repo stringclasses 454
values | file_path stringlengths 5 201 | extension stringclasses 1
value | content stringlengths 8 509k | num_lines int64 3 16.9k | size_bytes int64 8 511k |
|---|---|---|---|---|---|
sglang | python/sglang/kernels/aot/tests/speculative/test_eagle_utils.py | .py | import sys
import pytest
import torch
import torch.nn.functional as F
from sgl_kernel import verify_tree_greedy
def test_verify_tree_greedy():
candidates = torch.tensor(
[
[0, 1, 2, 3, 4, 5],
[7, 8, 9, 10, 11, 12],
],
dtype=torch.int64,
device="cuda",
)... | 90 | 2,410 |
sglang | python/sglang/kernels/aot/tests/speculative/test_ngram_utils.py | .py | import sys
import pytest
import torch
import torch.nn.functional as F
from sgl_kernel import reconstruct_indices_from_tree_mask
def test_reconstruct_indices_from_tree_mask():
bs = 1
num_branch_token = 4
seq_lens = torch.tensor([12], device="cuda", dtype=torch.int64)
retrive_index = torch.full(
... | 79 | 1,968 |
sglang | python/sglang/kernels/aot/tests/speculative/test_speculative_sampling.py | .py | import sys
import pytest
import torch
import torch.nn.functional as F
from sgl_kernel import tree_speculative_sampling_target_only
test_cases = [
(
1,
1,
[3, -1, -1, 4, 5, 18, 11, -1, -1, -1, 12, 18],
[[0, 3, 4, 5], [6, 10, 11, -1]],
[3, 2],
),
(
0, # thres... | 132 | 4,085 |
sglang | python/sglang/kernels/aot/tests/spatial/test_greenctx_stream.py | .py | import sys
import pytest
import torch
import torch.nn.functional as F
from sgl_kernel import create_greenctx_stream_by_value, get_sm_available
def test_green_ctx():
A = torch.randn(5120, 5120).cuda()
B = torch.randn(5120, 5120).cuda()
C = torch.matmul(A, B)
sm_counts = get_sm_available(0)
stream_... | 28 | 801 |
sglang | python/sglang/kernels/aot/benchmark/bench_sum_scale.py | .py | import os
import torch
import triton
import triton.language as tl
from sgl_kernel import moe_sum_reduce as moe_sum_reduce_cuda
from triton.testing import do_bench
from sglang.utils import is_in_ci
IS_CI = is_in_ci()
@triton.jit
def _moe_sum_reduce_kernel(
input_ptr,
input_stride_0,
input_stride_1,
... | 250 | 7,904 |
sglang | python/sglang/kernels/aot/benchmark/bench_dsv4_norm_rope.py | .py | """Benchmark for DeepSeek-V4 fused norm + RoPE kernels."""
import itertools
import sgl_kernel
import torch
import triton
import triton.testing
try:
from sglang.utils import is_in_ci
IS_CI = is_in_ci()
except ImportError:
IS_CI = False
batch_sizes = [1] if IS_CI else [1, 4, 16, 64, 256]
num_heads_list =... | 76 | 2,153 |
sglang | python/sglang/kernels/aot/benchmark/bench_fp4_gemm.py | .py | import argparse
import csv
import logging
from functools import partial
from typing import List, Tuple
import torch
import triton
from flashinfer import fp4_quantize, mm_fp4
from flashinfer.autotuner import autotune
from flashinfer.jit.core import logger as flashinfer_logger
from flashinfer.testing import bench_gpu_ti... | 463 | 13,132 |
sglang | python/sglang/kernels/aot/benchmark/bench_moe_topk_softmax.py | .py | import itertools
import os
import pytest
import torch
import triton
from sgl_kernel import topk_softmax
from sglang.utils import is_in_ci
# Optional vLLM import
try:
from vllm import _custom_ops as vllm_custom_ops
VLLM_AVAILABLE = True
except ImportError:
vllm_custom_ops = None
VLLM_AVAILABLE = Fals... | 221 | 6,463 |
sglang | python/sglang/kernels/aot/benchmark/bench_fp8_gemm_swap_ab.py | .py | """Targeted benchmark for the SM90 FP8 swap-AB dispatch path.
Sweeps small batch sizes (M = 1..128) across N/K shapes that exercise each
dispatch bucket in `fp8_gemm_sm90_dispatch.cuh`. Output style matches
`bench_fp8_gemm.py`: `triton.testing.perf_report` + GB/s table per (N, K).
Compare against `main` by:
1. Run ... | 133 | 4,133 |
sglang | python/sglang/kernels/aot/benchmark/bench_per_tensor_quant_fp8.py | .py | import itertools
import math
import os
from typing import Any, Dict, List, Optional, Tuple
import numpy as np
import torch
import triton
import triton.testing
from sglang.kernels.ops.quantization.per_tensor_quant_fp8 import (
per_tensor_quant_fp8,
)
from sglang.utils import is_in_ci
# Optional imports
try:
f... | 137 | 3,816 |
sglang | python/sglang/kernels/aot/benchmark/bench_int8_gemm.py | .py | import argparse
import copy
import itertools
import os
import torch
import triton
from sgl_kernel import int8_scaled_mm
from sglang.utils import is_in_ci
# Optional vLLM import
try:
from vllm._custom_ops import cutlass_scaled_mm as vllm_scaled_mm
VLLM_AVAILABLE = True
except ImportError:
vllm_scaled_mm ... | 184 | 5,131 |
sglang | python/sglang/kernels/aot/benchmark/bench_amd_deterministic_allreduce.py | .py | """
Benchmark latency comparison between different all-reduce implementations.
Compares:
- NCCL all-reduce (may be non-deterministic)
- Reduce-scatter + all-gather (RS+AG, deterministic but slower)
- Deterministic 1-stage kernel (forces fixed accumulation order, deterministic)
Note: The "deterministic kernel" is NOT ... | 690 | 29,516 |
sglang | python/sglang/kernels/aot/benchmark/bench_moe_topk_sigmoid.py | .py | import itertools
import os
import pytest
import torch
import triton
from sgl_kernel import topk_sigmoid
from sglang.utils import is_in_ci
# Optional MUSA import
try:
from sglang.srt.utils import is_musa
if is_musa():
from sglang.srt.hardware_backend.musa.kernels.topk import (
topk_sigmoi... | 243 | 6,938 |
sglang | python/sglang/kernels/aot/benchmark/bench_per_token_group_quant_8bit.py | .py | import itertools
import os
import torch
import triton
from sgl_kernel.test_utils import create_per_token_group_quant_test_data
from sglang.kernels.ops.quantization.fp8_kernel import (
per_token_group_quant_8bit as triton_per_token_group_quant_8bit,
)
from sglang.kernels.ops.quantization.fp8_kernel import (
sg... | 242 | 7,277 |
sglang | python/sglang/kernels/aot/benchmark/bench_moe_ep_post_reorder.py | .py | import torch
import triton
from sglang.kernels.ops.moe.ep_moe_kernels import post_reorder_triton_kernel
from sglang.utils import is_in_ci
IS_CI = is_in_ci()
# CI environment uses simplified parameters
if IS_CI:
batch_sizes = [64, 128] # Only test 2 values in CI
else:
batch_sizes = [64, 128, 256, 512, 640, 7... | 86 | 2,481 |
sglang | python/sglang/kernels/aot/benchmark/bench_activation.py | .py | # Benchmarks SGLang kernels versus vLLM across
# (kernel, dtype, batch_size, seq_len, dim) and prints speed-up.
import argparse
import itertools
import os
import re
from typing import List, Tuple
import sgl_kernel
import torch
import torch.nn.functional as F
import triton
import triton.testing
from sgl_kernel import g... | 210 | 6,890 |
sglang | python/sglang/kernels/aot/benchmark/bench_fp8_blockwise_group_gemm.py | .py | import argparse
import random
from dataclasses import dataclass
from typing import List, Tuple
import deep_gemm
import torch
from sgl_kernel import fp8_blockwise_scaled_grouped_mm
from sglang.utils import is_in_ci
IS_CI = is_in_ci()
def get_m_alignment_for_contiguous_layout():
return 128
def ceil_div(x: int,... | 341 | 11,549 |
sglang | python/sglang/kernels/aot/benchmark/bench_es_fp8_blockwise_grouped_gemm.py | .py | import argparse
import random
from dataclasses import dataclass
from typing import List, Tuple
import numpy as np
import torch
from sgl_kernel import (
es_fp8_blockwise_scaled_grouped_mm,
fp8_blockwise_scaled_grouped_mm,
)
random.seed(28)
def ceil_div(x: int, y: int) -> int:
return (x + y - 1) // y
de... | 371 | 11,631 |
sglang | python/sglang/kernels/aot/benchmark/bench_rotary_embedding.py | .py | import itertools
import os
import torch
import triton
from sgl_kernel.testing.rotary_embedding import (
FlashInferRotaryEmbedding,
FusedSetKVBufferArg,
MHATokenToKVPool,
RotaryEmbedding,
create_inputs,
)
from sglang.srt.utils.bench_utils import bench_kineto
from sglang.utils import is_in_ci
IS_CI... | 109 | 2,800 |
sglang | python/sglang/kernels/aot/benchmark/bench_awq_dequant.py | .py | import itertools
import os
from typing import List, Tuple
import torch
import triton
import triton.testing
from sgl_kernel import awq_dequantize
from sglang.utils import is_in_ci
# Optional vLLM import
try:
from vllm import _custom_ops as ops
VLLM_AVAILABLE = True
except ImportError:
ops = None
VLLM... | 152 | 4,432 |
sglang | python/sglang/kernels/aot/benchmark/bench_mrope.py | .py | # Adapted from vLLM benchmark_mrope.py
# This script benchmarks the mrope kernel (mainly for Qwen2VL and Qwen2.5VL models).
# It generates test data, runs benchmarks, and saves results to a CSV file.
#
# The CSV file (named with current date/time) contains these columns:
# model_name, tp_size, num_tokens, num_heads, n... | 251 | 8,031 |
sglang | python/sglang/kernels/aot/benchmark/bench_top_k_top_p_sampling.py | .py | import itertools
import os
import flashinfer.sampling
import sgl_kernel
import torch
import triton
import triton.testing
from sglang.utils import is_in_ci
IS_CI = is_in_ci()
def torch_top_k_top_p_joint_sampling_from_probs(
normalized_prob, top_k, top_p, eps=1e-4
):
"""Reference PyTorch implementation of jo... | 147 | 4,667 |
sglang | python/sglang/kernels/aot/benchmark/bench_cutlass_mla.py | .py | import argparse
import copy
import itertools
import os
import torch
import triton
from sgl_kernel import cutlass_mla_decode, cutlass_mla_get_workspace_size
from sglang.srt.utils import get_device_capability
from sglang.utils import is_in_ci
IS_CI = is_in_ci()
# CI environment uses simplified parameters
if IS_CI:
... | 176 | 5,184 |
sglang | python/sglang/kernels/aot/benchmark/bench_rmsnorm.py | .py | # Benchmarks SGLang RMSNorm kernels versus vLLM and FlashInfer across
# (batch_size, seq_len, hidden_size) and prints speed-up.
import argparse
import itertools
import os
import re
from typing import List, Optional, Tuple, Union
import sgl_kernel
import torch
import torch.nn as nn
import triton
import triton.testing
f... | 396 | 12,192 |
sglang | python/sglang/kernels/aot/benchmark/bench_fp8_gemm.py | .py | import argparse
import copy
import itertools
import os
from typing import Optional, Tuple
import torch
import triton
from sgl_kernel import fp8_scaled_mm as sgl_scaled_mm
from sglang.kernels.ops.quantization.per_tensor_quant_fp8 import per_tensor_quant_fp8
from sglang.utils import is_in_ci
# Optional vLLM import
try... | 247 | 7,255 |
sglang | python/sglang/kernels/aot/benchmark/bench_moe_align_block_size.py | .py | import argparse
import itertools
import os
import torch
import triton
import triton.language as tl
from sgl_kernel import moe_align_block_size as sgl_moe_align_block_size
from sglang.utils import is_in_ci
try:
from vllm import _custom_ops as ops
VLLM_AVAILABLE = True
except ImportError:
ops = None
V... | 444 | 13,708 |
sglang | python/sglang/kernels/ops/__init__.py | .py | """Public operator groups for the ``sglang.kernels`` namespace.
Each submodule corresponds to one operator group from RFC #29630. Importing a
group registers its :class:`~sglang.kernels.spec.KernelSpec` metadata and
exposes thin, lazily-dispatched wrapper callables.
Importing this package eagerly imports every group ... | 46 | 1,149 |
sglang | python/sglang/kernels/ops/speculative/ngram_embedding.py | .py | from __future__ import annotations
from typing import TYPE_CHECKING
from sglang.kernel_api_logging import debug_kernel_api
from sglang.kernels.jit.utils import cache_once, load_jit
if TYPE_CHECKING:
import torch
from tvm_ffi.module import Module
@cache_once
def _jit_ngram_embedding_module() -> Module:
... | 166 | 4,642 |
sglang | python/sglang/kernels/ops/speculative/fused_kv_materialize.py | .py | # Copyright 2023-2024 SGLang Team
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writin... | 458 | 17,056 |
sglang | python/sglang/kernels/ops/speculative/eagle.py | .py | import torch
import triton
import triton.language as tl
from sglang.srt.utils import is_cpu, next_power_of_2
_is_cpu = is_cpu()
if _is_cpu:
from sgl_kernel import fill_accept_out_cache_loc_cpu, fill_bonus_tokens_cpu
@triton.jit
def fill_bonus_tokens(
accept_tokens,
accept_lens,
bonus_tokens_ptr,
... | 92 | 2,351 |
sglang | python/sglang/kernels/ops/speculative/__init__.py | .py | """Speculative-decoding kernels (Triton).
The Triton kernels migrated here live in this package
(``sglang.kernels.ops.speculative.<module>``); import them from there. Their
``KernelSpec`` metadata is registered below for inventory (backend = Triton).
"""
from sglang.kernels.registry import register_kernel
from sglang... | 37 | 1,331 |
sglang | python/sglang/kernels/ops/speculative/ngram_corpus.py | .py | from __future__ import annotations
from collections.abc import Iterable, Sequence
from typing import Dict, List, Tuple
import numpy as np
import torch
import tvm_ffi
from sglang.kernels.jit.utils import cache_once, load_jit
_MATCH_TYPE_MAP = {"BFS": 0, "PROB": 1}
def _to_csr(batch_tokens: List[List[int]]) -> Tupl... | 141 | 5,026 |
sglang | python/sglang/kernels/ops/speculative/ragged_verify_kernels.py | .py | from __future__ import annotations
import msgspec
import torch
import triton
import triton.language as tl
class PaddedToBucket:
@classmethod
def execute(
cls,
*,
verify_lens: torch.Tensor,
graph_num_tokens: int,
bs: int,
padded_bs: int,
) -> torch.Tensor:
... | 200 | 5,703 |
sglang | python/sglang/kernels/ops/speculative/multi_layer_eagle.py | .py | # Copyright 2023-2024 SGLang Team
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writin... | 797 | 26,944 |
sglang | python/sglang/kernels/ops/speculative/dflash.py | .py | import torch
import triton
import triton.language as tl
@triton.jit
def _dflash_accept_bonus_contig_kernel(
candidates_ptr,
target_top1_ptr,
accept_lens_out_ptr,
commit_lens_out_ptr,
bonus_ids_out_ptr,
out_tokens_ptr,
prefix_lens_ptr,
new_seq_lens_out_ptr,
candidates_row_stride,
... | 247 | 8,042 |
sglang | python/sglang/kernels/ops/speculative/cache_locs.py | .py | from __future__ import annotations
import torch
import triton
import triton.language as tl
from sglang.srt.utils import (
is_cpu,
is_cuda,
is_hip,
is_musa,
is_npu,
is_xpu,
next_power_of_2,
)
_is_cpu = is_cpu()
_is_cuda = is_cuda()
_is_hip = is_hip()
_is_npu = is_npu()
_is_musa = is_musa()... | 491 | 15,261 |
sglang | python/sglang/kernels/ops/speculative/reject_sampling.py | .py | import triton
import triton.language as tl
@triton.jit
def speculative_sampling_classic_kernel(
# Pointers
Predicts,
AcceptIndex,
AcceptTokenNum,
Candidates,
RetriveIndex,
UniformSamples,
UniformSamplesFinal,
TargetProbs,
DraftProbs,
# Strides
stride_cand_b,
stride_... | 209 | 6,349 |
sglang | python/sglang/kernels/ops/speculative/gather_spec_extras.py | .py | from __future__ import annotations
from typing import Optional
import torch
import triton
import triton.language as tl
@triton.jit
def _gather_rows_kernel(
idx_ptr,
s0,
d0,
n0,
s1,
d1,
n1,
s2,
d2,
n2,
s3,
d3,
n3,
HAS3: tl.constexpr,
BLOCK: tl.constexpr,
):... | 118 | 3,790 |
sglang | python/sglang/kernels/ops/speculative/topk1.py | .py | from __future__ import annotations
import torch
import triton
import triton.language as tl
_DRAFT_TOPK1_BLOCK = 8192
@triton.jit
def _draft_topk1_partial_argmax_kernel(
logits,
partial_vals,
partial_indices,
logits_row_stride,
vocab_size: tl.constexpr,
num_splits: tl.constexpr,
BLOCK: tl... | 151 | 4,881 |
sglang | python/sglang/kernels/ops/speculative/spec_tree.py | .py | # Copyright 2023-2026 SGLang Team
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writin... | 282 | 10,723 |
sglang | python/sglang/kernels/ops/speculative/dspark/dspark_schedule.py | .py | from __future__ import annotations
from typing import TYPE_CHECKING
import torch
import triton
import triton.language as tl
from sglang.kernels.ops.speculative.dspark.dispatch import (
inputs_on_cuda,
)
if TYPE_CHECKING:
from sglang.srt.speculative.dspark_components.dspark_planner import (
DSparkSch... | 261 | 7,836 |
sglang | python/sglang/kernels/ops/speculative/dspark/dspark_draft_model.py | .py | from __future__ import annotations
from typing import Optional
import msgspec
import torch
import torch.nn.functional as F
import triton
import triton.language as tl
from sglang.kernels.ops.speculative.dspark.dispatch import inputs_on_cuda
_BLOCK_V = 1024
_IDX_SENTINEL = tl.constexpr(2147483647)
class SampleStepT... | 448 | 14,288 |
sglang | python/sglang/kernels/ops/speculative/dspark/dspark_accept.py | .py | from __future__ import annotations
from typing import Optional
import msgspec
import torch
import triton
import triton.language as tl
from sglang.kernels.ops.speculative.dspark.dispatch import inputs_on_cuda
from sglang.kernels.ops.speculative.reject_sampling import (
chain_speculative_sampling_triton,
)
from sg... | 865 | 26,608 |
sglang | python/sglang/kernels/ops/speculative/dspark/dispatch.py | .py | from __future__ import annotations
import torch
def inputs_on_cuda(*args, **kwargs) -> bool:
"""Route kernel dispatch by input placement: the first tensor argument
decides. CUDA inputs take the fused triton kernel; CPU inputs take the
torch reference implementation (triton is CUDA-only, and CPU-side call... | 15 | 583 |
sglang | python/sglang/kernels/ops/speculative/dspark/fused_kv_write.py | .py | from typing import Optional
import torch
import triton
import triton.language as tl
@triton.jit
def _fused_kv_norm_rope_write_kernel(
kv_ptr,
meta_ptr,
knw_ptr,
cos_sin_ptr,
pos_ptr,
loc_ptr,
commit_lens_ptr,
locs_row_width,
KV: tl.constexpr,
D: tl.constexpr,
NH: tl.conste... | 127 | 4,130 |
sglang | python/sglang/kernels/ops/speculative/dspark/dspark_verify_window.py | .py | from __future__ import annotations
import msgspec
import torch
import triton
import triton.language as tl
from sglang.kernels.ops.speculative.cache_locs import assign_extend_cache_locs_func
from sglang.kernels.ops.speculative.dspark.dispatch import inputs_on_cuda
from sglang.srt.managers.schedule_batch import Schedul... | 908 | 27,198 |
sglang | python/sglang/kernels/ops/speculative/dspark/dspark_attn_metadata.py | .py | from __future__ import annotations
from typing import Tuple
import msgspec
import torch
import triton
import triton.language as tl
from sglang.kernels.ops.speculative.dspark.dispatch import inputs_on_cuda
from sglang.srt.utils import ceil_align
class DsparkWindowGather(msgspec.Struct, frozen=True):
num_q: int
... | 492 | 15,426 |
sglang | python/sglang/kernels/ops/activation/__init__.py | .py | """Fused gated-activation kernels (``act(x[:h]) * x[h:]``).
Each operator is a :class:`~sglang.kernels.fused_op.BaseFusedOp` with a
pure-``torch`` reference (``forward_native``) plus AOT (``sgl_kernel``) and
JIT CUDA backends behind one ``(input, out)`` signature. The JIT backend
additionally accepts ``expert_ids`` / ... | 333 | 10,949 |
sglang | python/sglang/kernels/ops/activation/activation.py | .py | from __future__ import annotations
from typing import TYPE_CHECKING, Optional
import torch
from sglang.kernels.jit.utils import (
cache_once,
get_jit_cuda_arch,
is_arch_support_pdl,
is_hip_runtime,
load_jit,
make_cpp_args,
)
from sglang.srt.utils.custom_op import register_custom_op
if TYPE_C... | 230 | 7,363 |
sglang | python/sglang/kernels/ops/activation/softcap.py | .py | import torch
import triton
import triton.language as tl
from triton.language.extra import libdevice
softcap_out_autotune = triton.autotune(
configs=[
triton.Config(kwargs={"BLOCK_SIZE": 128}, num_warps=4),
triton.Config(kwargs={"BLOCK_SIZE": 128}, num_warps=8),
triton.Config(kwargs={"BLOCK_... | 121 | 3,880 |
sglang | python/sglang/kernels/ops/mm/__init__.py | .py | """Multimodal kernels."""
__all__ = ["process"]
| 4 | 49 |
sglang | python/sglang/kernels/ops/mm/process/__init__.py | .py | """Multimodal input-processing kernels."""
from sglang.kernels.ops.mm.process.image import normalize_and_patchify
__all__ = ["normalize_and_patchify"]
| 6 | 153 |
sglang | python/sglang/kernels/ops/mm/process/image.py | .py | import torch
import torch.nn.functional as F
import triton
import triton.language as tl
_MAX_TRITON_ELEMENTS = 2**31 - 1
@triton.jit
def _normalize_and_patchify_kernel(
input_ptr,
scale_ptr,
bias_ptr,
output_ptr,
channels: tl.constexpr,
input_height,
input_width,
grid_height,
grid... | 125 | 3,687 |
sglang | python/sglang/kernels/ops/kvcache/triton_store_cache.py | .py | from typing import Literal
import torch
import triton
import triton.language as tl
from sglang.kernels.ops.quantization.fp8_kernel import is_fp8_fnuz
from sglang.srt.layers.attention.dsa.utils import (
INDEXER_K_CACHE_PRESHUFFLE_TILE,
aiter_can_use_preshuffle_paged_mqa,
)
_FP8_DTYPE = torch.float8_e4m3fnuz i... | 261 | 7,844 |
sglang | python/sglang/kernels/ops/kvcache/hisparse.py | .py | from __future__ import annotations
import functools
from typing import TYPE_CHECKING
import torch
from sglang.kernels.jit.utils import load_jit, make_cpp_args
if TYPE_CHECKING:
from tvm_ffi.module import Module
@functools.cache
def _jit_sparse_module(
item_size_bytes: int,
block_size: int,
num_top... | 330 | 9,339 |
sglang | python/sglang/kernels/ops/kvcache/mla_buffer.py | .py | from __future__ import annotations
import torch
import triton
import triton.language as tl
from sglang.kernels.jit.utils import is_arch_support_pdl
from sglang.srt.runtime_context import get_parallel
@triton.jit
def set_mla_kv_buffer_kernel(
kv_buffer_ptr,
cache_k_nope_ptr,
cache_k_rope_ptr,
loc_ptr... | 387 | 11,578 |
sglang | python/sglang/kernels/ops/kvcache/aiter_unified_attention.py | .py | import triton
import triton.language as tl
@triton.jit
def scatter_ragged_to_page_table_kernel(
kv_flat_ptr,
kv_indptr_ptr,
dest_ptr,
dest_stride,
sw_page_table_ptr,
swa_slot_mapping_ptr,
PAGE_SIZE: tl.constexpr,
BLOCK_SIZE: tl.constexpr,
HAS_SWA: tl.constexpr,
):
"""Scatter ra... | 98 | 2,789 |
sglang | python/sglang/kernels/ops/kvcache/trtllm_mha_page_table.py | .py | """Device-side page-table builder for the trtllm_mha attention backend.
trtllm_mha builds its block (page) table from the global ``req_to_token`` pool.
Doing it with a host-max PyTorch gather forces a ``seq_lens.max().item()`` D2H
sync (the CPU must know the page-table width before launching). This kernel
instead deri... | 139 | 5,714 |
sglang | python/sglang/kernels/ops/kvcache/set_mla_kv_buffer.py | .py | """JIT TMA bulk-store path for ``set_mla_kv_buffer``.
Each warp scatter-writes one item's (nope, rope) row via a single
``cp.async.bulk.global.shared::cta`` store. Requires SM90+ (Hopper or later)
for the TMA bulk-store hardware. The host-side wrapper in
``sglang.srt.mem_cache.utils`` falls back to a Triton kernel for... | 118 | 4,013 |
sglang | python/sglang/kernels/ops/kvcache/cache_ops.py | .py | import torch
import triton
import triton.language as tl
@triton.jit
def concat_and_cast_mha_k_kernel(
k_ptr,
k_nope_ptr,
k_rope_ptr,
head_cnt: tl.constexpr,
k_stride0: tl.constexpr,
k_stride1: tl.constexpr,
nope_stride0: tl.constexpr,
nope_stride1: tl.constexpr,
rope_stride0: tl.co... | 796 | 30,948 |
sglang | python/sglang/kernels/ops/kvcache/__init__.py | .py | """KV-cache write/transfer kernels.
This group wraps the Triton ``reshape_and_cache`` launcher, whose implementation
now lives in this package (``sglang.kernels.ops.kvcache.cache_ops``) after being
migrated out of ``sglang.srt.layers.attention.triton_ops`` (RFC #29630).
"""
from __future__ import annotations
from ty... | 89 | 3,056 |
sglang | python/sglang/kernels/ops/kvcache/kvcache.py | .py | from __future__ import annotations
import logging
from typing import TYPE_CHECKING
import torch
from sglang.kernels.jit.utils import (
cache_once,
is_arch_support_pdl,
load_jit,
make_cpp_args,
)
from sglang.srt.utils.custom_op import register_custom_op
if TYPE_CHECKING:
from tvm_ffi.module impor... | 111 | 3,840 |
sglang | python/sglang/kernels/ops/kvcache/kv_indices.py | .py | import triton
import triton.language as tl
_FLASHMLA_CREATE_KV_BLOCK_SIZE = 4096
FLASHMLA_CREATE_KV_BLOCK_SIZE_TRITON = tl.constexpr(_FLASHMLA_CREATE_KV_BLOCK_SIZE)
@triton.jit
def create_flashinfer_kv_indices_triton(
req_to_token_ptr, # [max_batch, max_context_len]
req_pool_indices_ptr,
page_kernel_len... | 167 | 6,003 |
sglang | python/sglang/kernels/ops/kvcache/rope_cache.py | .py | import torch
import triton
import triton.language as tl
@triton.jit
def _get_gptj_rotated_x(
x,
x_rotated_mask,
BLOCK_D: tl.constexpr,
BLOCK_D_HALF: tl.constexpr,
):
# GPT-J rotary layout:
# Pair adjacent dimensions and apply:
# [x0, x1, x2, x3] -> [-x1, x0, -x3, x2]
# Apply sign inve... | 737 | 25,651 |
sglang | python/sglang/kernels/ops/kvcache/hicache.py | .py | from __future__ import annotations
import logging
from typing import TYPE_CHECKING
from sglang.kernel_api_logging import debug_kernel_api
from sglang.kernels.jit.utils import cache_once, load_jit, make_cpp_args
if TYPE_CHECKING:
import torch
from tvm_ffi.module import Module
DEFAULT_BLOCK_QUOTA = 2
@cache... | 351 | 10,855 |
sglang | python/sglang/kernels/ops/kvcache/zero_pages.py | .py | """Zero whole page envelopes of the unified pool by physical page id.
The pool is viewed as int64 words (the MLA page envelope is always
8-byte-aligned: entry bytes per layer = kv_cache_dim * itemsize, a multiple
of 8), one wide element per lane; grid = (num_pages, page word blocks).
"""
from __future__ import annota... | 49 | 1,480 |
sglang | python/sglang/kernels/ops/kvcache/minimax_store_kv_index.py | .py | from __future__ import annotations
from typing import TYPE_CHECKING, Optional
import torch
from sglang.kernels.jit.utils import (
cache_once,
is_arch_support_pdl,
load_jit,
make_cpp_args,
)
if TYPE_CHECKING:
from tvm_ffi.module import Module
@cache_once
def _jit_module(head_bytes: int) -> Modu... | 80 | 2,331 |
sglang | python/sglang/kernels/ops/kvcache/trtllm_mha_graph_metadata.py | .py | """Fused CUDA-graph metadata update for the TRTLLM MHA backend.
`TRTLLMHAAttnBackend._apply_cuda_graph_metadata` used to rebuild the
page table(s) and seqlen buffers with ~25 small aten ops per graph
replay (index gathers, floor_divide, cumsum, dtype casts, copies).
On some CPUs that is ~0.7-1.0 ms of pure host dispat... | 207 | 8,421 |
sglang | python/sglang/kernels/ops/kvcache/cache_move.py | .py | import torch
import triton
import triton.language as tl
from sglang.srt.utils import is_cpu
_is_cpu = is_cpu()
if _is_cpu:
from sgl_kernel import copy_all_layer_kv_cache_cpu
@triton.jit
def set_kv_buffer_prefix_valid_tiled(
src_k_ptr,
src_v_ptr,
dst_k_ptr,
dst_v_ptr,
loc_2d_ptr,
commit_... | 307 | 10,403 |
sglang | python/sglang/kernels/ops/kvcache/fused_fp8_qkv_kv_cache.py | .py | from __future__ import annotations
from typing import TYPE_CHECKING, Optional
import torch
from sglang.kernels.jit.utils import (
cache_once,
is_arch_support_pdl,
load_jit,
make_cpp_args,
)
if TYPE_CHECKING:
from tvm_ffi.module import Module
@cache_once
def _jit_fused_fp8_qkv_kv_cache_module(d... | 71 | 2,036 |
sglang | python/sglang/kernels/ops/embeddings/__init__.py | .py | """Embedding kernels."""
from sglang.kernels.registry import register_kernel
from sglang.kernels.spec import KernelBackend, KernelSpec
register_kernel(
KernelSpec(
op="embeddings.vocab_parallel_embedding",
backend=KernelBackend.TRITON,
target=(
"sglang.kernels.ops.embeddings.vo... | 18 | 417 |
sglang | python/sglang/kernels/ops/embeddings/vocab_parallel_embedding.py | .py | """Fused Triton vocabulary-parallel embedding lookup."""
import torch
import triton
import triton.language as tl
@triton.jit
def _vocab_parallel_embedding_kernel(
input_ptr,
weight_ptr,
out_ptr,
# The scalar params are tl.constexpr on purpose: it lets the compiler fold
# the vocab-window comparis... | 97 | 2,965 |
sglang | python/sglang/kernels/ops/mamba/transfer_mamba.py | .py | """JIT-compiled Mamba KV cache transfer kernel.
Provides ``transfer_kv_mamba_pf_lf`` (load: page_first -> layer_first)
and ``transfer_kv_mamba_lf_pf`` (backup: layer_first -> page_first).
Uses the shared ``load_jit`` + ``cache_once`` infrastructure from
``sglang.kernels.jit.utils`` — the same mechanism used by ``hica... | 84 | 2,131 |
sglang | python/sglang/kernels/ops/mamba/__init__.py | .py | """State-space / Mamba kernels (causal conv1d)."""
from __future__ import annotations
from typing import TYPE_CHECKING, Optional
from sglang.kernels.registry import register_kernel
from sglang.kernels.selector import get_kernel
from sglang.kernels.spec import FormatSignature, KernelBackend, KernelSpec
if TYPE_CHECK... | 104 | 2,904 |
sglang | python/sglang/kernels/ops/mamba/causal_conv1d_triton.py | .py | # SPDX-License-Identifier: Apache-2.0
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
# Copyright (c) 2024, Tri Dao.
# Adapted from https://github.com/Dao-AILab/causal-conv1d/blob/main/causal_conv1d/causal_conv1d_interface.py
# and https://github.com/vllm-project/vllm/blob/main/vllm/model_executor/... | 1,208 | 48,091 |
sglang | python/sglang/kernels/ops/mamba/mamba_state_scatter_triton.py | .py | """
Fused Triton kernel for Mamba state scatter operations.
This kernel replaces the expensive advanced indexing operations in
`update_mamba_state_after_mtp_verify` with a single fused gather-scatter kernel,
avoiding multiple `index_elementwise_kernel` launches.
"""
import torch
import triton
import triton.language a... | 890 | 31,220 |
sglang | python/sglang/kernels/ops/mamba/mamba_state_indices_triton.py | .py | """Fused replay-prep state-indices kernel for the mamba cuda-graph path.
``MambaAttnBackendBase._replay_metadata`` refreshes the captured per-bs
``state_indices_list`` buffer before every cuda-graph replay. The reference
form is a chain of dispatched aten ops whose host cost shows up in the bs=1
MTP inter-phase seam:
... | 76 | 2,904 |
sglang | python/sglang/kernels/ops/mamba/inkling_sconv.py | .py | """CUDA-JIT implementations of the Inkling short-convolution kernels.
Their signatures match the Triton entrypoints so model layers can select either
backend without adapting arguments.
"""
from __future__ import annotations
from typing import TYPE_CHECKING
import torch
from sglang.kernels.jit.utils import cache_o... | 240 | 7,677 |
sglang | python/sglang/kernels/ops/mamba/triton_ops/ssd_state_passing.py | .py | # Adapted from: https://github.com/vllm-project/vllm/tree/main/vllm/model_executor/layers/mamba/ops/ssd_state_passing.py
# SPDX-License-Identifier: Apache-2.0
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
# Copyright (c) 2024, Tri Dao, Albert Gu.
# Adapted from https://github.com/state-spaces/m... | 265 | 10,060 |
sglang | python/sglang/kernels/ops/mamba/triton_ops/ssd_chunk_scan.py | .py | # Adapted from: https://github.com/vllm-project/vllm/tree/main/vllm/model_executor/layers/mamba/ops/ssd_chunk_scan.py
# SPDX-License-Identifier: Apache-2.0
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
# Copyright (c) 2024, Tri Dao, Albert Gu.
# Adapted from https://github.com/state-spaces/mamb... | 563 | 18,405 |
sglang | python/sglang/kernels/ops/mamba/triton_ops/ssd_combined.py | .py | # Adapted from: https://github.com/vllm-project/vllm/tree/main/vllm/model_executor/layers/mamba/ops/ssd_combined.py
# SPDX-License-Identifier: Apache-2.0
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
# Copyright (c) 2024, Tri Dao, Albert Gu.
# Adapted from https://github.com/state-spaces/mamba/... | 276 | 9,323 |
sglang | python/sglang/kernels/ops/mamba/triton_ops/__init__.py | .py | from .mamba_ssm import PAD_SLOT_ID
from .ssd_combined import mamba_chunk_scan_combined
from .ssu_dispatch import (
initialize_mamba_selective_state_update_backend,
selective_state_update,
)
__all__ = [
"PAD_SLOT_ID",
"selective_state_update",
"mamba_chunk_scan_combined",
"initialize_mamba_selec... | 14 | 350 |
sglang | python/sglang/kernels/ops/mamba/triton_ops/ssd_chunk_state.py | .py | # Adapted from: https://github.com/vllm-project/vllm/tree/main/vllm/model_executor/layers/mamba/ops/ssd_chunk_state.py
# SPDX-License-Identifier: Apache-2.0
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
# Copyright (c) 2024, Tri Dao, Albert Gu.
# Adapted from https://github.com/state-spaces/mam... | 647 | 20,854 |
sglang | python/sglang/kernels/ops/mamba/triton_ops/ssu_dispatch.py | .py | from __future__ import annotations
import logging
from abc import ABC, abstractmethod
from typing import TYPE_CHECKING
import torch
if TYPE_CHECKING:
from sglang.srt.server_args import ServerArgs
logger = logging.getLogger(__name__)
class MambaSSUBackend(ABC):
@property
@abstractmethod
def name(se... | 306 | 9,971 |
sglang | python/sglang/kernels/ops/mamba/triton_ops/ssd_bmm.py | .py | # Adapted from: https://github.com/vllm-project/vllm/tree/main/vllm/model_executor/layers/mamba/ops/ssd_bmm.py
# SPDX-License-Identifier: Apache-2.0
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
# Copyright (c) 2024, Tri Dao, Albert Gu.
# Adapted from https://github.com/state-spaces/mamba/blob/... | 215 | 6,821 |
sglang | python/sglang/kernels/ops/mamba/triton_ops/mamba_ssm.py | .py | # Adapted from: https://github.com/vllm-project/vllm/tree/main/vllm/model_executor/layers/mamba/ops/mamba_ssm.py
# SPDX-License-Identifier: Apache-2.0
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
# Copyright (c) 2024, Tri Dao, Albert Gu.
# Adapted from https://github.com/state-spaces/mamba/blo... | 569 | 18,579 |
sglang | python/sglang/kernels/ops/elementwise/add3.py | .py | """CUDA JIT elementwise 3-way add: out = bf16(bf16(a + b) + c)."""
from __future__ import annotations
from typing import TYPE_CHECKING
import torch
from sglang.kernels.jit.utils import (
cache_once,
is_arch_support_pdl,
load_jit,
make_cpp_args,
)
from sglang.srt.utils import is_npu
if TYPE_CHECKING... | 70 | 2,004 |
sglang | python/sglang/kernels/ops/elementwise/__init__.py | .py | """Generic elementwise / fused-pointwise kernels.
Home for cross-cutting pointwise kernels that do not belong to a single
functional group: the fused-pointwise Triton collection (``elementwise``:
sigmoid-mul, gated-activation and fused-rmsnorm variants shared across models)
and the ``add_constant`` JIT reference kerne... | 12 | 470 |
sglang | python/sglang/kernels/ops/elementwise/add_constant.py | .py | from __future__ import annotations
from typing import TYPE_CHECKING
import torch
from sglang.kernels.jit.utils import cache_once, load_jit, make_cpp_args
if TYPE_CHECKING:
from tvm_ffi.module import Module
@cache_once
def _jit_add_constant_module(constant: int) -> Module:
args = make_cpp_args(constant)
... | 29 | 691 |
sglang | python/sglang/kernels/ops/elementwise/elementwise.py | .py | import torch
import triton
import triton.language as tl
from sglang.kernels.jit.utils import is_arch_support_pdl
from sglang.srt.utils import is_hip
_is_hip = is_hip()
rmsnorm_autotune = triton.autotune(
configs=[
triton.Config(kwargs={"BLOCK_SIZE": 1024}, num_warps=4, num_stages=1),
triton.Conf... | 548 | 17,571 |
sglang | python/sglang/kernels/ops/diffusion/modulate_scale_shift.py | .py | """Fused adaLN modulate: ``x * (1 + scale) + shift`` in one CUDA kernel.
Numerical contract: the kernel reproduces each eager op's
fp32-opmath/round-to-storage-dtype boundary (fp16/bf16), so its output is
bit-exact vs the eager chain (``torch.equal``) and needs no quality gate.
"""
from __future__ import annotations
... | 131 | 4,029 |
sglang | python/sglang/kernels/ops/diffusion/residual_gate_add.py | .py | from __future__ import annotations
import logging
from typing import TYPE_CHECKING
import torch
from sglang.kernels.jit.utils import cache_once, load_jit, make_cpp_args
from sglang.srt.utils.custom_op import register_custom_op
if TYPE_CHECKING:
from tvm_ffi.module import Module
_SUPPORTED_DTYPES = (torch.floa... | 137 | 4,215 |
sglang | python/sglang/kernels/ops/diffusion/timestep_embedding.py | .py | from __future__ import annotations
from typing import TYPE_CHECKING
import torch
from sglang.kernel_api_logging import debug_kernel_api
from sglang.kernels.jit.utils import cache_once, load_jit, make_cpp_args
if TYPE_CHECKING:
from tvm_ffi.module import Module
@cache_once
def _jit_timestep_embedding_module(dt... | 54 | 1,374 |
sglang | python/sglang/kernels/ops/diffusion/fused_ln_modulate.py | .py | """LayerNorm + adaLN modulate folded into one affine LN call.
``layer_norm(x, weight=(1 + scale), bias=shift)`` replaces the affine-free
LayerNorm + modulate pair: one kernel and one HBM pass per site instead of
two. ``1 + scale`` keeps the eager rounding of the [1, D] modulation row,
but scale/shift then apply in fp... | 82 | 2,551 |
sglang | python/sglang/kernels/ops/diffusion/bitexact_gate.py | .py | """Shared first-sight verification for bit-exact diffusion fast paths."""
from __future__ import annotations
import logging
from collections.abc import Callable
from typing import Any, TypeVar
import torch
T = TypeVar("T")
EqualFn = Callable[[Any, Any], bool]
DiagnosticHintFn = Callable[[], str | None]
def flashi... | 178 | 6,462 |
sglang | python/sglang/kernels/ops/diffusion/__init__.py | .py | """Registered diffusion-model kernels and their public wrappers.
Hot paths import concrete implementations from submodules. The package-level
wrappers remain available for backward compatibility.
"""
from __future__ import annotations
from typing import TYPE_CHECKING
from sglang.kernels.registry import register_ker... | 123 | 3,622 |
sglang | python/sglang/kernels/ops/diffusion/usp_relayout.py | .py | from __future__ import annotations
from typing import TYPE_CHECKING
import torch
from sglang.kernels.jit.utils import cache_once, load_jit, make_cpp_args
from sglang.srt.utils.custom_op import register_custom_op
if TYPE_CHECKING:
from tvm_ffi.module import Module
_SUPPORTED_DTYPES = (torch.float16, torch.bflo... | 86 | 2,599 |
sglang | python/sglang/kernels/ops/diffusion/ltx2_rmsnorm_modulate.py | .py | """Weightless RMSNorm + adaLN modulate folded into one kernel for LTX-2.
``rms_norm(x) * (1 + scale) + shift`` at the LTX-2 transformer-block adaLN
sites is otherwise an aten ``F.rms_norm`` plus a separate ``mul``/``add``
modulate (one reduction kernel plus several pointwise passes per site). This
folds the whole chai... | 81 | 2,945 |
sglang | python/sglang/kernels/ops/diffusion/quality_gate.py | .py | """Shared module-site protocol for request-scoped diffusion fast paths."""
from __future__ import annotations
import logging
from collections.abc import Callable, Iterator
from typing import Any
from torch import nn
RejectReason = Callable[[nn.Module], str | None]
class QualityGatedFusion:
"""Track and toggle... | 83 | 2,732 |
sglang | python/sglang/kernels/ops/diffusion/ltx2_qknorm_split_rope.py | .py | from __future__ import annotations
from typing import TYPE_CHECKING
import torch
from sglang.kernels.jit.utils import cache_once, load_jit
from sglang.srt.utils.custom_op import register_custom_op
if TYPE_CHECKING:
from tvm_ffi.module import Module
@cache_once
def _jit_ltx2_qknorm_split_rope_module() -> Modul... | 206 | 4,809 |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.