") = R.matmul(layer_norm19, permute_dims38, out_dtype="float32") + add56: R.Tensor((1, seq_len, 3072), dtype="float32") = R.add(matmul38, gpt_neox_layers_9_mlp_dense_h_to_4h_bias1) + gelu9: R.Tensor((1, seq_len, 3072), dtype="float32") = R.nn.gelu(add56) + astype18: R.Tensor((1, seq_len, 3072), dtype="float16") = R.astype(gelu9, dtype="float16") + permute_dims39: R.Tensor((3072, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_9_mlp_dense_4h_to_h_weight1, axes=None) + matmul39: R.Tensor((1, seq_len, 768), dtype="float32") = R.matmul(astype18, permute_dims39, out_dtype="float32") + add57: R.Tensor((1, seq_len, 768), dtype="float32") = R.add(matmul39, gpt_neox_layers_9_mlp_dense_4h_to_h_bias1) + astype19: R.Tensor((1, seq_len, 768), dtype="float16") = R.astype(add57, dtype="float16") + add58: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(astype19, add55) + add59: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(add58, add53) + layer_norm20: R.Tensor((1, seq_len, 768), dtype="float16") = R.nn.layer_norm(add59, gpt_neox_layers_10_input_layernorm_weight1, gpt_neox_layers_10_input_layernorm_bias1, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims40: R.Tensor((768, 2304), dtype="float16") = R.permute_dims(gpt_neox_layers_10_attention_query_key_value_weight1, axes=None) + matmul40: R.Tensor((1, seq_len, 2304), dtype="float16") = R.matmul(layer_norm20, permute_dims40, out_dtype="void") + add60: R.Tensor((1, seq_len, 2304), dtype="float16") = R.add(matmul40, gpt_neox_layers_10_attention_query_key_value_bias1) + reshape40: R.Tensor((1, seq_len, 36, 64), dtype="float16") = R.reshape(add60, R.shape([1, seq_len, 36, 64])) + reshape41: R.Tensor((seq_len, 36, 64), dtype="float16") = R.reshape(reshape40, R.shape([seq_len, 36, 64])) + lv10 = R.call_dps_packed("vm.builtin.attention_kv_cache_attention_with_fused_qkv", (paged_kv_cache, R.prim_value(10), R.prim_value(T.float32(1.0)), reshape41), out_sinfo=R.Tensor((seq_len, 12, 64), dtype="float16")) + reshape42: R.Tensor((1, seq_len, 12, 64), dtype="float16") = R.reshape(lv10, R.shape([1, seq_len, 12, 64])) + reshape43: R.Tensor((1, seq_len, 768), dtype="float16") = R.reshape(reshape42, R.shape([1, seq_len, 768])) + permute_dims41: R.Tensor((768, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_10_attention_dense_weight1, axes=None) + matmul41: R.Tensor((1, seq_len, 768), dtype="float16") = R.matmul(reshape43, permute_dims41, out_dtype="void") + add61: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(matmul41, gpt_neox_layers_10_attention_dense_bias1) + layer_norm21: R.Tensor((1, seq_len, 768), dtype="float16") = R.nn.layer_norm(add59, gpt_neox_layers_10_post_attention_layernorm_weight1, gpt_neox_layers_10_post_attention_layernorm_bias1, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims42: R.Tensor((768, 3072), dtype="float16") = R.permute_dims(gpt_neox_layers_10_mlp_dense_h_to_4h_weight1, axes=None) + matmul42: R.Tensor((1, seq_len, 3072), dtype="float32") = R.matmul(layer_norm21, permute_dims42, out_dtype="float32") + add62: R.Tensor((1, seq_len, 3072), dtype="float32") = R.add(matmul42, gpt_neox_layers_10_mlp_dense_h_to_4h_bias1) + gelu10: R.Tensor((1, seq_len, 3072), dtype="float32") = R.nn.gelu(add62) + astype20: R.Tensor((1, seq_len, 3072), dtype="float16") = R.astype(gelu10, dtype="float16") + permute_dims43: R.Tensor((3072, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_10_mlp_dense_4h_to_h_weight1, axes=None) + matmul43: R.Tensor((1, seq_len, 768), dtype="float32") = R.matmul(astype20, permute_dims43, out_dtype="float32") + add63: R.Tensor((1, seq_len, 768), dtype="float32") = R.add(matmul43, gpt_neox_layers_10_mlp_dense_4h_to_h_bias1) + astype21: R.Tensor((1, seq_len, 768), dtype="float16") = R.astype(add63, dtype="float16") + add64: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(astype21, add61) + add65: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(add64, add59) + layer_norm22: R.Tensor((1, seq_len, 768), dtype="float16") = R.nn.layer_norm(add65, gpt_neox_layers_11_input_layernorm_weight1, gpt_neox_layers_11_input_layernorm_bias1, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims44: R.Tensor((768, 2304), dtype="float16") = R.permute_dims(gpt_neox_layers_11_attention_query_key_value_weight1, axes=None) + matmul44: R.Tensor((1, seq_len, 2304), dtype="float16") = R.matmul(layer_norm22, permute_dims44, out_dtype="void") + add66: R.Tensor((1, seq_len, 2304), dtype="float16") = R.add(matmul44, gpt_neox_layers_11_attention_query_key_value_bias1) + reshape44: R.Tensor((1, seq_len, 36, 64), dtype="float16") = R.reshape(add66, R.shape([1, seq_len, 36, 64])) + reshape45: R.Tensor((seq_len, 36, 64), dtype="float16") = R.reshape(reshape44, R.shape([seq_len, 36, 64])) + lv11 = R.call_dps_packed("vm.builtin.attention_kv_cache_attention_with_fused_qkv", (paged_kv_cache, R.prim_value(11), R.prim_value(T.float32(1.0)), reshape45), out_sinfo=R.Tensor((seq_len, 12, 64), dtype="float16")) + reshape46: R.Tensor((1, seq_len, 12, 64), dtype="float16") = R.reshape(lv11, R.shape([1, seq_len, 12, 64])) + reshape47: R.Tensor((1, seq_len, 768), dtype="float16") = R.reshape(reshape46, R.shape([1, seq_len, 768])) + permute_dims45: R.Tensor((768, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_11_attention_dense_weight1, axes=None) + matmul45: R.Tensor((1, seq_len, 768), dtype="float16") = R.matmul(reshape47, permute_dims45, out_dtype="void") + add67: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(matmul45, gpt_neox_layers_11_attention_dense_bias1) + layer_norm23: R.Tensor((1, seq_len, 768), dtype="float16") = R.nn.layer_norm(add65, gpt_neox_layers_11_post_attention_layernorm_weight1, gpt_neox_layers_11_post_attention_layernorm_bias1, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + permute_dims46: R.Tensor((768, 3072), dtype="float16") = R.permute_dims(gpt_neox_layers_11_mlp_dense_h_to_4h_weight1, axes=None) + matmul46: R.Tensor((1, seq_len, 3072), dtype="float32") = R.matmul(layer_norm23, permute_dims46, out_dtype="float32") + add68: R.Tensor((1, seq_len, 3072), dtype="float32") = R.add(matmul46, gpt_neox_layers_11_mlp_dense_h_to_4h_bias1) + gelu11: R.Tensor((1, seq_len, 3072), dtype="float32") = R.nn.gelu(add68) + astype22: R.Tensor((1, seq_len, 3072), dtype="float16") = R.astype(gelu11, dtype="float16") + permute_dims47: R.Tensor((3072, 768), dtype="float16") = R.permute_dims(gpt_neox_layers_11_mlp_dense_4h_to_h_weight1, axes=None) + matmul47: R.Tensor((1, seq_len, 768), dtype="float32") = R.matmul(astype22, permute_dims47, out_dtype="float32") + add69: R.Tensor((1, seq_len, 768), dtype="float32") = R.add(matmul47, gpt_neox_layers_11_mlp_dense_4h_to_h_bias1) + astype23: R.Tensor((1, seq_len, 768), dtype="float16") = R.astype(add69, dtype="float16") + add70: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(astype23, add67) + add71: R.Tensor((1, seq_len, 768), dtype="float16") = R.add(add70, add65) + layer_norm24: R.Tensor((1, seq_len, 768), dtype="float16") = R.nn.layer_norm(add71, gpt_neox_final_layer_norm_weight1, gpt_neox_final_layer_norm_bias1, axes=[-1], epsilon=1.0000000000000001e-05, center=True, scale=True) + lv12 = R.call_tir(cls.index, (layer_norm24,), out_sinfo=R.Tensor((1, 1, 768), dtype="float16")) + permute_dims48: R.Tensor((768, vocab_size), dtype="float16") = R.permute_dims(embed_out_weight1, axes=None) + matmul48: R.Tensor((1, 1, vocab_size), dtype="float16") = R.matmul(lv12, permute_dims48, out_dtype="void") + astype24: R.Tensor((1, 1, vocab_size), dtype="float32") = R.astype(matmul48, dtype="float32") + gv1: R.Tuple(R.Tensor((1, 1, vocab_size), dtype="float32"), R.Object) = astype24, paged_kv_cache + R.output(gv1) + return gv1 + + @R.function + def softmax_with_temperature(logits: R.Tensor(("batch_size", 1, "vocab_size"), dtype="float32"), temperature: R.Tensor(("batch_size",), dtype="float32")) -> R.Tensor(("batch_size", 1, "vocab_size"), dtype="float32"): + batch_size = T.int64(is_size_var=True) + vocab_size = T.int64(is_size_var=True) + R.func_attr({"relax.memory_plan_dynamic_func_output": True, "tir_non_negative_var": ["vocab_size"], "tir_var_upper_bound": {"batch_size": 128, "seq_len": 2048, "total_seq_len": 2048}}) + cls = Module + with R.dataflow(): + lv: R.Tensor((batch_size, vocab_size), dtype="float32") = R.call_pure_packed("vm.builtin.reshape", logits, R.shape([batch_size, vocab_size]), sinfo_args=(R.Tensor((batch_size, vocab_size), dtype="float32"),)) + lv1 = R.call_tir(cls.chunk_lse, (lv, temperature), out_sinfo=[R.Tensor((batch_size, (vocab_size + 4096 - 1) // 4096), dtype="float32"), R.Tensor((batch_size, (vocab_size + 4096 - 1) // 4096), dtype="float32")]) + lv2: R.Tensor((batch_size, (vocab_size + 4096 - 1) // 4096), dtype="float32") = lv1[0] + lv3: R.Tensor((batch_size, (vocab_size + 4096 - 1) // 4096), dtype="float32") = lv1[1] + lv4 = R.call_tir(cls.softmax_with_chunked_sum, (lv, temperature, lv2, lv3), out_sinfo=R.Tensor((batch_size, vocab_size), dtype="float32")) + gv: R.Tensor((batch_size, 1, vocab_size), dtype="float32") = R.call_pure_packed("vm.builtin.reshape", lv4, R.shape([batch_size, 1, vocab_size]), sinfo_args=(R.Tensor((batch_size, 1, vocab_size), dtype="float32"),)) + R.output(gv) + return gv \ No newline at end of file