Add files using upload-large-folder tool
Browse filesThis view is limited to 50 files because it contains too many changes. See raw diff
- .gitattributes +31 -0
- runs/s1_numina_q17/asft_lr1e-5/args.json +26 -0
- runs/s1_numina_q17/asft_lr1e-5/eval.log +87 -0
- runs/s1_numina_q17/asft_lr1e-5/eval.sbatch +13 -0
- runs/s1_numina_q17/asft_lr1e-5/eval/aime24.jsonl +0 -0
- runs/s1_numina_q17/asft_lr1e-5/eval/aime25.jsonl +0 -0
- runs/s1_numina_q17/asft_lr1e-5/eval/amc23.jsonl +0 -0
- runs/s1_numina_q17/asft_lr1e-5/eval/gsm8k.jsonl +0 -0
- runs/s1_numina_q17/asft_lr1e-5/eval/math500.jsonl +0 -0
- runs/s1_numina_q17/asft_lr1e-5/eval/minerva.jsonl +0 -0
- runs/s1_numina_q17/asft_lr1e-5/eval/olympiadbench.jsonl +0 -0
- runs/s1_numina_q17/asft_lr1e-5/eval/scores.json +63 -0
- runs/s1_numina_q17/asft_lr1e-5/final/chat_template.jinja +85 -0
- runs/s1_numina_q17/asft_lr1e-5/final/config.json +63 -0
- runs/s1_numina_q17/asft_lr1e-5/final/generation_config.json +7 -0
- runs/s1_numina_q17/asft_lr1e-5/final/tokenizer.json +3 -0
- runs/s1_numina_q17/asft_lr1e-5/final/tokenizer_config.json +30 -0
- runs/s1_numina_q17/asft_lr1e-5/train.log +121 -0
- runs/s1_numina_q17/asft_lr1e-5/train.sbatch +12 -0
- runs/s1_numina_q17/asft_lr1e-5/train_log.jsonl +0 -0
- runs/s1_numina_q17/asft_lr1e-6/args.json +30 -0
- runs/s1_numina_q17/asft_lr1e-6/eval.log +83 -0
- runs/s1_numina_q17/asft_lr1e-6/eval.sbatch +13 -0
- runs/s1_numina_q17/asft_lr1e-6/eval/aime24.jsonl +0 -0
- runs/s1_numina_q17/asft_lr1e-6/eval/aime25.jsonl +0 -0
- runs/s1_numina_q17/asft_lr1e-6/eval/amc23.jsonl +0 -0
- runs/s1_numina_q17/asft_lr1e-6/eval/gsm8k.jsonl +0 -0
- runs/s1_numina_q17/asft_lr1e-6/eval/math500.jsonl +0 -0
- runs/s1_numina_q17/asft_lr1e-6/eval/minerva.jsonl +0 -0
- runs/s1_numina_q17/asft_lr1e-6/eval/olympiadbench.jsonl +0 -0
- runs/s1_numina_q17/asft_lr1e-6/eval/per_item.json +1 -0
- runs/s1_numina_q17/asft_lr1e-6/eval/scores.json +63 -0
- runs/s1_numina_q17/asft_lr1e-6/final/chat_template.jinja +85 -0
- runs/s1_numina_q17/asft_lr1e-6/final/config.json +63 -0
- runs/s1_numina_q17/asft_lr1e-6/final/generation_config.json +7 -0
- runs/s1_numina_q17/asft_lr1e-6/final/tokenizer.json +3 -0
- runs/s1_numina_q17/asft_lr1e-6/final/tokenizer_config.json +30 -0
- runs/s1_numina_q17/asft_lr1e-6/train.log +121 -0
- runs/s1_numina_q17/asft_lr1e-6/train.sbatch +12 -0
- runs/s1_numina_q17/asft_lr1e-6/train_log.jsonl +0 -0
- runs/s1_numina_q17/asft_lr2e-5/args.json +30 -0
- runs/s1_numina_q17/asft_lr2e-5/eval.log +85 -0
- runs/s1_numina_q17/asft_lr2e-5/eval.sbatch +13 -0
- runs/s1_numina_q17/asft_lr2e-5/eval/aime24.jsonl +0 -0
- runs/s1_numina_q17/asft_lr2e-5/eval/aime25.jsonl +0 -0
- runs/s1_numina_q17/asft_lr2e-5/eval/amc23.jsonl +0 -0
- runs/s1_numina_q17/asft_lr2e-5/eval/gsm8k.jsonl +0 -0
- runs/s1_numina_q17/asft_lr2e-5/eval/math500.jsonl +0 -0
- runs/s1_numina_q17/asft_lr2e-5/eval/minerva.jsonl +0 -0
- runs/s1_numina_q17/asft_lr2e-5/eval/olympiadbench.jsonl +0 -0
.gitattributes
CHANGED
|
@@ -94,3 +94,34 @@ runs/s1_numina_q17/psft_lr2e-5/final/tokenizer.json filter=lfs diff=lfs merge=lf
|
|
| 94 |
runs/s1_numina_q17/basew-a1_lr2e-6_s1/final/tokenizer.json filter=lfs diff=lfs merge=lfs -text
|
| 95 |
runs/s1_numina_q17/priftC-empirical-g1_lr1e-6/final/tokenizer.json filter=lfs diff=lfs merge=lfs -text
|
| 96 |
runs/s1_numina_q17/asft_lr2e-6/final/tokenizer.json filter=lfs diff=lfs merge=lfs -text
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 94 |
runs/s1_numina_q17/basew-a1_lr2e-6_s1/final/tokenizer.json filter=lfs diff=lfs merge=lfs -text
|
| 95 |
runs/s1_numina_q17/priftC-empirical-g1_lr1e-6/final/tokenizer.json filter=lfs diff=lfs merge=lfs -text
|
| 96 |
runs/s1_numina_q17/asft_lr2e-6/final/tokenizer.json filter=lfs diff=lfs merge=lfs -text
|
| 97 |
+
runs/s1_numina_q17/nll_lr1e-6/final/tokenizer.json filter=lfs diff=lfs merge=lfs -text
|
| 98 |
+
runs/s1_numina_q17/psft_lr2e-6/final/tokenizer.json filter=lfs diff=lfs merge=lfs -text
|
| 99 |
+
runs/s1_numina_q17/trimsft_lr2e-5/final/tokenizer.json filter=lfs diff=lfs merge=lfs -text
|
| 100 |
+
runs/s1_numina_q17/basew-a0.5_lr2e-6/final/tokenizer.json filter=lfs diff=lfs merge=lfs -text
|
| 101 |
+
runs/s1_numina_q17/focal_lr2e-5/final/tokenizer.json filter=lfs diff=lfs merge=lfs -text
|
| 102 |
+
runs/s1_numina_q17/dft_lr2e-5/final/tokenizer.json filter=lfs diff=lfs merge=lfs -text
|
| 103 |
+
runs/s1_numina_q17/dft_lr5e-6/final/tokenizer.json filter=lfs diff=lfs merge=lfs -text
|
| 104 |
+
runs/s1_numina_q17/dft_lr1e-6/final/tokenizer.json filter=lfs diff=lfs merge=lfs -text
|
| 105 |
+
runs/s1_numina_q17/priftC-true-g10_lr1e-6/final/tokenizer.json filter=lfs diff=lfs merge=lfs -text
|
| 106 |
+
runs/s1_numina_q17/trimsft_lr5e-6/final/tokenizer.json filter=lfs diff=lfs merge=lfs -text
|
| 107 |
+
runs/s1_numina_q17/basew-a1_lr2e-6_s2/final/tokenizer.json filter=lfs diff=lfs merge=lfs -text
|
| 108 |
+
runs/s1_numina_q17/nll_lr5e-7/final/tokenizer.json filter=lfs diff=lfs merge=lfs -text
|
| 109 |
+
runs/s1_numina_q17/asft_lr2e-5/final/tokenizer.json filter=lfs diff=lfs merge=lfs -text
|
| 110 |
+
runs/s1_numina_q17/focal_lr5e-6/final/tokenizer.json filter=lfs diff=lfs merge=lfs -text
|
| 111 |
+
runs/s1_numina_q17/priftC-empirical-g1_lr2e-6/final/tokenizer.json filter=lfs diff=lfs merge=lfs -text
|
| 112 |
+
runs/s1_numina_q17/prift_lr1e-6/final/tokenizer.json filter=lfs diff=lfs merge=lfs -text
|
| 113 |
+
runs/s1_numina_q17/basewC-a0.25-true-g1_lr2e-6/final/tokenizer.json filter=lfs diff=lfs merge=lfs -text
|
| 114 |
+
runs/s1_numina_q17/prift_lr5e-5/final/tokenizer.json filter=lfs diff=lfs merge=lfs -text
|
| 115 |
+
runs/s1_numina_q17/priftC-true-g10_lr2e-6/final/tokenizer.json filter=lfs diff=lfs merge=lfs -text
|
| 116 |
+
runs/s1_numina_q17/asft_lr1e-5/final/tokenizer.json filter=lfs diff=lfs merge=lfs -text
|
| 117 |
+
runs/s1_numina_q17/prift_lr2e-6/final/tokenizer.json filter=lfs diff=lfs merge=lfs -text
|
| 118 |
+
runs/s1_numina_q17/priftC-empirical-g10_lr2e-6/final/tokenizer.json filter=lfs diff=lfs merge=lfs -text
|
| 119 |
+
runs/s1_numina_q17/asft_lr1e-6/final/tokenizer.json filter=lfs diff=lfs merge=lfs -text
|
| 120 |
+
runs/s1_numina_q17/priftC-true-g1_lr2e-6/final/tokenizer.json filter=lfs diff=lfs merge=lfs -text
|
| 121 |
+
runs/s1_numina_q17/basew-a1_lr5e-6/final/tokenizer.json filter=lfs diff=lfs merge=lfs -text
|
| 122 |
+
runs/s2_tulu_q17/trimsft_lr2e-6/final/tokenizer.json filter=lfs diff=lfs merge=lfs -text
|
| 123 |
+
runs/s1_numina_q17/prift_lr2e-5/final/tokenizer.json filter=lfs diff=lfs merge=lfs -text
|
| 124 |
+
runs/s1_numina_q17/priftC-empirical-g10_lr1e-6/final/tokenizer.json filter=lfs diff=lfs merge=lfs -text
|
| 125 |
+
runs/s1_numina_q17/prift_lr1e-5/final/tokenizer.json filter=lfs diff=lfs merge=lfs -text
|
| 126 |
+
runs/s1_numina_q17/priftC-true-g1_lr1e-6/final/tokenizer.json filter=lfs diff=lfs merge=lfs -text
|
| 127 |
+
runs/s1_numina_q17/prift_lr5e-6/final/tokenizer.json filter=lfs diff=lfs merge=lfs -text
|
runs/s1_numina_q17/asft_lr1e-5/args.json
ADDED
|
@@ -0,0 +1,26 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"model": "/fsx_0/user/lyw/projects/arl_data/models/Qwen3-1.7B-Base",
|
| 3 |
+
"data": "/fsx_0/user/lyw/projects/arl_data/sft/v1/numina_cot_20k.jsonl",
|
| 4 |
+
"out": "/fsx_0/user/lyw/projects/arl_data/runs/s1_numina_q17/asft_lr1e-5",
|
| 5 |
+
"method": "asft",
|
| 6 |
+
"method_args": "{}",
|
| 7 |
+
"template": "plain",
|
| 8 |
+
"max_len": 4096,
|
| 9 |
+
"limit": 0,
|
| 10 |
+
"lr": 1e-05,
|
| 11 |
+
"epochs": 3.0,
|
| 12 |
+
"seqs_per_step": 64,
|
| 13 |
+
"micro_tokens": 16384,
|
| 14 |
+
"chunk": 2048,
|
| 15 |
+
"warmup": 0.05,
|
| 16 |
+
"clip": 1.0,
|
| 17 |
+
"seed": 0,
|
| 18 |
+
"ckpt_every_min": 40,
|
| 19 |
+
"stop_at": 0,
|
| 20 |
+
"save_at": "",
|
| 21 |
+
"zero1": false,
|
| 22 |
+
"no_save": false,
|
| 23 |
+
"method_cfg": {
|
| 24 |
+
"lam": 0.05
|
| 25 |
+
}
|
| 26 |
+
}
|
runs/s1_numina_q17/asft_lr1e-5/eval.log
ADDED
|
@@ -0,0 +1,87 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
(MainProcess pid=3547341) INFO 11-12 09:24:09 [api_utils.py:286] non-default args: {'dtype': 'bfloat16', 'max_model_len': 6144, 'gpu_memory_utilization': 0.85, 'disable_log_stats': True, 'model': '/fsx_0/user/lyw/projects/arl_data/runs/s1_numina_q17/asft_lr1e-5/final'}
|
| 2 |
+
(MainProcess pid=3547341) INFO 11-12 09:24:09 [model.py:733] Resolved architecture: Qwen3ForCausalLM
|
| 3 |
+
(MainProcess pid=3547341) INFO 11-12 09:24:09 [model.py:2495] Downcasting torch.float32 to torch.bfloat16.
|
| 4 |
+
(MainProcess pid=3547341) INFO 11-12 09:24:09 [model.py:2152] Using max model len 6144
|
| 5 |
+
(MainProcess pid=3547341) INFO 11-12 09:24:09 [scheduler.py:307] Chunked prefill is enabled with max_num_batched_tokens=16384.
|
| 6 |
+
(MainProcess pid=3547341) INFO 11-12 09:24:09 [kernel.py:467] Final IR op priority after setting platform defaults: IrOpPriorityConfig(rms_norm=['native'], fused_add_rms_norm=['native'], gelu_and_mul_sparse=['triton', 'native'])
|
| 7 |
+
(EngineCore pid=3550619) INFO 11-12 09:24:12 [core.py:133] Initializing a V1 LLM engine (v0.32.0) with config: model='/fsx_0/user/lyw/projects/arl_data/runs/s1_numina_q17/asft_lr1e-5/final', speculative_config=None, tokenizer='/fsx_0/user/lyw/projects/arl_data/runs/s1_numina_q17/asft_lr1e-5/final', skip_tokenizer_init=False, tokenizer_mode=auto, revision=None, tokenizer_revision=None, trust_remote_code=False, dtype=torch.bfloat16, max_seq_len=6144, download_dir=None, load_format=auto, tensor_parallel_size=1, pipeline_parallel_size=1, data_parallel_size=1, decode_context_parallel_size=1, dcp_comm_backend=ag_rs, disable_custom_all_reduce=False, quantization=None, quantization_config=None, enforce_eager=False, aux_output_config=AuxOutputConfig(enable_return_routed_experts=False, max_bytes=None), kv_cache_dtype=auto, device_config=cuda, structured_outputs_config=StructuredOutputsConfig(backend='auto', disable_any_whitespace=False, disable_additional_properties=False, reasoning_parser='', reasoning_parser_plugin='', enable_in_reasoning=False), observability_config=ObservabilityConfig(show_hidden_metrics_for_version=None, otlp_traces_endpoint=None, collect_detailed_traces=None, per_request_spec_decode_metrics='none', kv_cache_metrics=False, kv_cache_metrics_sample=0.01, custom_histogram_buckets=None, cudagraph_metrics=False, enable_layerwise_nvtx_tracing=False, enable_mfu_metrics=False, enable_mm_processor_stats=False, enable_logging_iteration_details=False, jit_monitor_mode='warn', jit_monitor_verbose=False), seed=0, served_model_name=/fsx_0/user/lyw/projects/arl_data/runs/s1_numina_q17/asft_lr1e-5/final, enable_prefix_caching=True, enable_chunked_prefill=True, pooler_config=None, compilation_config={'mode': <CompilationMode.VLLM_COMPILE: 3>, 'debug_dump_path': None, 'cache_dir': '', 'compile_cache_save_format': 'binary', 'backend': 'inductor', 'custom_ops': ['none'], 'ir_enable_torch_wrap': True, 'splitting_ops': ['vllm::unified_attention_with_output', 'vllm::unified_mla_attention_with_output', 'vllm::mamba_mixer2', 'vllm::mamba_mixer', 'vllm::short_conv', 'vllm::qwen4_exp_ple_short_conv', 'vllm::qwen4_exp_qsa_with_output', 'vllm::qwen4_exp_qsa_prepare_with_output', 'vllm::linear_attention', 'vllm::qwen_gdn_attention_core', 'vllm::qwen_gdn_attention_core_fused_norm_packed', 'vllm::gdn_attention_core_xpu', 'vllm::olmo_hybrid_gdn_full_forward', 'vllm::sparse_attn_indexer', 'vllm::rocm_aiter_sparse_attn_indexer', 'vllm::deepseek_v4_attention', 'vllm::hpc_rope_norm_forward', 'vllm::unified_kv_cache_update', 'vllm::unified_mla_kv_cache_update'], 'compile_mm_encoder': False, 'cudagraph_mm_encoder': False, 'encoder_cudagraph_token_budgets': [], 'encoder_cudagraph_max_vision_items_per_batch': 0, 'encoder_cudagraph_max_frames_per_batch': None, 'cudagraph_decoder_replay': True, 'decoder_replay_cudagraph_capture_sizes': [], 'decoder_replay_trim_threshold': 768, 'compile_sizes': [], 'compile_ranges_endpoints': [16384], 'inductor_compile_config': {'enable_auto_functionalized_v2': False, 'combo_kernels': True, 'benchmark_combo_kernel': True}, 'inductor_passes': {}, 'cudagraph_mode': <CUDAGraphMode.FULL_AND_PIECEWISE: (2, 1)>, 'cudagraph_num_of_warmups': 1, 'cudagraph_capture_sizes': [1, 2, 4, 8, 16, 24, 32, 40, 48, 56, 64, 72, 80, 88, 96, 104, 112, 120, 128, 136, 144, 152, 160, 168, 176, 184, 192, 200, 208, 216, 224, 232, 240, 248, 256, 272, 288, 304, 320, 336, 352, 368, 384, 400, 416, 432, 448, 464, 480, 496, 512], 'cudagraph_copy_inputs': False, 'cudagraph_specialize_lora': True, 'use_inductor_graph_partition': False, 'pass_config': {'fuse_norm_quant': False, 'fuse_act_quant': False, 'fuse_attn_quant': False, 'fuse_allreduce_rms': False, 'enable_qk_norm_rope_fusion': False, 'fuse_rope_kvcache_cat_mla': False, 'fuse_act_padding': False, 'fuse_qk_norm_rope_kvcache': False}, 'max_cudagraph_capture_size': 512, 'dynamic_shapes_config': {'type': <DynamicShapesType.BACKED: 'backed'>, 'evaluate_guards': False, 'assume_32_bit_indexing': False}, 'local_cache_dir': None, 'fast_moe_cold_start': False, 'static_all_moe_layers': []}, kernel_config=KernelConfig(ir_op_priority=IrOpPriorityConfig(rms_norm=['native'], fused_add_rms_norm=['native'], gelu_and_mul_sparse=['triton', 'native']), enable_flashinfer_autotune=True, enable_rocm_segmented_attn_autotune=False, enable_cutedsl_warmup=True, enable_jit_warmup=True, moe_backend='auto', sparse_indexer_topk_backend='auto', linear_backend='auto', linear_backend_per_quant=None)
|
| 8 |
+
(EngineCore pid=3550619) INFO 11-12 09:24:15 [parallel_state.py:1906] world_size=1 rank=0 local_rank=0 distributed_init_method=file:///fsx_0/user/lyw/projects/jiaruiliu/adaptive-reasoning-loss/tmp/vllm_dist_c2b95be9a8884ffdbfe141c023b74a06 backend=nccl
|
| 9 |
+
[rank0]:[W1112 09:24:15.972180667 Utils.hpp:137] Warning: Environment variable TORCH_NCCL_TRACE_BUFFER_SIZE is deprecated; use TORCH_FR_BUFFER_SIZE instead (function operator())
|
| 10 |
+
(EngineCore pid=3550619) INFO 11-12 09:24:15 [parallel_state.py:2350] rank 0 in world size 1 is assigned as DP rank 0, PP rank 0, PCP rank 0, TP rank 0, ETP rank 0, EP rank N/A, EPLB rank N/A
|
| 11 |
+
(EngineCore pid=3550619) INFO 11-12 09:24:15 [gpu_worker.py:518] Using V2 Model Runner
|
| 12 |
+
(EngineCore pid=3550619) WARNING 11-12 09:24:15 [flashinfer.py:143] FlashInfer's precompiled kernels are missing or do not match flashinfer-python 0.7.0.post1 (flashinfer-cubin 0.7.0.post1, flashinfer-jit-cache not installed). FlashInfer then downloads and compiles kernels at startup, which can take several minutes, or fails to import mismatched ones. Run `vllm download-kernels` in this Python environment to install them.
|
| 13 |
+
(EngineCore pid=3550619) INFO 11-12 09:24:16 [model_runner.py:392] Loading model from scratch...
|
| 14 |
+
(EngineCore pid=3550619) INFO 11-12 09:24:19 [cuda.py:506] Using FLASH_ATTN attention backend out of potential backends: ['FLASH_ATTN', 'FLASHINFER', 'TRITON_ATTN', 'FLEX_ATTENTION'].
|
| 15 |
+
(EngineCore pid=3550619) INFO 11-12 09:24:19 [flash_attn.py:1117] Using FlashAttention version 3
|
| 16 |
+
(EngineCore pid=3550619) INFO 11-12 09:24:20 [base_loader.py:82] Initializing model took 3.453413 seconds
|
| 17 |
+
(EngineCore pid=3550619) INFO 11-12 09:24:20 [weight_utils.py:1108] Filesystem type for checkpoints: LUSTRE. Checkpoint size: 3.78 GiB. Available RAM: 1672.11 GiB.
|
| 18 |
+
(EngineCore pid=3550619) INFO 11-12 09:24:20 [weight_utils.py:1070] Prefetching checkpoint files into page cache started (in background, num_threads=8, block_size=16777216 bytes)
|
| 19 |
+
|
| 20 |
+
(EngineCore pid=3550619) INFO 11-12 09:24:23 [weight_utils.py:1042] Prefetching checkpoint files: 10% (1/1)
|
| 21 |
+
(EngineCore pid=3550619) INFO 11-12 09:24:23 [weight_utils.py:1065] Prefetching checkpoint files into page cache finished in 3.04s
|
| 22 |
+
|
| 23 |
+
|
| 24 |
+
|
| 25 |
+
(EngineCore pid=3550619) INFO 11-12 09:24:23 [default_loader.py:501] Loading weights took 3.50 seconds
|
| 26 |
+
(EngineCore pid=3550619) INFO 11-12 09:24:26 [model_runner.py:424] Model loading took 3.21 GiB memory and 9.684512 seconds
|
| 27 |
+
(EngineCore pid=3550619) INFO 11-12 09:24:26 [topk_topp_sampler.py:124] Using FlashInfer for top-p & top-k sampling.
|
| 28 |
+
(EngineCore pid=3550619) INFO 11-12 09:24:26 [torch_utils.py:295] Reducing Torch threads from 8 to 1 for serving. Set OMP_NUM_THREADS in the external environment to override.
|
| 29 |
+
(EngineCore pid=3550619) INFO 11-12 09:24:26 [utils.py:347] Using LBNHC KV cache layout.
|
| 30 |
+
(EngineCore pid=3550619) INFO 11-12 09:24:32 [caching.py:342] reconstructed serializable fn from standalone compile artifacts. num_artifacts=3 num_submods=29
|
| 31 |
+
(EngineCore pid=3550619) INFO 11-12 09:24:32 [decorators.py:319] [backbone] Directly load AOT compilation from path /home/lyw/.cache/vllm/torch_compile_cache/torch_aot_compile/a38eaf45b86616f3cf4452eb5cdac65a39ca9e5c18a0d2a077700eb310bcd821/rank_0_0/model
|
| 32 |
+
(EngineCore pid=3550619) INFO 11-12 09:24:32 [monitor.py:62] [backbone] torch.compile took 0.93 s in total
|
| 33 |
+
(EngineCore pid=3550619) INFO 11-12 09:24:32 [monitor.py:91] [backbone] Initial profiling/warmup run took 0.19 s
|
| 34 |
+
|
| 35 |
+
|
| 36 |
+
(EngineCore pid=3550619) INFO 11-12 09:24:37 [model_runner.py:1127] Graph capturing finished in 4 secs, took 0.41 GiB
|
| 37 |
+
(EngineCore pid=3550619) INFO 11-12 09:24:37 [gpu_worker.py:759] Available KV cache memory: 61.82 GiB
|
| 38 |
+
(EngineCore pid=3550619) INFO 11-12 09:24:37 [gpu_worker.py:774] CUDA graph memory profiling is enabled (default since v0.21.0). The current --gpu-memory-utilization=0.8500 is equivalent to --gpu-memory-utilization=0.8427 without CUDA graph memory profiling. To maintain the same effective KV cache size as before, increase --gpu-memory-utilization to 0.8573. To disable, set VLLM_MEMORY_PROFILER_ESTIMATE_CUDAGRAPHS=0.
|
| 39 |
+
(EngineCore pid=3550619) INFO 11-12 09:24:37 [kv_cache_utils.py:2327] GPU KV cache size: 578,752 tokens, Maximum concurrency for 6,144 tokens per request: 94.20x
|
| 40 |
+
(EngineCore pid=3550619) INFO 11-12 09:24:38 [kernel_warmup.py:176] JIT kernel warmup starting.
|
| 41 |
+
(EngineCore pid=3550619) INFO 11-12 09:24:38 [kernel_warmup.py:189] JIT kernel warmup finished in 0.00s.
|
| 42 |
+
(EngineCore pid=3550619) INFO 11-12 09:24:38 [kernel.py:467] Final IR op priority after setting platform defaults: IrOpPriorityConfig(rms_norm=['native'], fused_add_rms_norm=['native'], gelu_and_mul_sparse=['triton', 'native'])
|
| 43 |
+
(EngineCore pid=3550619) INFO 11-12 09:24:38 [kernel_warmup.py:460] Using FlashInfer autotune cache file: /home/lyw/.cache/vllm/flashinfer_autotune_cache/0.7.0.post1/90a/30141a2b37a986207443ec41d6eb9145984353fa6ac00fc1bbb9e30dd5933af9/autotune_configs_dp0_rank0.json
|
| 44 |
+
2026-11-12 09:24:38,272 - INFO - autotuner.py:1046 - flashinfer.jit: [Autotuner]: Autotuning process starts ...
|
| 45 |
+
(EngineCore pid=3550619) INFO 11-12 09:24:38 [kernel_warmup.py:367] Running FlashInfer autotune with 16384 tokens and token buckets (1, 2, 4, 8, 16, 32, 64, 128, 256, 512, 768, 1024, 1280, 1536, 1792, 2048, 2560, 3072, 3584, 4096, 8192, 16384).
|
| 46 |
+
2026-11-12 09:24:38,299 - INFO - autotuner.py:1069 - flashinfer.jit: [Autotuner]: Autotuning process ends
|
| 47 |
+
|
| 48 |
+
|
| 49 |
+
(EngineCore pid=3550619) INFO 11-12 09:24:44 [model_runner.py:1127] Graph capturing finished in 5 secs, took 0.27 GiB
|
| 50 |
+
(EngineCore pid=3550619) INFO 11-12 09:24:44 [gpu_worker.py:971] CUDA graph pool memory: 0.27 GiB (actual), 0.58 GiB (estimated), difference: 0.3 GiB (110.7%).
|
| 51 |
+
(EngineCore pid=3550619) INFO 11-12 09:24:44 [gpu_worker.py:1034] Free memory on device (78.66/79.18 GiB) on startup. Desired GPU memory utilization is (0.85, 67.3 GiB). Actual usage is 4.16 GiB for consumed memory (weights + non-torch), 1.33 GiB for peak activation, and 0.27 GiB for CUDAGraph memory. Replace gpu_memory_utilization config with `--kv-cache-memory=65925157991` (61.4 GiB) to fit into requested memory, or `--kv-cache-memory=78125379072` (72.76 GiB) to fully utilize gpu memory. Current kv cache memory in use is 61.82 GiB.
|
| 52 |
+
(EngineCore pid=3550619) INFO 11-12 09:24:47 [jit_monitor.py:84] Kernel JIT monitor activated; monitored JIT compilations during inference will use mode=warn.
|
| 53 |
+
(EngineCore pid=3550619) INFO 11-12 09:24:47 [core.py:395] init engine (profile, create kv cache, warmup model) took 21.20 s (compilation: 0.93 s)
|
| 54 |
+
(EngineCore pid=3550619) INFO 11-12 09:24:47 [kernel.py:467] Final IR op priority after setting platform defaults: IrOpPriorityConfig(rms_norm=['native'], fused_add_rms_norm=['native'], gelu_and_mul_sparse=['triton', 'native'])
|
| 55 |
+
(MainProcess pid=3547341) INFO 11-12 09:24:48 [hf.py:648] Detected the chat template content format to be 'string'. You can set `--chat-template-content-format` to override this.
|
| 56 |
+
|
| 57 |
+
(EngineCore pid=3550619) WARNING 11-12 09:24:48 [jit_monitor.py:140] Triton kernel JIT compilation during inference: _topp_sb_step_kernel. This causes a latency spike; consider extending warmup to cover this shape/config.
|
| 58 |
+
(EngineCore pid=3550619) WARNING 11-12 09:24:48 [jit_monitor.py:140] Triton kernel JIT compilation during inference: _topp_sb_mask_kernel. This causes a latency spike; consider extending warmup to cover this shape/config.
|
| 59 |
+
|
| 60 |
+
|
| 61 |
+
|
| 62 |
+
|
| 63 |
+
wrote /fsx_0/user/lyw/projects/arl_data/runs/s1_numina_q17/asft_lr1e-5/eval/gsm8k.jsonl
|
| 64 |
+
(MainProcess pid=3547341) INFO 11-12 09:25:18 [utils.py:628] [shutdown] Process manager: send sigterm to process EngineCore
|
| 65 |
+
(EngineCore pid=3550619) INFO 11-12 09:25:18 [core.py:1454] [shutdown] EngineCore: trigger received signal=SIGTERM
|
| 66 |
+
(EngineCore pid=3550619) INFO 11-12 09:25:18 [core.py:1635] [shutdown] EngineCore: start mode=abort timeout=0s
|
| 67 |
+
(EngineCore pid=3550619) INFO 11-12 09:25:18 [core.py:1666] [shutdown] EngineCore: request processing complete; starting resource teardown
|
| 68 |
+
(EngineCore pid=3550619) INFO 11-12 09:25:18 [core.py:1499] [shutdown] EngineCore: exiting busy loop
|
| 69 |
+
aime24 {"pass1": 0.00625, "n": 16, "kind": "math", "mean_tokens": 1361.7583333333334, "truncated": 0.21458333333333332, "pass4": 0.025, "pass8": 0.05, "pass16": 0.1}
|
| 70 |
+
Timeout during comparison
|
| 71 |
+
Timeout during comparison
|
| 72 |
+
Timeout during comparison
|
| 73 |
+
Timeout during comparison
|
| 74 |
+
Timeout during comparison
|
| 75 |
+
Timeout during comparison
|
| 76 |
+
Timeout during comparison
|
| 77 |
+
aime25 {"pass1": 0.014583333333333334, "n": 16, "kind": "math", "mean_tokens": 1214.9625, "truncated": 0.16041666666666668, "pass4": 0.055, "pass8": 0.10111111111111111, "pass16": 0.16666666666666666}
|
| 78 |
+
amc23 {"pass1": 0.29375, "n": 16, "kind": "math", "mean_tokens": 759.9421875, "truncated": 0.0640625, "pass4": 0.4777472527472527, "pass8": 0.5909615384615385, "pass16": 0.7}
|
| 79 |
+
gsm8k {"pass1": 0.7581501137225171, "n": 2, "kind": "math", "mean_tokens": 218.17172100075814, "truncated": 0.00265352539802881}
|
| 80 |
+
math500 {"pass1": 0.5565, "n": 4, "kind": "math", "mean_tokens": 508.8855, "truncated": 0.0385, "pass4": 0.712}
|
| 81 |
+
Timeout during comparison
|
| 82 |
+
Timeout during comparison
|
| 83 |
+
Timeout during comparison
|
| 84 |
+
Timeout during comparison
|
| 85 |
+
minerva {"pass1": 0.1792279411764706, "n": 4, "kind": "math", "mean_tokens": 787.9926470588235, "truncated": 0.07996323529411764, "pass4": 0.3272058823529412}
|
| 86 |
+
Timeout during comparison
|
| 87 |
+
olympiadbench {"pass1": 0.23516320474777447, "n": 4, "kind": "math", "mean_tokens": 907.3056379821959, "truncated": 0.08160237388724036, "pass4": 0.37833827893175076}
|
runs/s1_numina_q17/asft_lr1e-5/eval.sbatch
ADDED
|
@@ -0,0 +1,13 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
#!/bin/bash
|
| 2 |
+
#SBATCH --job-name=arl-s1_numina_q17-asft_lr1e-5-eval
|
| 3 |
+
#SBATCH --gres=gpu:1
|
| 4 |
+
#SBATCH --cpus-per-task=8
|
| 5 |
+
#SBATCH --mem=150G
|
| 6 |
+
#SBATCH --time=3:00:00
|
| 7 |
+
#SBATCH --requeue
|
| 8 |
+
#SBATCH --output=/fsx_0/user/lyw/projects/arl_data/runs/s1_numina_q17/asft_lr1e-5/eval.log
|
| 9 |
+
export HF_HUB_OFFLINE=1 TRANSFORMERS_OFFLINE=1 ARL_EVAL_SETS=/fsx_0/user/lyw/projects/arl_data/sft/v1/eval_sets NLTK_DATA=/fsx_0/user/lyw/projects/envs/nltk_data
|
| 10 |
+
cd /fsx_0/user/lyw/projects/arl_data/runs/s1_numina_q17/code-c1870e2
|
| 11 |
+
[ -e /fsx_0/user/lyw/projects/arl_data/runs/s1_numina_q17/asft_lr1e-5/final/config.json ] || { echo "no final model"; exit 1; }
|
| 12 |
+
PYTHONPATH=/fsx_0/user/lyw/projects/arl_data/runs/s1_numina_q17/code-c1870e2:/fsx_0/user/lyw/projects/envs/shim:/fsx_0/user/lyw/projects/envs/vllm32-py312 python3 -s -m arl.eval.generate --model /fsx_0/user/lyw/projects/arl_data/runs/s1_numina_q17/asft_lr1e-5/final --out /fsx_0/user/lyw/projects/arl_data/runs/s1_numina_q17/asft_lr1e-5/eval --bench math500:4,minerva:4,olympiadbench:4,amc23:16,aime24:16,aime25:16,gsm8k:2 --max_tokens 4096 --template plain
|
| 13 |
+
PYTHONPATH=/fsx_0/user/lyw/projects/arl_data/runs/s1_numina_q17/code-c1870e2:/fsx_0/user/lyw/projects/envs/arl-py312:/fsx_0/user/lyw/projects/envs/awe-py312:/fsx_0/user/lyw/projects/envs/arl-eval-py312 python3 -s -m arl.eval.grade --gen_dir /fsx_0/user/lyw/projects/arl_data/runs/s1_numina_q17/asft_lr1e-5/eval
|
runs/s1_numina_q17/asft_lr1e-5/eval/aime24.jsonl
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
runs/s1_numina_q17/asft_lr1e-5/eval/aime25.jsonl
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
runs/s1_numina_q17/asft_lr1e-5/eval/amc23.jsonl
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
runs/s1_numina_q17/asft_lr1e-5/eval/gsm8k.jsonl
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
runs/s1_numina_q17/asft_lr1e-5/eval/math500.jsonl
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
runs/s1_numina_q17/asft_lr1e-5/eval/minerva.jsonl
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
runs/s1_numina_q17/asft_lr1e-5/eval/olympiadbench.jsonl
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
runs/s1_numina_q17/asft_lr1e-5/eval/scores.json
ADDED
|
@@ -0,0 +1,63 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"aime24": {
|
| 3 |
+
"pass1": 0.00625,
|
| 4 |
+
"n": 16,
|
| 5 |
+
"kind": "math",
|
| 6 |
+
"mean_tokens": 1361.7583333333334,
|
| 7 |
+
"truncated": 0.21458333333333332,
|
| 8 |
+
"pass4": 0.025,
|
| 9 |
+
"pass8": 0.05,
|
| 10 |
+
"pass16": 0.1
|
| 11 |
+
},
|
| 12 |
+
"aime25": {
|
| 13 |
+
"pass1": 0.014583333333333334,
|
| 14 |
+
"n": 16,
|
| 15 |
+
"kind": "math",
|
| 16 |
+
"mean_tokens": 1214.9625,
|
| 17 |
+
"truncated": 0.16041666666666668,
|
| 18 |
+
"pass4": 0.055,
|
| 19 |
+
"pass8": 0.10111111111111111,
|
| 20 |
+
"pass16": 0.16666666666666666
|
| 21 |
+
},
|
| 22 |
+
"amc23": {
|
| 23 |
+
"pass1": 0.29375,
|
| 24 |
+
"n": 16,
|
| 25 |
+
"kind": "math",
|
| 26 |
+
"mean_tokens": 759.9421875,
|
| 27 |
+
"truncated": 0.0640625,
|
| 28 |
+
"pass4": 0.4777472527472527,
|
| 29 |
+
"pass8": 0.5909615384615385,
|
| 30 |
+
"pass16": 0.7
|
| 31 |
+
},
|
| 32 |
+
"gsm8k": {
|
| 33 |
+
"pass1": 0.7581501137225171,
|
| 34 |
+
"n": 2,
|
| 35 |
+
"kind": "math",
|
| 36 |
+
"mean_tokens": 218.17172100075814,
|
| 37 |
+
"truncated": 0.00265352539802881
|
| 38 |
+
},
|
| 39 |
+
"math500": {
|
| 40 |
+
"pass1": 0.5565,
|
| 41 |
+
"n": 4,
|
| 42 |
+
"kind": "math",
|
| 43 |
+
"mean_tokens": 508.8855,
|
| 44 |
+
"truncated": 0.0385,
|
| 45 |
+
"pass4": 0.712
|
| 46 |
+
},
|
| 47 |
+
"minerva": {
|
| 48 |
+
"pass1": 0.1792279411764706,
|
| 49 |
+
"n": 4,
|
| 50 |
+
"kind": "math",
|
| 51 |
+
"mean_tokens": 787.9926470588235,
|
| 52 |
+
"truncated": 0.07996323529411764,
|
| 53 |
+
"pass4": 0.3272058823529412
|
| 54 |
+
},
|
| 55 |
+
"olympiadbench": {
|
| 56 |
+
"pass1": 0.23516320474777447,
|
| 57 |
+
"n": 4,
|
| 58 |
+
"kind": "math",
|
| 59 |
+
"mean_tokens": 907.3056379821959,
|
| 60 |
+
"truncated": 0.08160237388724036,
|
| 61 |
+
"pass4": 0.37833827893175076
|
| 62 |
+
}
|
| 63 |
+
}
|
runs/s1_numina_q17/asft_lr1e-5/final/chat_template.jinja
ADDED
|
@@ -0,0 +1,85 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{%- if tools %}
|
| 2 |
+
{{- '<|im_start|>system\n' }}
|
| 3 |
+
{%- if messages[0].role == 'system' %}
|
| 4 |
+
{{- messages[0].content + '\n\n' }}
|
| 5 |
+
{%- endif %}
|
| 6 |
+
{{- "# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>" }}
|
| 7 |
+
{%- for tool in tools %}
|
| 8 |
+
{{- "\n" }}
|
| 9 |
+
{{- tool | tojson }}
|
| 10 |
+
{%- endfor %}
|
| 11 |
+
{{- "\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call><|im_end|>\n" }}
|
| 12 |
+
{%- else %}
|
| 13 |
+
{%- if messages[0].role == 'system' %}
|
| 14 |
+
{{- '<|im_start|>system\n' + messages[0].content + '<|im_end|>\n' }}
|
| 15 |
+
{%- endif %}
|
| 16 |
+
{%- endif %}
|
| 17 |
+
{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
|
| 18 |
+
{%- for message in messages[::-1] %}
|
| 19 |
+
{%- set index = (messages|length - 1) - loop.index0 %}
|
| 20 |
+
{%- if ns.multi_step_tool and message.role == "user" and not(message.content.startswith('<tool_response>') and message.content.endswith('</tool_response>')) %}
|
| 21 |
+
{%- set ns.multi_step_tool = false %}
|
| 22 |
+
{%- set ns.last_query_index = index %}
|
| 23 |
+
{%- endif %}
|
| 24 |
+
{%- endfor %}
|
| 25 |
+
{%- for message in messages %}
|
| 26 |
+
{%- if (message.role == "user") or (message.role == "system" and not loop.first) %}
|
| 27 |
+
{{- '<|im_start|>' + message.role + '\n' + message.content + '<|im_end|>' + '\n' }}
|
| 28 |
+
{%- elif message.role == "assistant" %}
|
| 29 |
+
{%- set content = message.content %}
|
| 30 |
+
{%- set reasoning_content = '' %}
|
| 31 |
+
{%- if message.reasoning_content is defined and message.reasoning_content is not none %}
|
| 32 |
+
{%- set reasoning_content = message.reasoning_content %}
|
| 33 |
+
{%- else %}
|
| 34 |
+
{%- if '</think>' in message.content %}
|
| 35 |
+
{%- set content = message.content.split('</think>')[-1].lstrip('\n') %}
|
| 36 |
+
{%- set reasoning_content = message.content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
|
| 37 |
+
{%- endif %}
|
| 38 |
+
{%- endif %}
|
| 39 |
+
{%- if loop.index0 > ns.last_query_index %}
|
| 40 |
+
{%- if loop.last or (not loop.last and reasoning_content) %}
|
| 41 |
+
{{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content.strip('\n') + '\n</think>\n\n' + content.lstrip('\n') }}
|
| 42 |
+
{%- else %}
|
| 43 |
+
{{- '<|im_start|>' + message.role + '\n' + content }}
|
| 44 |
+
{%- endif %}
|
| 45 |
+
{%- else %}
|
| 46 |
+
{{- '<|im_start|>' + message.role + '\n' + content }}
|
| 47 |
+
{%- endif %}
|
| 48 |
+
{%- if message.tool_calls %}
|
| 49 |
+
{%- for tool_call in message.tool_calls %}
|
| 50 |
+
{%- if (loop.first and content) or (not loop.first) %}
|
| 51 |
+
{{- '\n' }}
|
| 52 |
+
{%- endif %}
|
| 53 |
+
{%- if tool_call.function %}
|
| 54 |
+
{%- set tool_call = tool_call.function %}
|
| 55 |
+
{%- endif %}
|
| 56 |
+
{{- '<tool_call>\n{"name": "' }}
|
| 57 |
+
{{- tool_call.name }}
|
| 58 |
+
{{- '", "arguments": ' }}
|
| 59 |
+
{%- if tool_call.arguments is string %}
|
| 60 |
+
{{- tool_call.arguments }}
|
| 61 |
+
{%- else %}
|
| 62 |
+
{{- tool_call.arguments | tojson }}
|
| 63 |
+
{%- endif %}
|
| 64 |
+
{{- '}\n</tool_call>' }}
|
| 65 |
+
{%- endfor %}
|
| 66 |
+
{%- endif %}
|
| 67 |
+
{{- '<|im_end|>\n' }}
|
| 68 |
+
{%- elif message.role == "tool" %}
|
| 69 |
+
{%- if loop.first or (messages[loop.index0 - 1].role != "tool") %}
|
| 70 |
+
{{- '<|im_start|>user' }}
|
| 71 |
+
{%- endif %}
|
| 72 |
+
{{- '\n<tool_response>\n' }}
|
| 73 |
+
{{- message.content }}
|
| 74 |
+
{{- '\n</tool_response>' }}
|
| 75 |
+
{%- if loop.last or (messages[loop.index0 + 1].role != "tool") %}
|
| 76 |
+
{{- '<|im_end|>\n' }}
|
| 77 |
+
{%- endif %}
|
| 78 |
+
{%- endif %}
|
| 79 |
+
{%- endfor %}
|
| 80 |
+
{%- if add_generation_prompt %}
|
| 81 |
+
{{- '<|im_start|>assistant\n' }}
|
| 82 |
+
{%- if enable_thinking is defined and enable_thinking is false %}
|
| 83 |
+
{{- '<think>\n\n</think>\n\n' }}
|
| 84 |
+
{%- endif %}
|
| 85 |
+
{%- endif %}
|
runs/s1_numina_q17/asft_lr1e-5/final/config.json
ADDED
|
@@ -0,0 +1,63 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"architectures": [
|
| 3 |
+
"Qwen3ForCausalLM"
|
| 4 |
+
],
|
| 5 |
+
"attention_bias": false,
|
| 6 |
+
"attention_dropout": 0.0,
|
| 7 |
+
"bos_token_id": 151643,
|
| 8 |
+
"dtype": "float32",
|
| 9 |
+
"eos_token_id": 151643,
|
| 10 |
+
"head_dim": 128,
|
| 11 |
+
"hidden_act": "silu",
|
| 12 |
+
"hidden_size": 2048,
|
| 13 |
+
"initializer_range": 0.02,
|
| 14 |
+
"intermediate_size": 6144,
|
| 15 |
+
"layer_types": [
|
| 16 |
+
"full_attention",
|
| 17 |
+
"full_attention",
|
| 18 |
+
"full_attention",
|
| 19 |
+
"full_attention",
|
| 20 |
+
"full_attention",
|
| 21 |
+
"full_attention",
|
| 22 |
+
"full_attention",
|
| 23 |
+
"full_attention",
|
| 24 |
+
"full_attention",
|
| 25 |
+
"full_attention",
|
| 26 |
+
"full_attention",
|
| 27 |
+
"full_attention",
|
| 28 |
+
"full_attention",
|
| 29 |
+
"full_attention",
|
| 30 |
+
"full_attention",
|
| 31 |
+
"full_attention",
|
| 32 |
+
"full_attention",
|
| 33 |
+
"full_attention",
|
| 34 |
+
"full_attention",
|
| 35 |
+
"full_attention",
|
| 36 |
+
"full_attention",
|
| 37 |
+
"full_attention",
|
| 38 |
+
"full_attention",
|
| 39 |
+
"full_attention",
|
| 40 |
+
"full_attention",
|
| 41 |
+
"full_attention",
|
| 42 |
+
"full_attention",
|
| 43 |
+
"full_attention"
|
| 44 |
+
],
|
| 45 |
+
"max_position_embeddings": 32768,
|
| 46 |
+
"max_window_layers": 28,
|
| 47 |
+
"model_type": "qwen3",
|
| 48 |
+
"num_attention_heads": 16,
|
| 49 |
+
"num_hidden_layers": 28,
|
| 50 |
+
"num_key_value_heads": 8,
|
| 51 |
+
"pad_token_id": null,
|
| 52 |
+
"rms_norm_eps": 1e-06,
|
| 53 |
+
"rope_parameters": {
|
| 54 |
+
"rope_theta": 1000000,
|
| 55 |
+
"rope_type": "default"
|
| 56 |
+
},
|
| 57 |
+
"sliding_window": null,
|
| 58 |
+
"tie_word_embeddings": true,
|
| 59 |
+
"transformers_version": "5.20.2",
|
| 60 |
+
"use_cache": false,
|
| 61 |
+
"use_sliding_window": false,
|
| 62 |
+
"vocab_size": 151936
|
| 63 |
+
}
|
runs/s1_numina_q17/asft_lr1e-5/final/generation_config.json
ADDED
|
@@ -0,0 +1,7 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"bos_token_id": 151643,
|
| 3 |
+
"do_sample": false,
|
| 4 |
+
"eos_token_id": 151643,
|
| 5 |
+
"max_new_tokens": 2048,
|
| 6 |
+
"transformers_version": "5.20.2"
|
| 7 |
+
}
|
runs/s1_numina_q17/asft_lr1e-5/final/tokenizer.json
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:be75606093db2094d7cd20f3c2f385c212750648bd6ea4fb2bf507a6a4c55506
|
| 3 |
+
size 11422650
|
runs/s1_numina_q17/asft_lr1e-5/final/tokenizer_config.json
ADDED
|
@@ -0,0 +1,30 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"add_prefix_space": false,
|
| 3 |
+
"backend": "tokenizers",
|
| 4 |
+
"bos_token": null,
|
| 5 |
+
"clean_up_tokenization_spaces": false,
|
| 6 |
+
"eos_token": "<|endoftext|>",
|
| 7 |
+
"errors": "replace",
|
| 8 |
+
"extra_special_tokens": [
|
| 9 |
+
"<|im_start|>",
|
| 10 |
+
"<|im_end|>",
|
| 11 |
+
"<|object_ref_start|>",
|
| 12 |
+
"<|object_ref_end|>",
|
| 13 |
+
"<|box_start|>",
|
| 14 |
+
"<|box_end|>",
|
| 15 |
+
"<|quad_start|>",
|
| 16 |
+
"<|quad_end|>",
|
| 17 |
+
"<|vision_start|>",
|
| 18 |
+
"<|vision_end|>",
|
| 19 |
+
"<|vision_pad|>",
|
| 20 |
+
"<|image_pad|>",
|
| 21 |
+
"<|video_pad|>"
|
| 22 |
+
],
|
| 23 |
+
"is_local": true,
|
| 24 |
+
"local_files_only": true,
|
| 25 |
+
"model_max_length": 131072,
|
| 26 |
+
"pad_token": "<|endoftext|>",
|
| 27 |
+
"split_special_tokens": false,
|
| 28 |
+
"tokenizer_class": "Qwen2Tokenizer",
|
| 29 |
+
"unk_token": null
|
| 30 |
+
}
|
runs/s1_numina_q17/asft_lr1e-5/train.log
ADDED
|
@@ -0,0 +1,121 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
|
| 2 |
+
*****************************************
|
| 3 |
+
Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed.
|
| 4 |
+
*****************************************
|
| 5 |
+
examples 20000 (dropped 0 longer than 4096), steps/epoch 312, total 936, method asft {'lam': 0.05}
|
| 6 |
+
|
| 7 |
+
|
| 8 |
+
|
| 9 |
+
|
| 10 |
+
NCCL version 2.27.7+cuda13.0
|
| 11 |
+
[rank0]:[W1111 21:51:45.930385338 Utils.hpp:137] Warning: Environment variable TORCH_NCCL_TRACE_BUFFER_SIZE is deprecated; use TORCH_FR_BUFFER_SIZE instead (function operator())
|
| 12 |
+
[rank2]:[W1111 21:51:45.943144251 Utils.hpp:137] Warning: Environment variable TORCH_NCCL_TRACE_BUFFER_SIZE is deprecated; use TORCH_FR_BUFFER_SIZE instead (function operator())
|
| 13 |
+
[rank3]:[W1111 21:51:45.957347526 Utils.hpp:137] Warning: Environment variable TORCH_NCCL_TRACE_BUFFER_SIZE is deprecated; use TORCH_FR_BUFFER_SIZE instead (function operator())
|
| 14 |
+
[rank1]:[W1111 21:51:45.962334158 Utils.hpp:137] Warning: Environment variable TORCH_NCCL_TRACE_BUFFER_SIZE is deprecated; use TORCH_FR_BUFFER_SIZE instead (function operator())
|
| 15 |
+
|
| 16 |
+
|
| 17 |
+
|
| 18 |
+
|
| 19 |
+
{'step': 1, 'epoch': 0.003, 'loss': 0.0869249450707501, 'gnorm': 1.6092534065246582, 'lr': 4.347813841927667e-07, 'ntok': 25530, 'sec': 3, 'mem_gb': 35.3, 'nll': 0.522143334740991, 'w': 0.800987196435566, 'kl': 0.0008342181100034732}
|
| 20 |
+
{'step': 10, 'epoch': 0.032, 'loss': 0.07645117406215006, 'gnorm': 1.2988877296447754, 'lr': 2.3906309338295734e-06, 'ntok': 26790, 'sec': 20, 'mem_gb': 44.6, 'nll': 0.5311293057694102, 'w': 0.8146049132138857, 'kl': 0.0031241257095479306}
|
| 21 |
+
{'step': 20, 'epoch': 0.064, 'loss': 0.054915812421478506, 'gnorm': 0.496549129486084, 'lr': 4.560076405307406e-06, 'ntok': 30147, 'sec': 37, 'mem_gb': 44.6, 'nll': 0.6404619318464524, 'w': 0.8449389657345673, 'kl': 0.15525677111839487}
|
| 22 |
+
{'step': 30, 'epoch': 0.096, 'loss': 0.052274906056553425, 'gnorm': 0.5359945893287659, 'lr': 6.72206303118935e-06, 'ntok': 25645, 'sec': 55, 'mem_gb': 44.6, 'nll': 0.7631954937609671, 'w': 0.8500199539383896, 'kl': 0.32015995137941117}
|
| 23 |
+
{'step': 40, 'epoch': 0.128, 'loss': 0.04911098116605286, 'gnorm': 0.3338779807090759, 'lr': 8.872940037943724e-06, 'ntok': 25388, 'sec': 73, 'mem_gb': 44.6, 'nll': 0.6108501525080747, 'w': 0.859573712851544, 'kl': 0.18663348733899873}
|
| 24 |
+
{'step': 50, 'epoch': 0.16, 'loss': 0.05050832097789174, 'gnorm': 0.30983272194862366, 'lr': 9.929756110423195e-06, 'ntok': 25361, 'sec': 92, 'mem_gb': 44.8, 'nll': 0.6071092795039628, 'w': 0.856183121968771, 'kl': 0.18721946113742557}
|
| 25 |
+
{'step': 60, 'epoch': 0.192, 'loss': 0.05038423639923915, 'gnorm': 0.29168790578842163, 'lr': 9.89895326021134e-06, 'ntok': 27190, 'sec': 110, 'mem_gb': 44.8, 'nll': 0.605541947292203, 'w': 0.8562637918352335, 'kl': 0.19244571979013425}
|
| 26 |
+
{'step': 70, 'epoch': 0.224, 'loss': 0.05167431157470353, 'gnorm': 0.29360806941986084, 'lr': 9.862632043093667e-06, 'ntok': 26478, 'sec': 128, 'mem_gb': 44.8, 'nll': 0.6592856864661228, 'w': 0.8463369058746885, 'kl': 0.1963347593403014}
|
| 27 |
+
{'step': 80, 'epoch': 0.256, 'loss': 0.05076745427819646, 'gnorm': 0.32857441902160645, 'lr': 9.820833372667813e-06, 'ntok': 24050, 'sec': 146, 'mem_gb': 44.8, 'nll': 0.6136317161512475, 'w': 0.8532589981808731, 'kl': 0.19496674408783785}
|
| 28 |
+
{'step': 90, 'epoch': 0.288, 'loss': 0.05052022031107374, 'gnorm': 0.36228615045547485, 'lr': 9.77360433254273e-06, 'ntok': 23006, 'sec': 164, 'mem_gb': 44.8, 'nll': 0.6038413722235504, 'w': 0.8550294114850474, 'kl': 0.19839964897064896}
|
| 29 |
+
{'step': 100, 'epoch': 0.321, 'loss': 0.04993581712986076, 'gnorm': 0.306488037109375, 'lr': 9.720998123301924e-06, 'ntok': 27542, 'sec': 182, 'mem_gb': 44.8, 'nll': 0.6540435686860068, 'w': 0.8488452721071454, 'kl': 0.1992994506117929}
|
| 30 |
+
{'step': 110, 'epoch': 0.353, 'loss': 0.048128938029920604, 'gnorm': 0.30867767333984375, 'lr': 9.663074002576303e-06, 'ntok': 25505, 'sec': 200, 'mem_gb': 44.8, 'nll': 0.5879737505513625, 'w': 0.8578186262497549, 'kl': 0.17991754067829838}
|
| 31 |
+
{'step': 120, 'epoch': 0.385, 'loss': 0.04932013393909048, 'gnorm': 0.31235310435295105, 'lr': 9.599897218294122e-06, 'ntok': 26525, 'sec': 217, 'mem_gb': 44.8, 'nll': 0.5983383894910462, 'w': 0.8567021088595664, 'kl': 0.1925671536286522}
|
| 32 |
+
{'step': 130, 'epoch': 0.417, 'loss': 0.04930229487956872, 'gnorm': 0.36421942710876465, 'lr': 9.531538935183252e-06, 'ntok': 21929, 'sec': 236, 'mem_gb': 44.8, 'nll': 0.5784057800885813, 'w': 0.8589314520725979, 'kl': 0.1977530823908979}
|
| 33 |
+
{'step': 140, 'epoch': 0.449, 'loss': 0.051007250488523254, 'gnorm': 0.3492606580257416, 'lr': 9.458076154608515e-06, 'ntok': 24212, 'sec': 255, 'mem_gb': 44.8, 'nll': 0.6674265827637123, 'w': 0.8438903617008096, 'kl': 0.19937312770397944}
|
| 34 |
+
{'step': 150, 'epoch': 0.481, 'loss': 0.04837100720855714, 'gnorm': 0.29547688364982605, 'lr': 9.37959162783444e-06, 'ntok': 28019, 'sec': 273, 'mem_gb': 44.8, 'nll': 0.6222071128073807, 'w': 0.8537434977158357, 'kl': 0.19748692781572325}
|
| 35 |
+
{'step': 160, 'epoch': 0.513, 'loss': 0.05121278232343412, 'gnorm': 0.3024562895298004, 'lr': 9.296173762811084e-06, 'ntok': 25888, 'sec': 291, 'mem_gb': 45.1, 'nll': 0.6461467577883575, 'w': 0.8476092476485244, 'kl': 0.19999687279700054}
|
| 36 |
+
{'step': 170, 'epoch': 0.545, 'loss': 0.04500653003678426, 'gnorm': 0.28221747279167175, 'lr': 9.207916524587971e-06, 'ntok': 28205, 'sec': 310, 'mem_gb': 45.1, 'nll': 0.5555439527566034, 'w': 0.8659080144256337, 'kl': 0.18144415260481297}
|
| 37 |
+
{'step': 180, 'epoch': 0.577, 'loss': 0.05157039086195447, 'gnorm': 0.3170209527015686, 'lr': 9.114919329468283e-06, 'ntok': 24563, 'sec': 327, 'mem_gb': 45.1, 'nll': 0.6473778444077474, 'w': 0.8460040094603672, 'kl': 0.2058355642569617}
|
| 38 |
+
{'step': 190, 'epoch': 0.609, 'loss': 0.04937389498004711, 'gnorm': 0.30696117877960205, 'lr': 9.017286933022573e-06, 'ntok': 27330, 'sec': 345, 'mem_gb': 45.1, 'nll': 0.596971548550128, 'w': 0.8564865475210391, 'kl': 0.20972324654683497}
|
| 39 |
+
{'step': 200, 'epoch': 0.641, 'loss': 0.04722432254810916, 'gnorm': 0.28340932726860046, 'lr': 8.915129312088112e-06, 'ntok': 28334, 'sec': 363, 'mem_gb': 45.1, 'nll': 0.5778510634308605, 'w': 0.8614705389284958, 'kl': 0.19740927970812452}
|
| 40 |
+
{'step': 210, 'epoch': 0.673, 'loss': 0.05085136258646363, 'gnorm': 0.32734566926956177, 'lr': 8.808561540886796e-06, 'ntok': 26149, 'sec': 381, 'mem_gb': 45.1, 'nll': 0.6245464692722476, 'w': 0.8499246505602509, 'kl': 0.20275949396965467}
|
| 41 |
+
{'step': 220, 'epoch': 0.705, 'loss': 0.046241153183803764, 'gnorm': 0.29649481177330017, 'lr': 8.697703661401187e-06, 'ntok': 26970, 'sec': 398, 'mem_gb': 45.1, 'nll': 0.5508384243488135, 'w': 0.8662438878846866, 'kl': 0.18635336282211717}
|
| 42 |
+
{'step': 230, 'epoch': 0.737, 'loss': 0.046972601049980175, 'gnorm': 0.2797365188598633, 'lr': 8.58268054815466e-06, 'ntok': 29633, 'sec': 416, 'mem_gb': 45.1, 'nll': 0.6138335797506159, 'w': 0.8564593184751796, 'kl': 0.19161937150938144}
|
| 43 |
+
{'step': 240, 'epoch': 0.769, 'loss': 0.04627784934423673, 'gnorm': 0.2873570919036865, 'lr': 8.463621767547998e-06, 'ntok': 24847, 'sec': 435, 'mem_gb': 45.1, 'nll': 0.6029216775415543, 'w': 0.86005478528595, 'kl': 0.200119398688725}
|
| 44 |
+
{'step': 250, 'epoch': 0.801, 'loss': 0.053068325674462596, 'gnorm': 0.3199753165245056, 'lr': 8.340661431910888e-06, 'ntok': 25935, 'sec': 453, 'mem_gb': 45.1, 'nll': 0.7196819422353962, 'w': 0.8351310969732022, 'kl': 0.22699672860034703}
|
| 45 |
+
{'step': 260, 'epoch': 0.833, 'loss': 0.05189735597758791, 'gnorm': 0.3245716094970703, 'lr': 8.213938048432697e-06, 'ntok': 24228, 'sec': 471, 'mem_gb': 45.1, 'nll': 0.6195990789324336, 'w': 0.8482880248318062, 'kl': 0.20849687974012507}
|
| 46 |
+
{'step': 270, 'epoch': 0.865, 'loss': 0.05209419275020096, 'gnorm': 0.3078349828720093, 'lr': 8.083594363142717e-06, 'ntok': 27368, 'sec': 489, 'mem_gb': 45.3, 'nll': 0.6857735311312482, 'w': 0.8406954232589156, 'kl': 0.21620042182590068}
|
| 47 |
+
{'step': 280, 'epoch': 0.897, 'loss': 0.046213636571634825, 'gnorm': 0.3100213408470154, 'lr': 7.949777200115617e-06, 'ntok': 27402, 'sec': 505, 'mem_gb': 45.3, 'nll': 0.5656448719527771, 'w': 0.8656896194848916, 'kl': 0.1913894108721535}
|
| 48 |
+
{'step': 290, 'epoch': 0.929, 'loss': 0.04871173608668769, 'gnorm': 0.30705976486206055, 'lr': 7.812637296083235e-06, 'ntok': 27545, 'sec': 524, 'mem_gb': 46.1, 'nll': 0.5947980719050644, 'w': 0.8567397525639862, 'kl': 0.1928982332887094}
|
| 49 |
+
{'step': 300, 'epoch': 0.962, 'loss': 0.051505881432879874, 'gnorm': 0.30369871854782104, 'lr': 7.672329130639007e-06, 'ntok': 24350, 'sec': 542, 'mem_gb': 46.1, 'nll': 0.6823836948151951, 'w': 0.8412628336755646, 'kl': 0.21354249550821355}
|
| 50 |
+
{'step': 310, 'epoch': 0.994, 'loss': 0.04887367138368232, 'gnorm': 0.28973305225372314, 'lr': 7.529010752226277e-06, 'ntok': 28463, 'sec': 561, 'mem_gb': 46.1, 'nll': 0.6063566214251835, 'w': 0.8550761761980465, 'kl': 0.20082558524444016}
|
| 51 |
+
{'step': 320, 'epoch': 1.026, 'loss': 0.04694921459374749, 'gnorm': 0.29557719826698303, 'lr': 7.382843600106539e-06, 'ntok': 24906, 'sec': 578, 'mem_gb': 46.1, 'nll': 0.5869611144152012, 'w': 0.8673059139865896, 'kl': 0.25921403931908976}
|
| 52 |
+
{'step': 330, 'epoch': 1.058, 'loss': 0.04469769661702049, 'gnorm': 0.2619975209236145, 'lr': 7.2339923225081296e-06, 'ntok': 30453, 'sec': 596, 'mem_gb': 46.1, 'nll': 0.5920419375759367, 'w': 0.8694886895379765, 'kl': 0.2403342663797491}
|
| 53 |
+
{'step': 340, 'epoch': 1.09, 'loss': 0.046964208311104726, 'gnorm': 0.2862684726715088, 'lr': 7.082624591160201e-06, 'ntok': 25891, 'sec': 614, 'mem_gb': 46.1, 'nll': 0.6300273487288054, 'w': 0.8618324176354718, 'kl': 0.24691568664666294}
|
| 54 |
+
{'step': 350, 'epoch': 1.122, 'loss': 0.0433732533722146, 'gnorm': 0.26563572883605957, 'lr': 6.9289109124209276e-06, 'ntok': 29529, 'sec': 633, 'mem_gb': 46.4, 'nll': 0.5559517497629449, 'w': 0.8748866315740459, 'kl': 0.2305125198692387}
|
| 55 |
+
{'step': 360, 'epoch': 1.154, 'loss': 0.04573662390806192, 'gnorm': 0.28397566080093384, 'lr': 6.773024435212678e-06, 'ntok': 26129, 'sec': 651, 'mem_gb': 46.4, 'nll': 0.5407757170552834, 'w': 0.8734701804030005, 'kl': 0.2459196775194707}
|
| 56 |
+
{'step': 370, 'epoch': 1.186, 'loss': 0.044722015961199714, 'gnorm': 0.2755327820777893, 'lr': 6.615140755980491e-06, 'ntok': 26398, 'sec': 669, 'mem_gb': 46.4, 'nll': 0.5773221234208084, 'w': 0.869648623948784, 'kl': 0.2261409925704599}
|
| 57 |
+
{'step': 380, 'epoch': 1.218, 'loss': 0.04527906011577884, 'gnorm': 0.2894454598426819, 'lr': 6.455437720893565e-06, 'ntok': 26427, 'sec': 687, 'mem_gb': 46.4, 'nll': 0.5659982126471034, 'w': 0.8714354932928444, 'kl': 0.2441024523333144}
|
| 58 |
+
{'step': 390, 'epoch': 1.25, 'loss': 0.05000578866911086, 'gnorm': 0.278126060962677, 'lr': 6.294095225512604e-06, 'ntok': 29253, 'sec': 704, 'mem_gb': 46.4, 'nll': 0.6942353535321164, 'w': 0.8487797992940895, 'kl': 0.26457305794491165}
|
| 59 |
+
{'step': 400, 'epoch': 1.282, 'loss': 0.04335125797791259, 'gnorm': 0.2853747606277466, 'lr': 6.131295012148613e-06, 'ntok': 28291, 'sec': 721, 'mem_gb': 46.4, 'nll': 0.5655493976109893, 'w': 0.8742920948799972, 'kl': 0.22955487307138311}
|
| 60 |
+
{'step': 410, 'epoch': 1.314, 'loss': 0.042886677663480616, 'gnorm': 0.2933399975299835, 'lr': 5.9672204651414564e-06, 'ntok': 25456, 'sec': 738, 'mem_gb': 46.4, 'nll': 0.5167372899561989, 'w': 0.8815629480770741, 'kl': 0.2383377775315741}
|
| 61 |
+
{'step': 420, 'epoch': 1.346, 'loss': 0.04329177058400614, 'gnorm': 0.2508086860179901, 'lr': 5.8020564042888015e-06, 'ntok': 31491, 'sec': 756, 'mem_gb': 46.4, 'nll': 0.5614648580149884, 'w': 0.8749929915491728, 'kl': 0.22705361874225968}
|
| 62 |
+
{'step': 430, 'epoch': 1.378, 'loss': 0.04746842057737585, 'gnorm': 0.3095942735671997, 'lr': 5.6359888766580555e-06, 'ntok': 23360, 'sec': 774, 'mem_gb': 46.4, 'nll': 0.6270307148972603, 'w': 0.8616621461633134, 'kl': 0.24966349144504496}
|
| 63 |
+
{'step': 440, 'epoch': 1.41, 'loss': 0.04302719944859888, 'gnorm': 0.2884642779827118, 'lr': 5.469204947015897e-06, 'ntok': 26461, 'sec': 793, 'mem_gb': 46.4, 'nll': 0.5537861144703526, 'w': 0.8761159554391369, 'kl': 0.23856634675796456}
|
| 64 |
+
{'step': 450, 'epoch': 1.442, 'loss': 0.04557458274522478, 'gnorm': 0.3249262571334839, 'lr': 5.301892487111431e-06, 'ntok': 23690, 'sec': 811, 'mem_gb': 46.4, 'nll': 0.5576961287067328, 'w': 0.8730866135500212, 'kl': 0.24789779360357747}
|
| 65 |
+
{'step': 460, 'epoch': 1.474, 'loss': 0.04287392286701235, 'gnorm': 0.26531970500946045, 'lr': 5.1342399640503074e-06, 'ntok': 26674, 'sec': 828, 'mem_gb': 46.4, 'nll': 0.5496729976404926, 'w': 0.8756688102178151, 'kl': 0.22415327484160605}
|
| 66 |
+
{'step': 470, 'epoch': 1.506, 'loss': 0.040965230561366806, 'gnorm': 0.28454625606536865, 'lr': 4.9664362279982205e-06, 'ntok': 25812, 'sec': 847, 'mem_gb': 46.4, 'nll': 0.49713054369091897, 'w': 0.8861523044029909, 'kl': 0.22301604813168294}
|
| 67 |
+
{'step': 480, 'epoch': 1.538, 'loss': 0.050176814435056345, 'gnorm': 0.2888350188732147, 'lr': 4.798670299452926e-06, 'ntok': 26267, 'sec': 864, 'mem_gb': 46.4, 'nll': 0.6715781684242585, 'w': 0.848863132090646, 'kl': 0.2611861708012906}
|
| 68 |
+
{'step': 490, 'epoch': 1.571, 'loss': 0.047054003874546856, 'gnorm': 0.2758960425853729, 'lr': 4.631131156324359e-06, 'ntok': 27723, 'sec': 882, 'mem_gb': 46.4, 'nll': 0.6060231323044043, 'w': 0.8621516317678461, 'kl': 0.24877587117420552}
|
| 69 |
+
{'step': 500, 'epoch': 1.603, 'loss': 0.04684284895397742, 'gnorm': 0.3116913437843323, 'lr': 4.464007521062761e-06, 'ntok': 23316, 'sec': 900, 'mem_gb': 46.4, 'nll': 0.5900026152239878, 'w': 0.8662415964359238, 'kl': 0.2505452432074541}
|
| 70 |
+
{'step': 510, 'epoch': 1.635, 'loss': 0.049200274997611315, 'gnorm': 0.3321226239204407, 'lr': 4.297487648074538e-06, 'ntok': 26165, 'sec': 918, 'mem_gb': 46.4, 'nll': 0.6194836255493981, 'w': 0.8601253463596408, 'kl': 0.2613011029763042}
|
| 71 |
+
{'step': 520, 'epoch': 1.667, 'loss': 0.04566161185377681, 'gnorm': 0.2773717939853668, 'lr': 4.131759111665349e-06, 'ntok': 28536, 'sec': 936, 'mem_gb': 46.4, 'nll': 0.6185813670670381, 'w': 0.8630607226398234, 'kl': 0.24485672292236121}
|
| 72 |
+
{'step': 530, 'epoch': 1.699, 'loss': 0.04423050962825505, 'gnorm': 0.2713128328323364, 'lr': 3.9670085947493e-06, 'ntok': 28763, 'sec': 954, 'mem_gb': 46.4, 'nll': 0.6106256573114766, 'w': 0.8665904873448528, 'kl': 0.23591722044705873}
|
| 73 |
+
{'step': 540, 'epoch': 1.731, 'loss': 0.042339646955636294, 'gnorm': 0.28177711367607117, 'lr': 3.803421678562213e-06, 'ntok': 27283, 'sec': 971, 'mem_gb': 46.4, 'nll': 0.5237363783079206, 'w': 0.8802472292728073, 'kl': 0.2299480753525547}
|
| 74 |
+
{'step': 550, 'epoch': 1.763, 'loss': 0.043328484524194126, 'gnorm': 0.2702811658382416, 'lr': 3.6411826336158785e-06, 'ntok': 27858, 'sec': 989, 'mem_gb': 46.4, 'nll': 0.5667061105651877, 'w': 0.8747239768423792, 'kl': 0.23294826104709598}
|
| 75 |
+
{'step': 560, 'epoch': 1.795, 'loss': 0.044832826733144274, 'gnorm': 0.2808559238910675, 'lr': 3.480474212128766e-06, 'ntok': 25455, 'sec': 1007, 'mem_gb': 46.4, 'nll': 0.6006876411805147, 'w': 0.8661617161412296, 'kl': 0.2355272900400216}
|
| 76 |
+
{'step': 570, 'epoch': 1.827, 'loss': 0.04485901905665888, 'gnorm': 0.2817330062389374, 'lr': 3.3214774421669777e-06, 'ntok': 26750, 'sec': 1024, 'mem_gb': 46.4, 'nll': 0.5628308995327103, 'w': 0.8731412821261683, 'kl': 0.2492738390771028}
|
| 77 |
+
{'step': 580, 'epoch': 1.859, 'loss': 0.04527445244300513, 'gnorm': 0.2874723970890045, 'lr': 3.1643714237273628e-06, 'ntok': 26691, 'sec': 1042, 'mem_gb': 46.4, 'nll': 0.569718242455322, 'w': 0.8718350169532801, 'kl': 0.25287690973197896}
|
| 78 |
+
{'step': 590, 'epoch': 1.891, 'loss': 0.04502802309782609, 'gnorm': 0.28341302275657654, 'lr': 3.0093331269924954e-06, 'ntok': 27324, 'sec': 1060, 'mem_gb': 46.4, 'nll': 0.6279685727291027, 'w': 0.862925564522032, 'kl': 0.24682726194805849}
|
| 79 |
+
{'step': 600, 'epoch': 1.923, 'loss': 0.04225378922544558, 'gnorm': 0.29076245427131653, 'lr': 2.8565371929847286e-06, 'ntok': 26370, 'sec': 1079, 'mem_gb': 46.4, 'nll': 0.5195111040007584, 'w': 0.8816988410125142, 'kl': 0.2236919913727721}
|
| 80 |
+
{'step': 610, 'epoch': 1.955, 'loss': 0.04144875598429783, 'gnorm': 0.2883087396621704, 'lr': 2.7061557368439294e-06, 'ntok': 24093, 'sec': 1097, 'mem_gb': 46.4, 'nll': 0.4849832387778608, 'w': 0.8865335140134064, 'kl': 0.22342029777840036}
|
| 81 |
+
{'step': 620, 'epoch': 1.987, 'loss': 0.04314557267670375, 'gnorm': 0.2870231568813324, 'lr': 2.5583581539504464e-06, 'ntok': 26317, 'sec': 1115, 'mem_gb': 46.4, 'nll': 0.5261635035100886, 'w': 0.8804695218394953, 'kl': 0.24139903626933162}
|
| 82 |
+
{'step': 630, 'epoch': 2.019, 'loss': 0.04066903871025703, 'gnorm': 0.25142988562583923, 'lr': 2.4133109291117156e-06, 'ntok': 26349, 'sec': 1132, 'mem_gb': 46.4, 'nll': 0.5410024307136894, 'w': 0.8832499128287601, 'kl': 0.25014888047221906}
|
| 83 |
+
{'step': 640, 'epoch': 2.051, 'loss': 0.04227218270493541, 'gnorm': 0.23311755061149597, 'lr': 2.2711774490274767e-06, 'ntok': 28602, 'sec': 1150, 'mem_gb': 46.4, 'nll': 0.5947839229992657, 'w': 0.8777530308195232, 'kl': 0.28299935865498915}
|
| 84 |
+
{'step': 650, 'epoch': 2.083, 'loss': 0.04268250123863464, 'gnorm': 0.23196960985660553, 'lr': 2.132117818244771e-06, 'ntok': 29613, 'sec': 1169, 'mem_gb': 46.4, 'nll': 0.5980635220384628, 'w': 0.8751013728134603, 'kl': 0.2740291492017864}
|
| 85 |
+
{'step': 660, 'epoch': 2.115, 'loss': 0.04114492346914222, 'gnorm': 0.2708017826080322, 'lr': 1.996288678810105e-06, 'ntok': 25974, 'sec': 1187, 'mem_gb': 46.4, 'nll': 0.5325450901622777, 'w': 0.8859847678207053, 'kl': 0.2642094844243282}
|
| 86 |
+
{'step': 670, 'epoch': 2.147, 'loss': 0.04161535796683764, 'gnorm': 0.23944348096847534, 'lr': 1.8638430338219199e-06, 'ntok': 26636, 'sec': 1205, 'mem_gb': 46.4, 'nll': 0.5284975060186402, 'w': 0.8850079691254318, 'kl': 0.2658965830724865}
|
| 87 |
+
{'step': 680, 'epoch': 2.179, 'loss': 0.043215690039735695, 'gnorm': 0.22967204451560974, 'lr': 1.7349300750820758e-06, 'ntok': 29303, 'sec': 1222, 'mem_gb': 46.4, 'nll': 0.6243845286745384, 'w': 0.8697482313031089, 'kl': 0.2657402092469201}
|
| 88 |
+
{'step': 690, 'epoch': 2.212, 'loss': 0.04326565341190727, 'gnorm': 0.26724907755851746, 'lr': 1.6096950150405454e-06, 'ntok': 23957, 'sec': 1240, 'mem_gb': 46.4, 'nll': 0.5717127953207831, 'w': 0.8789802758588304, 'kl': 0.27585206372354637}
|
| 89 |
+
{'step': 700, 'epoch': 2.244, 'loss': 0.04268083955470402, 'gnorm': 0.262351393699646, 'lr': 1.4882789232226124e-06, 'ntok': 26015, 'sec': 1257, 'mem_gb': 46.4, 'nll': 0.5476288844777052, 'w': 0.8823384645877378, 'kl': 0.27192462581683646}
|
| 90 |
+
{'step': 710, 'epoch': 2.276, 'loss': 0.041071568522307283, 'gnorm': 0.238427996635437, 'lr': 1.3708185673227896e-06, 'ntok': 26516, 'sec': 1275, 'mem_gb': 46.4, 'nll': 0.5283958123161487, 'w': 0.8861970278558229, 'kl': 0.26880426042884487}
|
| 91 |
+
{'step': 720, 'epoch': 2.308, 'loss': 0.04219270299692623, 'gnorm': 0.2505238652229309, 'lr': 1.257446259144494e-06, 'ntok': 25925, 'sec': 1291, 'mem_gb': 46.4, 'nll': 0.5400798200940212, 'w': 0.8835752169720347, 'kl': 0.27520314760125364}
|
| 92 |
+
{'step': 730, 'epoch': 2.34, 'loss': 0.03766030950799093, 'gnorm': 0.24006663262844086, 'lr': 1.1482897055590275e-06, 'ntok': 27234, 'sec': 1310, 'mem_gb': 46.4, 'nll': 0.47836193671054567, 'w': 0.8977597798248513, 'kl': 0.25140146436325}
|
| 93 |
+
{'step': 740, 'epoch': 2.372, 'loss': 0.04391672122827788, 'gnorm': 0.28070691227912903, 'lr': 1.0434718646516917e-06, 'ntok': 23241, 'sec': 1328, 'mem_gb': 46.4, 'nll': 0.5634963534271331, 'w': 0.8772142307775053, 'kl': 0.2764013278997784}
|
| 94 |
+
{'step': 750, 'epoch': 2.404, 'loss': 0.039250085292024374, 'gnorm': 0.23895694315433502, 'lr': 9.431108072171346e-07, 'ntok': 28189, 'sec': 1345, 'mem_gb': 46.4, 'nll': 0.5171567574563659, 'w': 0.8900041904643655, 'kl': 0.2514202057429316}
|
| 95 |
+
{'step': 760, 'epoch': 2.436, 'loss': 0.037181874852455145, 'gnorm': 0.23781812191009521, 'lr': 8.473195837599419e-07, 'ntok': 26475, 'sec': 1363, 'mem_gb': 46.4, 'nll': 0.4533918348677998, 'w': 0.9011493006373937, 'kl': 0.2476263537240321}
|
| 96 |
+
{'step': 770, 'epoch': 2.468, 'loss': 0.04311137956983861, 'gnorm': 0.2705056369304657, 'lr': 7.562060971502383e-07, 'ntok': 27449, 'sec': 1380, 'mem_gb': 46.4, 'nll': 0.5737812909850997, 'w': 0.8782233073153849, 'kl': 0.28440764932807205}
|
| 97 |
+
{'step': 780, 'epoch': 2.5, 'loss': 0.039903676185575276, 'gnorm': 0.23965997993946075, 'lr': 6.698729810778065e-07, 'ntok': 29225, 'sec': 1398, 'mem_gb': 46.4, 'nll': 0.5196494733746793, 'w': 0.888994867408041, 'kl': 0.2568035246471343}
|
| 98 |
+
{'step': 790, 'epoch': 2.532, 'loss': 0.03826080892941618, 'gnorm': 0.23798969388008118, 'lr': 5.884174844416102e-07, 'ntok': 26952, 'sec': 1416, 'mem_gb': 46.4, 'nll': 0.5019454435199614, 'w': 0.8933141702146409, 'kl': 0.2456974291100846}
|
| 99 |
+
{'step': 800, 'epoch': 2.564, 'loss': 0.04202510223634234, 'gnorm': 0.2610495388507843, 'lr': 5.119313618049309e-07, 'ntok': 26098, 'sec': 1433, 'mem_gb': 46.4, 'nll': 0.5720556292627788, 'w': 0.878415012644647, 'kl': 0.26266685530859646}
|
| 100 |
+
{'step': 810, 'epoch': 2.596, 'loss': 0.0419120811940635, 'gnorm': 0.2524871230125427, 'lr': 4.405007700395497e-07, 'ntok': 29386, 'sec': 1450, 'mem_gb': 46.4, 'nll': 0.5647102095810931, 'w': 0.8797475572976928, 'kl': 0.2635017425655499}
|
| 101 |
+
{'step': 820, 'epoch': 2.628, 'loss': 0.04537449408680116, 'gnorm': 0.24934916198253632, 'lr': 3.742061712753825e-07, 'ntok': 31570, 'sec': 1469, 'mem_gb': 46.4, 'nll': 0.6478412307966424, 'w': 0.8638338414634147, 'kl': 0.2813230643213494}
|
| 102 |
+
{'step': 830, 'epoch': 2.66, 'loss': 0.04337035496556217, 'gnorm': 0.274399071931839, 'lr': 3.131222422648744e-07, 'ntok': 26860, 'sec': 1486, 'mem_gb': 46.4, 'nll': 0.5804544251442666, 'w': 0.8789059591399851, 'kl': 0.2889261957255212}
|
| 103 |
+
{'step': 840, 'epoch': 2.692, 'loss': 0.043290276082191013, 'gnorm': 0.2675422430038452, 'lr': 2.573177902642726e-07, 'ntok': 24548, 'sec': 1503, 'mem_gb': 46.4, 'nll': 0.5587324689765969, 'w': 0.8790747655104285, 'kl': 0.2795177830805972}
|
| 104 |
+
{'step': 850, 'epoch': 2.724, 'loss': 0.03974937668343767, 'gnorm': 0.24015042185783386, 'lr': 2.068556755265272e-07, 'ntok': 25478, 'sec': 1520, 'mem_gb': 46.4, 'nll': 0.523697298281851, 'w': 0.8893687078067353, 'kl': 0.27003697123498704}
|
| 105 |
+
{'step': 860, 'epoch': 2.756, 'loss': 0.041953977112899714, 'gnorm': 0.25849777460098267, 'lr': 1.6179274049310966e-07, 'ntok': 27551, 'sec': 1538, 'mem_gb': 46.4, 'nll': 0.5350083353939058, 'w': 0.8864492214438677, 'kl': 0.28431728742831475}
|
| 106 |
+
{'step': 870, 'epoch': 2.788, 'loss': 0.04466982187258415, 'gnorm': 0.29349109530448914, 'lr': 1.2217974576453072e-07, 'ntok': 21020, 'sec': 1556, 'mem_gb': 46.4, 'nll': 0.5745291872323978, 'w': 0.8764434727343007, 'kl': 0.2946213799357755}
|
| 107 |
+
{'step': 880, 'epoch': 2.821, 'loss': 0.04210429563783459, 'gnorm': 0.2540396749973297, 'lr': 8.80613129216762e-08, 'ntok': 24470, 'sec': 1573, 'mem_gb': 46.4, 'nll': 0.5401550367797303, 'w': 0.883850761136085, 'kl': 0.27755967527201675}
|
| 108 |
+
{'step': 890, 'epoch': 2.853, 'loss': 0.03799059834432152, 'gnorm': 0.2356002777814865, 'lr': 5.947587426236079e-08, 'ntok': 26847, 'sec': 1590, 'mem_gb': 46.4, 'nll': 0.4694109356957947, 'w': 0.8987395588892614, 'kl': 0.25466894505787424}
|
| 109 |
+
{'step': 900, 'epoch': 2.885, 'loss': 0.042558943913404396, 'gnorm': 0.2650890052318573, 'lr': 3.645562950973014e-08, 'ntok': 25072, 'sec': 1606, 'mem_gb': 46.4, 'nll': 0.532487529664566, 'w': 0.8842161138421346, 'kl': 0.2751308068699645}
|
| 110 |
+
{'step': 910, 'epoch': 2.917, 'loss': 0.03756796753405709, 'gnorm': 0.23625795543193817, 'lr': 1.9026509541272276e-08, 'ntok': 29234, 'sec': 1624, 'mem_gb': 46.4, 'nll': 0.4718315801485428, 'w': 0.8974278665252788, 'kl': 0.24951185237009987}
|
| 111 |
+
{'step': 920, 'epoch': 2.949, 'loss': 0.038428223104625805, 'gnorm': 0.23640362918376923, 'lr': 7.2081471792911914e-09, 'ntok': 27225, 'sec': 1643, 'mem_gb': 46.4, 'nll': 0.49215485824150595, 'w': 0.8942308023415978, 'kl': 0.2575388652146465}
|
| 112 |
+
{'step': 930, 'epoch': 2.981, 'loss': 0.04268174433748102, 'gnorm': 0.27044832706451416, 'lr': 1.0138550757493592e-09, 'ntok': 25687, 'sec': 1662, 'mem_gb': 46.4, 'nll': 0.5745237285640596, 'w': 0.8769259803110523, 'kl': 0.27234327249678825}
|
| 113 |
+
|
| 114 |
+
saved /fsx_0/user/lyw/projects/arl_data/runs/s1_numina_q17/asft_lr1e-5/final
|
| 115 |
+
/usr/local/lib/python3.12/site-packages/torch/distributed/distributed_c10d.py:4876: UserWarning: barrier(): using the device under current context. You can specify `device_id` in `init_process_group` to mute this warning.
|
| 116 |
+
warnings.warn( # warn only once
|
| 117 |
+
[W1111 22:19:48.280895111 AllocatorConfig.cpp:28] Warning: PYTORCH_CUDA_ALLOC_CONF is deprecated, use PYTORCH_ALLOC_CONF instead (function operator())
|
| 118 |
+
[W1111 22:19:48.280895051 AllocatorConfig.cpp:28] Warning: PYTORCH_CUDA_ALLOC_CONF is deprecated, use PYTORCH_ALLOC_CONF instead (function operator())
|
| 119 |
+
[W1111 22:19:48.281642942 AllocatorConfig.cpp:28] Warning: PYTORCH_CUDA_ALLOC_CONF is deprecated, use PYTORCH_ALLOC_CONF instead (function operator())
|
| 120 |
+
[W1111 22:19:48.282317121 AllocatorConfig.cpp:28] Warning: PYTORCH_CUDA_ALLOC_CONF is deprecated, use PYTORCH_ALLOC_CONF instead (function operator())
|
| 121 |
+
[W1111 22:19:52.520254653 AllocatorConfig.cpp:28] Warning: PYTORCH_CUDA_ALLOC_CONF is deprecated, use PYTORCH_ALLOC_CONF instead (function operator())
|
runs/s1_numina_q17/asft_lr1e-5/train.sbatch
ADDED
|
@@ -0,0 +1,12 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
#!/bin/bash
|
| 2 |
+
#SBATCH --job-name=arl-s1_numina_q17-asft_lr1e-5
|
| 3 |
+
#SBATCH --gres=gpu:4
|
| 4 |
+
#SBATCH --cpus-per-task=32
|
| 5 |
+
#SBATCH --mem=600G
|
| 6 |
+
#SBATCH --time=3:00:00
|
| 7 |
+
#SBATCH --requeue
|
| 8 |
+
#SBATCH --output=/fsx_0/user/lyw/projects/arl_data/runs/s1_numina_q17/asft_lr1e-5/train.log
|
| 9 |
+
export HF_HUB_OFFLINE=1 TRANSFORMERS_OFFLINE=1 PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True
|
| 10 |
+
cd /fsx_0/user/lyw/projects/arl_data/runs/s1_numina_q17/code-c1870e2
|
| 11 |
+
[ -e /fsx_0/user/lyw/projects/arl_data/runs/s1_numina_q17/asft_lr1e-5/final/config.json ] && exit 0
|
| 12 |
+
PYTHONPATH=/fsx_0/user/lyw/projects/arl_data/runs/s1_numina_q17/code-c1870e2:/fsx_0/user/lyw/projects/envs/arl-py312:/fsx_0/user/lyw/projects/envs/awe-py312:/fsx_0/user/lyw/projects/envs/arl-eval-py312 python3 -s -m torch.distributed.run --nproc_per_node 4 --master_port $((29500 + RANDOM % 500)) -m arl.train --model /fsx_0/user/lyw/projects/arl_data/models/Qwen3-1.7B-Base --data /fsx_0/user/lyw/projects/arl_data/sft/v1/numina_cot_20k.jsonl --out /fsx_0/user/lyw/projects/arl_data/runs/s1_numina_q17/asft_lr1e-5 --method asft --method_args '{}' --lr 1e-5 --epochs 3 --seqs_per_step 64 --max_len 4096 --micro_tokens 16384 --template plain --fisher_cache_dir /fsx_0/user/lyw/projects/arl_data/fisher
|
runs/s1_numina_q17/asft_lr1e-5/train_log.jsonl
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
runs/s1_numina_q17/asft_lr1e-6/args.json
ADDED
|
@@ -0,0 +1,30 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"model": "/fsx_0/user/lyw/projects/arl_data/models/Qwen3-1.7B-Base",
|
| 3 |
+
"data": "/fsx_0/user/lyw/projects/arl_data/sft/v1/numina_cot_20k.jsonl",
|
| 4 |
+
"out": "/fsx_0/user/lyw/projects/arl_data/runs/s1_numina_q17/asft_lr1e-6",
|
| 5 |
+
"method": "asft",
|
| 6 |
+
"method_args": "{}",
|
| 7 |
+
"template": "plain",
|
| 8 |
+
"max_len": 4096,
|
| 9 |
+
"limit": 0,
|
| 10 |
+
"lr": 1e-06,
|
| 11 |
+
"epochs": 3.0,
|
| 12 |
+
"seqs_per_step": 64,
|
| 13 |
+
"micro_tokens": 16384,
|
| 14 |
+
"chunk": 2048,
|
| 15 |
+
"warmup": 0.05,
|
| 16 |
+
"clip": 1.0,
|
| 17 |
+
"seed": 0,
|
| 18 |
+
"ckpt_every_min": 40,
|
| 19 |
+
"stop_at": 0,
|
| 20 |
+
"save_at": "",
|
| 21 |
+
"zero1": false,
|
| 22 |
+
"no_save": false,
|
| 23 |
+
"fisher": "",
|
| 24 |
+
"fisher_gamma": 1.0,
|
| 25 |
+
"fisher_n": 1024,
|
| 26 |
+
"fisher_cache_dir": "/fsx_0/user/lyw/projects/arl_data/fisher",
|
| 27 |
+
"method_cfg": {
|
| 28 |
+
"lam": 0.05
|
| 29 |
+
}
|
| 30 |
+
}
|
runs/s1_numina_q17/asft_lr1e-6/eval.log
ADDED
|
@@ -0,0 +1,83 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
(MainProcess pid=558908) INFO 11-14 07:32:25 [api_utils.py:286] non-default args: {'dtype': 'bfloat16', 'max_model_len': 6144, 'gpu_memory_utilization': 0.85, 'disable_log_stats': True, 'model': '/fsx_0/user/lyw/projects/arl_data/runs/s1_numina_q17/asft_lr1e-6/final'}
|
| 2 |
+
(MainProcess pid=558908) INFO 11-14 07:32:25 [model.py:733] Resolved architecture: Qwen3ForCausalLM
|
| 3 |
+
(MainProcess pid=558908) INFO 11-14 07:32:25 [model.py:2495] Downcasting torch.float32 to torch.bfloat16.
|
| 4 |
+
(MainProcess pid=558908) INFO 11-14 07:32:25 [model.py:2152] Using max model len 6144
|
| 5 |
+
(MainProcess pid=558908) INFO 11-14 07:32:25 [scheduler.py:307] Chunked prefill is enabled with max_num_batched_tokens=16384.
|
| 6 |
+
(MainProcess pid=558908) INFO 11-14 07:32:25 [kernel.py:467] Final IR op priority after setting platform defaults: IrOpPriorityConfig(rms_norm=['native'], fused_add_rms_norm=['native'], gelu_and_mul_sparse=['triton', 'native'])
|
| 7 |
+
(EngineCore pid=559244) INFO 11-14 07:32:28 [core.py:133] Initializing a V1 LLM engine (v0.32.0) with config: model='/fsx_0/user/lyw/projects/arl_data/runs/s1_numina_q17/asft_lr1e-6/final', speculative_config=None, tokenizer='/fsx_0/user/lyw/projects/arl_data/runs/s1_numina_q17/asft_lr1e-6/final', skip_tokenizer_init=False, tokenizer_mode=auto, revision=None, tokenizer_revision=None, trust_remote_code=False, dtype=torch.bfloat16, max_seq_len=6144, download_dir=None, load_format=auto, tensor_parallel_size=1, pipeline_parallel_size=1, data_parallel_size=1, decode_context_parallel_size=1, dcp_comm_backend=ag_rs, disable_custom_all_reduce=False, quantization=None, quantization_config=None, enforce_eager=False, aux_output_config=AuxOutputConfig(enable_return_routed_experts=False, max_bytes=None), kv_cache_dtype=auto, device_config=cuda, structured_outputs_config=StructuredOutputsConfig(backend='auto', disable_any_whitespace=False, disable_additional_properties=False, reasoning_parser='', reasoning_parser_plugin='', enable_in_reasoning=False), observability_config=ObservabilityConfig(show_hidden_metrics_for_version=None, otlp_traces_endpoint=None, collect_detailed_traces=None, per_request_spec_decode_metrics='none', kv_cache_metrics=False, kv_cache_metrics_sample=0.01, custom_histogram_buckets=None, cudagraph_metrics=False, enable_layerwise_nvtx_tracing=False, enable_mfu_metrics=False, enable_mm_processor_stats=False, enable_logging_iteration_details=False, jit_monitor_mode='warn', jit_monitor_verbose=False), seed=0, served_model_name=/fsx_0/user/lyw/projects/arl_data/runs/s1_numina_q17/asft_lr1e-6/final, enable_prefix_caching=True, enable_chunked_prefill=True, pooler_config=None, compilation_config={'mode': <CompilationMode.VLLM_COMPILE: 3>, 'debug_dump_path': None, 'cache_dir': '', 'compile_cache_save_format': 'binary', 'backend': 'inductor', 'custom_ops': ['none'], 'ir_enable_torch_wrap': True, 'splitting_ops': ['vllm::unified_attention_with_output', 'vllm::unified_mla_attention_with_output', 'vllm::mamba_mixer2', 'vllm::mamba_mixer', 'vllm::short_conv', 'vllm::qwen4_exp_ple_short_conv', 'vllm::qwen4_exp_qsa_with_output', 'vllm::qwen4_exp_qsa_prepare_with_output', 'vllm::linear_attention', 'vllm::qwen_gdn_attention_core', 'vllm::qwen_gdn_attention_core_fused_norm_packed', 'vllm::gdn_attention_core_xpu', 'vllm::olmo_hybrid_gdn_full_forward', 'vllm::sparse_attn_indexer', 'vllm::rocm_aiter_sparse_attn_indexer', 'vllm::deepseek_v4_attention', 'vllm::hpc_rope_norm_forward', 'vllm::unified_kv_cache_update', 'vllm::unified_mla_kv_cache_update'], 'compile_mm_encoder': False, 'cudagraph_mm_encoder': False, 'encoder_cudagraph_token_budgets': [], 'encoder_cudagraph_max_vision_items_per_batch': 0, 'encoder_cudagraph_max_frames_per_batch': None, 'cudagraph_decoder_replay': True, 'decoder_replay_cudagraph_capture_sizes': [], 'decoder_replay_trim_threshold': 768, 'compile_sizes': [], 'compile_ranges_endpoints': [16384], 'inductor_compile_config': {'enable_auto_functionalized_v2': False, 'combo_kernels': True, 'benchmark_combo_kernel': True}, 'inductor_passes': {}, 'cudagraph_mode': <CUDAGraphMode.FULL_AND_PIECEWISE: (2, 1)>, 'cudagraph_num_of_warmups': 1, 'cudagraph_capture_sizes': [1, 2, 4, 8, 16, 24, 32, 40, 48, 56, 64, 72, 80, 88, 96, 104, 112, 120, 128, 136, 144, 152, 160, 168, 176, 184, 192, 200, 208, 216, 224, 232, 240, 248, 256, 272, 288, 304, 320, 336, 352, 368, 384, 400, 416, 432, 448, 464, 480, 496, 512], 'cudagraph_copy_inputs': False, 'cudagraph_specialize_lora': True, 'use_inductor_graph_partition': False, 'pass_config': {'fuse_norm_quant': False, 'fuse_act_quant': False, 'fuse_attn_quant': False, 'fuse_allreduce_rms': False, 'enable_qk_norm_rope_fusion': False, 'fuse_rope_kvcache_cat_mla': False, 'fuse_act_padding': False, 'fuse_qk_norm_rope_kvcache': False}, 'max_cudagraph_capture_size': 512, 'dynamic_shapes_config': {'type': <DynamicShapesType.BACKED: 'backed'>, 'evaluate_guards': False, 'assume_32_bit_indexing': False}, 'local_cache_dir': None, 'fast_moe_cold_start': False, 'static_all_moe_layers': []}, kernel_config=KernelConfig(ir_op_priority=IrOpPriorityConfig(rms_norm=['native'], fused_add_rms_norm=['native'], gelu_and_mul_sparse=['triton', 'native']), enable_flashinfer_autotune=True, enable_rocm_segmented_attn_autotune=False, enable_cutedsl_warmup=True, enable_jit_warmup=True, moe_backend='auto', sparse_indexer_topk_backend='auto', linear_backend='auto', linear_backend_per_quant=None)
|
| 8 |
+
(EngineCore pid=559244) INFO 11-14 07:32:31 [parallel_state.py:1906] world_size=1 rank=0 local_rank=0 distributed_init_method=file:///fsx_0/user/lyw/projects/jiaruiliu/adaptive-reasoning-loss/tmp/vllm_dist_b49e53b7a5c14f4e8dc96bb04feb441b backend=nccl
|
| 9 |
+
[rank0]:[W1114 07:32:31.220670473 Utils.hpp:137] Warning: Environment variable TORCH_NCCL_TRACE_BUFFER_SIZE is deprecated; use TORCH_FR_BUFFER_SIZE instead (function operator())
|
| 10 |
+
(EngineCore pid=559244) INFO 11-14 07:32:31 [parallel_state.py:2350] rank 0 in world size 1 is assigned as DP rank 0, PP rank 0, PCP rank 0, TP rank 0, ETP rank 0, EP rank N/A, EPLB rank N/A
|
| 11 |
+
(EngineCore pid=559244) INFO 11-14 07:32:31 [gpu_worker.py:518] Using V2 Model Runner
|
| 12 |
+
(EngineCore pid=559244) WARNING 11-14 07:32:31 [flashinfer.py:143] FlashInfer's precompiled kernels are missing or do not match flashinfer-python 0.7.0.post1 (flashinfer-cubin 0.7.0.post1, flashinfer-jit-cache not installed). FlashInfer then downloads and compiles kernels at startup, which can take several minutes, or fails to import mismatched ones. Run `vllm download-kernels` in this Python environment to install them.
|
| 13 |
+
(EngineCore pid=559244) INFO 11-14 07:32:32 [model_runner.py:392] Loading model from scratch...
|
| 14 |
+
(EngineCore pid=559244) INFO 11-14 07:32:34 [cuda.py:506] Using FLASH_ATTN attention backend out of potential backends: ['FLASH_ATTN', 'FLASHINFER', 'TRITON_ATTN', 'FLEX_ATTENTION'].
|
| 15 |
+
(EngineCore pid=559244) INFO 11-14 07:32:34 [flash_attn.py:1117] Using FlashAttention version 3
|
| 16 |
+
(EngineCore pid=559244) INFO 11-14 07:32:35 [base_loader.py:82] Initializing model took 2.712448 seconds
|
| 17 |
+
(EngineCore pid=559244) INFO 11-14 07:32:35 [weight_utils.py:1108] Filesystem type for checkpoints: LUSTRE. Checkpoint size: 3.78 GiB. Available RAM: 1264.76 GiB.
|
| 18 |
+
(EngineCore pid=559244) INFO 11-14 07:32:35 [weight_utils.py:1070] Prefetching checkpoint files into page cache started (in background, num_threads=8, block_size=16777216 bytes)
|
| 19 |
+
|
| 20 |
+
(EngineCore pid=559244) INFO 11-14 07:32:38 [weight_utils.py:1042] Prefetching checkpoint files: 10% (1/1)
|
| 21 |
+
(EngineCore pid=559244) INFO 11-14 07:32:38 [weight_utils.py:1065] Prefetching checkpoint files into page cache finished in 3.28s
|
| 22 |
+
|
| 23 |
+
|
| 24 |
+
|
| 25 |
+
(EngineCore pid=559244) INFO 11-14 07:32:39 [default_loader.py:501] Loading weights took 3.78 seconds
|
| 26 |
+
(EngineCore pid=559244) INFO 11-14 07:32:41 [model_runner.py:424] Model loading took 3.21 GiB memory and 9.506310 seconds
|
| 27 |
+
(EngineCore pid=559244) INFO 11-14 07:32:41 [topk_topp_sampler.py:124] Using FlashInfer for top-p & top-k sampling.
|
| 28 |
+
(EngineCore pid=559244) INFO 11-14 07:32:41 [torch_utils.py:295] Reducing Torch threads from 8 to 1 for serving. Set OMP_NUM_THREADS in the external environment to override.
|
| 29 |
+
(EngineCore pid=559244) INFO 11-14 07:32:41 [utils.py:347] Using LBNHC KV cache layout.
|
| 30 |
+
(EngineCore pid=559244) INFO 11-14 07:32:47 [caching.py:342] reconstructed serializable fn from standalone compile artifacts. num_artifacts=3 num_submods=29
|
| 31 |
+
(EngineCore pid=559244) INFO 11-14 07:32:47 [decorators.py:319] [backbone] Directly load AOT compilation from path /home/lyw/.cache/vllm/torch_compile_cache/torch_aot_compile/bd8ad65ee4d491747dec5f9e3daea3287b3d3eb6e525a4046755b37238fcdcc9/rank_0_0/model
|
| 32 |
+
(EngineCore pid=559244) INFO 11-14 07:32:47 [monitor.py:62] [backbone] torch.compile took 0.86 s in total
|
| 33 |
+
(EngineCore pid=559244) INFO 11-14 07:32:47 [monitor.py:91] [backbone] Initial profiling/warmup run took 0.19 s
|
| 34 |
+
|
| 35 |
+
|
| 36 |
+
(EngineCore pid=559244) INFO 11-14 07:32:53 [model_runner.py:1127] Graph capturing finished in 4 secs, took 0.41 GiB
|
| 37 |
+
(EngineCore pid=559244) INFO 11-14 07:32:53 [gpu_worker.py:759] Available KV cache memory: 61.82 GiB
|
| 38 |
+
(EngineCore pid=559244) INFO 11-14 07:32:53 [gpu_worker.py:774] CUDA graph memory profiling is enabled (default since v0.21.0). The current --gpu-memory-utilization=0.8500 is equivalent to --gpu-memory-utilization=0.8427 without CUDA graph memory profiling. To maintain the same effective KV cache size as before, increase --gpu-memory-utilization to 0.8573. To disable, set VLLM_MEMORY_PROFILER_ESTIMATE_CUDAGRAPHS=0.
|
| 39 |
+
(EngineCore pid=559244) INFO 11-14 07:32:53 [kv_cache_utils.py:2327] GPU KV cache size: 578,752 tokens, Maximum concurrency for 6,144 tokens per request: 94.20x
|
| 40 |
+
(EngineCore pid=559244) INFO 11-14 07:32:53 [kernel_warmup.py:176] JIT kernel warmup starting.
|
| 41 |
+
(EngineCore pid=559244) INFO 11-14 07:32:53 [kernel_warmup.py:189] JIT kernel warmup finished in 0.00s.
|
| 42 |
+
(EngineCore pid=559244) INFO 11-14 07:32:53 [kernel.py:467] Final IR op priority after setting platform defaults: IrOpPriorityConfig(rms_norm=['native'], fused_add_rms_norm=['native'], gelu_and_mul_sparse=['triton', 'native'])
|
| 43 |
+
(EngineCore pid=559244) INFO 11-14 07:32:53 [kernel_warmup.py:460] Using FlashInfer autotune cache file: /home/lyw/.cache/vllm/flashinfer_autotune_cache/0.7.0.post1/90a/db99b1192fc74844ebd646236af799ac64cd592939582f8c379c7b2b3766d6d1/autotune_configs_dp0_rank0.json
|
| 44 |
+
2026-11-14 07:32:53,784 - INFO - autotuner.py:1046 - flashinfer.jit: [Autotuner]: Autotuning process starts ...
|
| 45 |
+
(EngineCore pid=559244) INFO 11-14 07:32:53 [kernel_warmup.py:367] Running FlashInfer autotune with 16384 tokens and token buckets (1, 2, 4, 8, 16, 32, 64, 128, 256, 512, 768, 1024, 1280, 1536, 1792, 2048, 2560, 3072, 3584, 4096, 8192, 16384).
|
| 46 |
+
2026-11-14 07:32:53,813 - INFO - autotuner.py:1069 - flashinfer.jit: [Autotuner]: Autotuning process ends
|
| 47 |
+
|
| 48 |
+
|
| 49 |
+
(EngineCore pid=559244) INFO 11-14 07:32:59 [model_runner.py:1127] Graph capturing finished in 5 secs, took 0.27 GiB
|
| 50 |
+
(EngineCore pid=559244) INFO 11-14 07:32:59 [gpu_worker.py:971] CUDA graph pool memory: 0.27 GiB (actual), 0.58 GiB (estimated), difference: 0.3 GiB (110.7%).
|
| 51 |
+
(EngineCore pid=559244) INFO 11-14 07:32:59 [gpu_worker.py:1034] Free memory on device (78.66/79.18 GiB) on startup. Desired GPU memory utilization is (0.85, 67.3 GiB). Actual usage is 4.16 GiB for consumed memory (weights + non-torch), 1.33 GiB for peak activation, and 0.27 GiB for CUDAGraph memory. Replace gpu_memory_utilization config with `--kv-cache-memory=65925157991` (61.4 GiB) to fit into requested memory, or `--kv-cache-memory=78125379072` (72.76 GiB) to fully utilize gpu memory. Current kv cache memory in use is 61.82 GiB.
|
| 52 |
+
(EngineCore pid=559244) INFO 11-14 07:33:01 [jit_monitor.py:84] Kernel JIT monitor activated; monitored JIT compilations during inference will use mode=warn.
|
| 53 |
+
(EngineCore pid=559244) INFO 11-14 07:33:02 [core.py:395] init engine (profile, create kv cache, warmup model) took 20.48 s (compilation: 0.86 s)
|
| 54 |
+
(EngineCore pid=559244) INFO 11-14 07:33:02 [kernel.py:467] Final IR op priority after setting platform defaults: IrOpPriorityConfig(rms_norm=['native'], fused_add_rms_norm=['native'], gelu_and_mul_sparse=['triton', 'native'])
|
| 55 |
+
(MainProcess pid=558908) INFO 11-14 07:33:03 [hf.py:648] Detected the chat template content format to be 'string'. You can set `--chat-template-content-format` to override this.
|
| 56 |
+
|
| 57 |
+
(EngineCore pid=559244) WARNING 11-14 07:33:03 [jit_monitor.py:140] Triton kernel JIT compilation during inference: _topp_sb_step_kernel. This causes a latency spike; consider extending warmup to cover this shape/config.
|
| 58 |
+
(EngineCore pid=559244) WARNING 11-14 07:33:03 [jit_monitor.py:140] Triton kernel JIT compilation during inference: _topp_sb_mask_kernel. This causes a latency spike; consider extending warmup to cover this shape/config.
|
| 59 |
+
|
| 60 |
+
|
| 61 |
+
|
| 62 |
+
|
| 63 |
+
wrote /fsx_0/user/lyw/projects/arl_data/runs/s1_numina_q17/asft_lr1e-6/eval/gsm8k.jsonl
|
| 64 |
+
(MainProcess pid=558908) INFO 11-14 07:33:33 [utils.py:628] [shutdown] Process manager: send sigterm to process EngineCore
|
| 65 |
+
(EngineCore pid=559244) INFO 11-14 07:33:33 [core.py:1454] [shutdown] EngineCore: trigger received signal=SIGTERM
|
| 66 |
+
(EngineCore pid=559244) INFO 11-14 07:33:33 [core.py:1635] [shutdown] EngineCore: start mode=abort timeout=0s
|
| 67 |
+
(EngineCore pid=559244) INFO 11-14 07:33:33 [core.py:1666] [shutdown] EngineCore: request processing complete; starting resource teardown
|
| 68 |
+
(EngineCore pid=559244) INFO 11-14 07:33:33 [core.py:1499] [shutdown] EngineCore: exiting busy loop
|
| 69 |
+
aime24 {"pass1": 0.05416666666666667, "n": 16, "kind": "math", "mean_tokens": 1414.15, "truncated": 0.2125, "pass4": 0.1171978021978022, "pass8": 0.15555296555296555, "pass16": 0.2}
|
| 70 |
+
Timeout during comparison
|
| 71 |
+
Timeout during comparison
|
| 72 |
+
Timeout during comparison
|
| 73 |
+
Timeout during comparison
|
| 74 |
+
Timeout during comparison
|
| 75 |
+
aime25 {"pass1": 0.022916666666666665, "n": 16, "kind": "math", "mean_tokens": 1265.7583333333334, "truncated": 0.17083333333333334, "pass4": 0.07926739926739927, "pass8": 0.13316239316239317, "pass16": 0.2}
|
| 76 |
+
amc23 {"pass1": 0.2921875, "n": 16, "kind": "math", "mean_tokens": 946.2734375, "truncated": 0.1, "pass4": 0.4859478021978022, "pass8": 0.5868745143745143, "pass16": 0.675}
|
| 77 |
+
gsm8k {"pass1": 0.7115238817285823, "n": 2, "kind": "math", "mean_tokens": 216.82031842304775, "truncated": 0.002274450341167551}
|
| 78 |
+
math500 {"pass1": 0.5565, "n": 4, "kind": "math", "mean_tokens": 578.323, "truncated": 0.056, "pass4": 0.726}
|
| 79 |
+
Timeout during comparison
|
| 80 |
+
Timeout during comparison
|
| 81 |
+
Timeout during comparison
|
| 82 |
+
minerva {"pass1": 0.16727941176470587, "n": 4, "kind": "math", "mean_tokens": 873.9420955882352, "truncated": 0.10845588235294118, "pass4": 0.3088235294117647}
|
| 83 |
+
olympiadbench {"pass1": 0.23701780415430268, "n": 4, "kind": "math", "mean_tokens": 990.3479228486647, "truncated": 0.10014836795252226, "pass4": 0.39020771513353114}
|
runs/s1_numina_q17/asft_lr1e-6/eval.sbatch
ADDED
|
@@ -0,0 +1,13 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
#!/bin/bash
|
| 2 |
+
#SBATCH --job-name=arl-s1_numina_q17-asft_lr1e-6-eval
|
| 3 |
+
#SBATCH --gres=gpu:1
|
| 4 |
+
#SBATCH --cpus-per-task=8
|
| 5 |
+
#SBATCH --mem=150G
|
| 6 |
+
#SBATCH --time=3:00:00
|
| 7 |
+
#SBATCH --requeue
|
| 8 |
+
#SBATCH --output=/fsx_0/user/lyw/projects/arl_data/runs/s1_numina_q17/asft_lr1e-6/eval.log
|
| 9 |
+
export HF_HUB_OFFLINE=1 TRANSFORMERS_OFFLINE=1 ARL_EVAL_SETS=/fsx_0/user/lyw/projects/arl_data/sft/v1/eval_sets NLTK_DATA=/fsx_0/user/lyw/projects/envs/nltk_data
|
| 10 |
+
cd /fsx_0/user/lyw/projects/arl_data/runs/s1_numina_q17/code-9c038dc
|
| 11 |
+
[ -e /fsx_0/user/lyw/projects/arl_data/runs/s1_numina_q17/asft_lr1e-6/final/config.json ] || { echo "no final model"; exit 1; }
|
| 12 |
+
PYTHONPATH=/fsx_0/user/lyw/projects/arl_data/runs/s1_numina_q17/code-9c038dc:/fsx_0/user/lyw/projects/envs/shim:/fsx_0/user/lyw/projects/envs/vllm32-py312 python3 -s -m arl.eval.generate --model /fsx_0/user/lyw/projects/arl_data/runs/s1_numina_q17/asft_lr1e-6/final --out /fsx_0/user/lyw/projects/arl_data/runs/s1_numina_q17/asft_lr1e-6/eval --bench math500:4,minerva:4,olympiadbench:4,amc23:16,aime24:16,aime25:16,gsm8k:2 --max_tokens 4096 --template plain
|
| 13 |
+
PYTHONPATH=/fsx_0/user/lyw/projects/arl_data/runs/s1_numina_q17/code-9c038dc:/fsx_0/user/lyw/projects/envs/arl-py312:/fsx_0/user/lyw/projects/envs/awe-py312:/fsx_0/user/lyw/projects/envs/arl-eval-py312 python3 -s -m arl.eval.grade --gen_dir /fsx_0/user/lyw/projects/arl_data/runs/s1_numina_q17/asft_lr1e-6/eval
|
runs/s1_numina_q17/asft_lr1e-6/eval/aime24.jsonl
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
runs/s1_numina_q17/asft_lr1e-6/eval/aime25.jsonl
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
runs/s1_numina_q17/asft_lr1e-6/eval/amc23.jsonl
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
runs/s1_numina_q17/asft_lr1e-6/eval/gsm8k.jsonl
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
runs/s1_numina_q17/asft_lr1e-6/eval/math500.jsonl
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
runs/s1_numina_q17/asft_lr1e-6/eval/minerva.jsonl
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
runs/s1_numina_q17/asft_lr1e-6/eval/olympiadbench.jsonl
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
runs/s1_numina_q17/asft_lr1e-6/eval/per_item.json
ADDED
|
@@ -0,0 +1 @@
|
|
|
|
|
|
|
| 1 |
+
{"aime24": [0.1875, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.6875, 0.0, 0.5, 0.0, 0.0, 0.0625, 0.0, 0.0625, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.125, 0.0, 0.0, 0.0], "aime25": [0.125, 0.0, 0.0, 0.0, 0.0, 0.125, 0.0, 0.0, 0.25, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0625, 0.0625, 0.0, 0.0625, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], "amc23": [0.75, 1.0, 0.9375, 0.375, 0.0625, 0.0, 0.0, 0.125, 0.25, 0.5625, 0.1875, 0.0, 0.0, 0.0625, 0.0, 0.1875, 0.375, 0.5, 0.0, 1.0, 0.0, 0.3125, 0.0625, 0.6875, 0.6875, 0.0, 0.1875, 0.0, 0.125, 0.1875, 0.875, 0.0, 0.0, 0.875, 1.0, 0.1875, 0.0625, 0.0, 0.0625, 0.0], "gsm8k": [1.0, 1.0, 0.5, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.5, 0.5, 1.0, 1.0, 1.0, 0.5, 0.0, 0.0, 0.5, 0.0, 1.0, 1.0, 1.0, 0.5, 0.5, 1.0, 0.5, 1.0, 1.0, 1.0, 1.0, 1.0, 0.5, 1.0, 1.0, 0.5, 1.0, 0.5, 1.0, 0.0, 1.0, 0.5, 0.0, 1.0, 1.0, 1.0, 1.0, 0.5, 1.0, 1.0, 0.5, 1.0, 0.5, 1.0, 1.0, 1.0, 0.5, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.5, 0.5, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.5, 1.0, 0.5, 1.0, 0.5, 0.0, 0.5, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.5, 1.0, 0.5, 1.0, 0.5, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.5, 0.5, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.5, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.5, 1.0, 0.5, 0.5, 1.0, 0.0, 0.5, 1.0, 1.0, 1.0, 1.0, 0.5, 1.0, 1.0, 0.5, 0.5, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.5, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.5, 1.0, 0.0, 1.0, 0.5, 0.0, 1.0, 1.0, 0.0, 1.0, 0.5, 1.0, 0.5, 1.0, 1.0, 1.0, 1.0, 0.5, 1.0, 0.0, 1.0, 1.0, 0.5, 0.0, 1.0, 1.0, 1.0, 0.0, 0.5, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.5, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.5, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.5, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.5, 0.5, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.5, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.5, 0.0, 1.0, 1.0, 0.5, 0.5, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.5, 1.0, 1.0, 0.0, 0.0, 1.0, 0.5, 1.0, 1.0, 1.0, 1.0, 0.5, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.5, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.5, 0.5, 1.0, 0.5, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.5, 1.0, 1.0, 0.5, 0.0, 0.5, 1.0, 1.0, 0.5, 0.0, 1.0, 0.5, 1.0, 1.0, 0.5, 0.0, 0.0, 1.0, 1.0, 0.0, 0.5, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.5, 1.0, 1.0, 0.5, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.5, 0.5, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.5, 1.0, 0.5, 0.0, 0.5, 0.0, 0.5, 1.0, 1.0, 1.0, 1.0, 1.0, 0.5, 0.5, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.5, 1.0, 1.0, 1.0, 0.5, 1.0, 1.0, 0.5, 1.0, 0.0, 1.0, 1.0, 0.5, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.5, 1.0, 1.0, 0.5, 0.0, 1.0, 1.0, 1.0, 0.0, 0.5, 1.0, 0.5, 0.5, 1.0, 1.0, 1.0, 1.0, 0.5, 0.5, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.5, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.5, 1.0, 1.0, 0.5, 1.0, 0.5, 1.0, 0.5, 0.0, 0.0, 0.0, 1.0, 0.5, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.5, 0.5, 1.0, 1.0, 0.5, 1.0, 1.0, 0.5, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.5, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.5, 0.5, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.5, 1.0, 0.0, 0.0, 1.0, 1.0, 0.5, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.5, 0.5, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.5, 0.0, 1.0, 1.0, 1.0, 1.0, 0.5, 1.0, 1.0, 0.5, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.5, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.5, 0.0, 1.0, 0.0, 0.5, 0.0, 0.5, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.5, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.5, 0.5, 0.5, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.5, 0.5, 1.0, 1.0, 0.5, 0.5, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.5, 0.5, 0.0, 0.0, 1.0, 0.5, 1.0, 0.0, 1.0, 0.5, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.5, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.5, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.5, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.5, 1.0, 0.0, 0.0, 0.0, 0.5, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.5, 1.0, 1.0, 0.5, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.5, 1.0, 0.0, 1.0, 0.5, 1.0, 0.5, 0.5, 1.0, 1.0, 1.0, 1.0, 0.5, 0.5, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.5, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.5, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.5, 0.0, 1.0, 1.0, 1.0, 1.0, 0.5, 0.5, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.5, 1.0, 0.5, 0.5, 0.5, 1.0, 1.0, 0.0, 1.0, 1.0, 0.5, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.5, 0.5, 0.5, 1.0, 0.5, 1.0, 0.5, 1.0, 1.0, 0.5, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.5, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.5, 1.0, 0.5, 1.0, 1.0, 1.0, 0.5, 0.0, 0.5, 0.5, 1.0, 1.0, 1.0, 1.0, 0.5, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.5, 0.0, 1.0, 0.0, 1.0, 0.5, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.5, 0.0, 0.0, 0.0, 1.0, 0.5, 0.5, 1.0, 1.0, 1.0, 0.5, 1.0, 0.5, 1.0, 0.5, 0.0, 1.0, 1.0, 0.5, 0.5, 1.0, 0.5, 0.0, 0.0, 0.5, 1.0, 0.5, 1.0, 0.5, 0.5, 1.0, 1.0, 0.5, 1.0, 0.5, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.5, 1.0, 0.5, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.5, 0.0, 1.0, 0.0, 1.0, 0.5, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.5, 0.5, 0.0, 1.0, 0.5, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.5, 0.0, 0.0, 0.5, 0.5, 0.0, 1.0, 0.5, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.5, 1.0, 0.0, 0.5, 0.0, 1.0, 0.5, 0.0, 1.0, 1.0, 1.0, 1.0, 0.5, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.5, 0.0, 1.0, 0.0, 0.5, 0.5, 0.5, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.5, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.5, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.5, 0.5, 1.0, 0.5, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.5, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.5, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.5, 1.0, 0.5, 0.0, 1.0, 0.5, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.5, 0.5, 0.5, 0.0, 1.0, 0.0, 0.5, 0.0, 1.0, 1.0, 0.5, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.5, 1.0, 0.5, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.5, 1.0, 0.5, 0.5, 1.0, 0.0, 0.5, 1.0, 1.0, 0.5, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.5, 0.5, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.5, 0.0, 0.5, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.5, 1.0, 0.0, 0.0, 0.5, 0.5, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.5, 1.0, 0.5, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.5, 0.0, 1.0, 1.0, 1.0, 1.0, 0.5, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.5, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.5, 1.0, 0.0, 1.0, 0.5, 1.0, 1.0, 1.0, 1.0], "math500": [1.0, 0.0, 1.0, 1.0, 0.25, 1.0, 1.0, 0.0, 1.0, 0.25, 0.5, 0.0, 1.0, 1.0, 0.0, 0.5, 1.0, 0.0, 0.0, 0.5, 1.0, 0.0, 0.25, 1.0, 0.0, 0.25, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.5, 1.0, 0.5, 1.0, 1.0, 1.0, 1.0, 0.25, 1.0, 0.0, 0.5, 1.0, 0.5, 0.75, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.5, 1.0, 1.0, 0.75, 0.75, 0.0, 1.0, 0.25, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.75, 1.0, 0.0, 1.0, 1.0, 1.0, 0.5, 0.25, 1.0, 0.25, 1.0, 0.0, 0.5, 0.0, 0.0, 0.5, 1.0, 1.0, 1.0, 0.25, 0.75, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.25, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.75, 0.25, 0.0, 0.0, 1.0, 0.75, 0.5, 0.75, 0.5, 1.0, 1.0, 0.0, 0.5, 0.25, 1.0, 0.5, 0.0, 1.0, 1.0, 0.0, 1.0, 0.25, 0.0, 0.0, 1.0, 1.0, 1.0, 0.25, 1.0, 0.25, 1.0, 0.0, 0.5, 1.0, 0.75, 0.25, 1.0, 0.75, 0.0, 1.0, 0.0, 1.0, 1.0, 0.25, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.75, 0.5, 1.0, 1.0, 1.0, 0.5, 0.0, 1.0, 0.5, 1.0, 0.0, 1.0, 0.5, 1.0, 0.75, 0.5, 1.0, 0.25, 1.0, 0.0, 1.0, 0.75, 0.25, 0.75, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.25, 0.0, 0.0, 1.0, 1.0, 0.75, 0.0, 1.0, 1.0, 0.0, 0.75, 1.0, 0.5, 1.0, 0.75, 0.75, 0.25, 0.0, 0.75, 0.75, 1.0, 0.75, 0.5, 1.0, 0.75, 0.0, 0.0, 1.0, 0.5, 0.0, 1.0, 0.25, 0.75, 0.5, 0.0, 0.25, 0.0, 0.5, 0.0, 0.5, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.75, 0.0, 0.75, 0.5, 0.0, 0.0, 0.75, 0.0, 1.0, 0.75, 0.0, 0.0, 0.25, 0.0, 0.0, 0.75, 0.25, 0.75, 0.75, 0.75, 0.75, 1.0, 1.0, 1.0, 1.0, 0.75, 1.0, 0.75, 1.0, 0.0, 1.0, 0.75, 0.25, 1.0, 0.25, 1.0, 0.5, 0.25, 1.0, 0.25, 1.0, 0.25, 1.0, 1.0, 0.0, 0.75, 0.0, 0.5, 1.0, 0.25, 0.0, 0.0, 0.5, 0.0, 1.0, 0.75, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.75, 0.0, 0.75, 1.0, 0.5, 0.0, 0.0, 0.5, 1.0, 0.0, 0.75, 0.0, 0.25, 1.0, 0.5, 1.0, 0.5, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.75, 0.75, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.25, 0.0, 0.0, 0.0, 0.5, 1.0, 1.0, 0.5, 1.0, 0.0, 0.25, 0.25, 1.0, 0.5, 1.0, 1.0, 1.0, 1.0, 1.0, 0.25, 1.0, 0.5, 0.0, 1.0, 0.75, 0.0, 0.5, 0.5, 0.25, 0.25, 0.25, 0.25, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.5, 0.75, 0.0, 0.0, 1.0, 1.0, 0.75, 1.0, 0.0, 1.0, 0.25, 0.0, 0.25, 0.0, 0.75, 1.0, 0.5, 1.0, 1.0, 1.0, 1.0, 0.25, 0.0, 0.0, 0.75, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.75, 0.75, 1.0, 0.5, 0.5, 0.75, 0.75, 0.0, 1.0, 1.0, 0.25, 0.75, 0.5, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.75, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.75, 0.5, 0.0, 1.0, 0.5, 0.0, 1.0, 0.75, 1.0, 0.75, 0.5, 0.0, 0.75, 0.75, 0.5, 1.0, 0.75, 0.75, 1.0, 0.75, 0.25, 1.0, 0.75, 1.0, 0.25, 0.25, 0.25, 1.0, 1.0, 1.0, 1.0, 1.0, 0.5, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.25, 1.0, 0.0, 0.25, 0.5, 0.0, 0.5, 0.75, 0.0, 0.25, 0.0, 0.25, 0.0, 1.0, 1.0, 1.0, 0.5, 0.0, 0.0, 0.5, 0.25, 0.0, 0.5, 0.75, 0.0, 0.25, 0.75], "minerva": [0.0, 0.0, 0.5, 0.25, 1.0, 0.0, 0.0, 0.5, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.5, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.25, 1.0, 0.0, 0.75, 0.0, 0.0, 0.0, 0.0, 0.0, 0.25, 0.0, 0.0, 0.0, 0.0, 0.5, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.75, 0.25, 0.0, 0.25, 0.25, 1.0, 0.25, 0.0, 0.0, 0.0, 0.25, 0.0, 1.0, 0.75, 1.0, 0.75, 0.0, 1.0, 0.0, 0.0, 0.5, 0.0, 1.0, 0.0, 1.0, 0.0, 0.5, 0.0, 0.5, 1.0, 0.5, 0.5, 0.0, 0.25, 0.0, 0.0, 0.0, 0.0, 0.0, 0.25, 0.0, 0.0, 1.0, 0.75, 0.0, 1.0, 0.0, 0.75, 0.0, 0.5, 0.5, 0.0, 0.25, 0.75, 0.5, 0.0, 0.0, 0.0, 0.75, 0.0, 0.25, 0.25, 0.75, 0.25, 0.5, 0.0, 0.0, 0.0, 0.25, 0.0, 0.25, 0.75, 0.0, 0.0, 0.25, 1.0, 0.0, 0.75, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.25, 0.5, 0.0, 0.25, 0.0, 0.25, 0.0, 0.0, 0.25, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.5, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.25, 0.25, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.25, 0.25, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.5, 0.25, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.5, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.25, 0.0, 0.0, 0.0, 0.0, 0.5, 0.0, 0.0, 0.5, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.25, 0.0, 0.25, 0.0, 0.0, 0.0, 0.0, 0.0, 0.5, 0.0, 0.0, 0.0, 0.0, 0.25, 0.0, 0.75, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.5, 0.5, 0.0, 0.25, 0.0, 0.75, 0.25, 0.75, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], "olympiadbench": [0.0, 0.0, 0.0, 0.25, 0.0, 0.0, 0.5, 0.0, 0.0, 0.0, 0.0, 0.25, 0.0, 0.0, 0.5, 0.0, 0.75, 0.0, 0.25, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.25, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.75, 0.0, 0.0, 0.25, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.5, 0.25, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.75, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.25, 0.0, 0.0, 0.0, 0.5, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.25, 0.0, 0.0, 0.0, 0.0, 0.25, 1.0, 0.0, 0.75, 0.75, 0.0, 0.5, 0.75, 0.25, 0.75, 1.0, 0.0, 0.0, 0.5, 0.0, 0.0, 1.0, 0.0, 0.25, 0.25, 0.75, 1.0, 0.0, 1.0, 0.0, 0.75, 0.75, 1.0, 0.25, 0.75, 0.0, 0.0, 1.0, 0.25, 0.0, 0.25, 0.0, 0.0, 0.25, 1.0, 0.0, 0.75, 0.0, 0.0, 0.0, 1.0, 0.75, 0.25, 0.75, 0.0, 1.0, 1.0, 0.5, 0.5, 0.0, 0.0, 0.0, 0.0, 0.25, 0.75, 0.25, 0.0, 1.0, 0.75, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.75, 0.0, 0.0, 0.5, 0.75, 0.0, 0.75, 0.25, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.75, 1.0, 0.5, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.25, 0.75, 0.0, 0.5, 1.0, 0.0, 0.25, 0.0, 0.25, 0.0, 0.0, 1.0, 1.0, 0.5, 0.25, 0.0, 0.0, 0.0, 0.75, 0.0, 0.0, 0.5, 0.75, 0.25, 0.0, 0.75, 0.0, 1.0, 0.0, 0.75, 0.0, 0.25, 0.25, 0.75, 1.0, 0.25, 0.0, 0.0, 1.0, 0.25, 0.5, 0.0, 0.0, 0.5, 0.0, 0.0, 0.0, 1.0, 0.0, 0.25, 1.0, 0.5, 0.0, 0.0, 0.0, 0.5, 0.25, 0.0, 0.0, 1.0, 0.0, 0.25, 0.0, 0.0, 0.0, 0.5, 0.0, 0.0, 0.25, 1.0, 1.0, 0.0, 0.5, 0.5, 0.25, 0.0, 0.5, 0.25, 0.0, 0.25, 0.0, 0.0, 0.0, 0.0, 0.75, 1.0, 0.25, 0.0, 0.0, 0.0, 0.0, 0.25, 0.75, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.25, 0.25, 0.5, 0.0, 0.0, 0.0, 0.25, 0.0, 0.0, 0.0, 0.0, 0.25, 0.75, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.5, 0.0, 0.25, 1.0, 0.5, 1.0, 0.0, 0.0, 0.75, 1.0, 0.75, 0.75, 0.0, 1.0, 0.5, 0.0, 0.0, 0.0, 0.0, 0.75, 0.25, 0.5, 0.0, 0.75, 0.0, 0.0, 0.25, 0.0, 0.0, 0.0, 1.0, 0.0, 0.5, 0.5, 0.0, 0.25, 0.0, 0.0, 0.0, 1.0, 0.25, 0.0, 0.0, 0.0, 0.0, 0.25, 0.0, 0.0, 0.0, 0.5, 0.5, 0.5, 0.0, 0.0, 0.25, 0.0, 0.5, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.25, 0.75, 0.25, 0.0, 0.0, 0.25, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.75, 1.0, 0.5, 0.75, 0.75, 0.75, 1.0, 1.0, 0.75, 0.75, 0.0, 0.75, 0.5, 0.5, 0.0, 0.0, 0.0, 0.75, 0.0, 0.0, 0.0, 0.0, 0.0, 0.25, 0.0, 1.0, 0.0, 0.0, 0.25, 0.25, 0.0, 0.0, 0.25, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.75, 0.5, 0.5, 0.75, 0.0, 1.0, 0.25, 0.0, 0.0, 0.25, 0.75, 0.0, 0.0, 0.0, 0.25, 0.0, 0.0, 0.25, 0.0, 0.0, 0.0, 0.25, 0.0, 0.5, 0.75, 0.0, 0.0, 0.75, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.25, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.75, 0.25, 0.25, 0.0, 0.0, 0.0, 0.75, 1.0, 0.0, 0.75, 0.5, 0.0, 0.5, 0.5, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.75, 1.0, 0.0, 0.25, 0.75, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.25, 0.0, 0.0, 0.0, 0.75, 0.75, 1.0, 0.75, 0.75, 0.0, 0.0, 0.75, 0.0, 0.5, 0.0, 0.25, 0.25, 0.0, 0.0, 0.0, 0.25, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.75, 0.25, 0.0, 0.25, 0.75, 0.0, 1.0, 0.25, 0.25, 0.0, 1.0, 0.0, 0.0, 0.0, 0.75, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.25, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.5, 0.75, 0.5, 0.25, 0.25, 0.0, 1.0, 0.0, 0.0, 0.25, 0.75, 0.75, 0.25, 0.0, 0.0, 0.75, 0.25, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.5, 0.5, 0.75, 1.0, 0.0, 1.0, 0.0, 0.0, 0.75, 0.5, 0.0, 0.0, 0.0, 0.0, 0.0, 0.5, 0.0, 0.0, 0.0, 0.0, 1.0, 0.25, 0.25, 0.0, 0.0, 0.0, 0.25, 0.75, 0.75, 1.0, 0.5, 1.0, 1.0]}
|
runs/s1_numina_q17/asft_lr1e-6/eval/scores.json
ADDED
|
@@ -0,0 +1,63 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"aime24": {
|
| 3 |
+
"pass1": 0.05416666666666667,
|
| 4 |
+
"n": 16,
|
| 5 |
+
"kind": "math",
|
| 6 |
+
"mean_tokens": 1414.15,
|
| 7 |
+
"truncated": 0.2125,
|
| 8 |
+
"pass4": 0.1171978021978022,
|
| 9 |
+
"pass8": 0.15555296555296555,
|
| 10 |
+
"pass16": 0.2
|
| 11 |
+
},
|
| 12 |
+
"aime25": {
|
| 13 |
+
"pass1": 0.022916666666666665,
|
| 14 |
+
"n": 16,
|
| 15 |
+
"kind": "math",
|
| 16 |
+
"mean_tokens": 1265.7583333333334,
|
| 17 |
+
"truncated": 0.17083333333333334,
|
| 18 |
+
"pass4": 0.07926739926739927,
|
| 19 |
+
"pass8": 0.13316239316239317,
|
| 20 |
+
"pass16": 0.2
|
| 21 |
+
},
|
| 22 |
+
"amc23": {
|
| 23 |
+
"pass1": 0.2921875,
|
| 24 |
+
"n": 16,
|
| 25 |
+
"kind": "math",
|
| 26 |
+
"mean_tokens": 946.2734375,
|
| 27 |
+
"truncated": 0.1,
|
| 28 |
+
"pass4": 0.4859478021978022,
|
| 29 |
+
"pass8": 0.5868745143745143,
|
| 30 |
+
"pass16": 0.675
|
| 31 |
+
},
|
| 32 |
+
"gsm8k": {
|
| 33 |
+
"pass1": 0.7115238817285823,
|
| 34 |
+
"n": 2,
|
| 35 |
+
"kind": "math",
|
| 36 |
+
"mean_tokens": 216.82031842304775,
|
| 37 |
+
"truncated": 0.002274450341167551
|
| 38 |
+
},
|
| 39 |
+
"math500": {
|
| 40 |
+
"pass1": 0.5565,
|
| 41 |
+
"n": 4,
|
| 42 |
+
"kind": "math",
|
| 43 |
+
"mean_tokens": 578.323,
|
| 44 |
+
"truncated": 0.056,
|
| 45 |
+
"pass4": 0.726
|
| 46 |
+
},
|
| 47 |
+
"minerva": {
|
| 48 |
+
"pass1": 0.16727941176470587,
|
| 49 |
+
"n": 4,
|
| 50 |
+
"kind": "math",
|
| 51 |
+
"mean_tokens": 873.9420955882352,
|
| 52 |
+
"truncated": 0.10845588235294118,
|
| 53 |
+
"pass4": 0.3088235294117647
|
| 54 |
+
},
|
| 55 |
+
"olympiadbench": {
|
| 56 |
+
"pass1": 0.23701780415430268,
|
| 57 |
+
"n": 4,
|
| 58 |
+
"kind": "math",
|
| 59 |
+
"mean_tokens": 990.3479228486647,
|
| 60 |
+
"truncated": 0.10014836795252226,
|
| 61 |
+
"pass4": 0.39020771513353114
|
| 62 |
+
}
|
| 63 |
+
}
|
runs/s1_numina_q17/asft_lr1e-6/final/chat_template.jinja
ADDED
|
@@ -0,0 +1,85 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{%- if tools %}
|
| 2 |
+
{{- '<|im_start|>system\n' }}
|
| 3 |
+
{%- if messages[0].role == 'system' %}
|
| 4 |
+
{{- messages[0].content + '\n\n' }}
|
| 5 |
+
{%- endif %}
|
| 6 |
+
{{- "# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>" }}
|
| 7 |
+
{%- for tool in tools %}
|
| 8 |
+
{{- "\n" }}
|
| 9 |
+
{{- tool | tojson }}
|
| 10 |
+
{%- endfor %}
|
| 11 |
+
{{- "\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call><|im_end|>\n" }}
|
| 12 |
+
{%- else %}
|
| 13 |
+
{%- if messages[0].role == 'system' %}
|
| 14 |
+
{{- '<|im_start|>system\n' + messages[0].content + '<|im_end|>\n' }}
|
| 15 |
+
{%- endif %}
|
| 16 |
+
{%- endif %}
|
| 17 |
+
{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
|
| 18 |
+
{%- for message in messages[::-1] %}
|
| 19 |
+
{%- set index = (messages|length - 1) - loop.index0 %}
|
| 20 |
+
{%- if ns.multi_step_tool and message.role == "user" and not(message.content.startswith('<tool_response>') and message.content.endswith('</tool_response>')) %}
|
| 21 |
+
{%- set ns.multi_step_tool = false %}
|
| 22 |
+
{%- set ns.last_query_index = index %}
|
| 23 |
+
{%- endif %}
|
| 24 |
+
{%- endfor %}
|
| 25 |
+
{%- for message in messages %}
|
| 26 |
+
{%- if (message.role == "user") or (message.role == "system" and not loop.first) %}
|
| 27 |
+
{{- '<|im_start|>' + message.role + '\n' + message.content + '<|im_end|>' + '\n' }}
|
| 28 |
+
{%- elif message.role == "assistant" %}
|
| 29 |
+
{%- set content = message.content %}
|
| 30 |
+
{%- set reasoning_content = '' %}
|
| 31 |
+
{%- if message.reasoning_content is defined and message.reasoning_content is not none %}
|
| 32 |
+
{%- set reasoning_content = message.reasoning_content %}
|
| 33 |
+
{%- else %}
|
| 34 |
+
{%- if '</think>' in message.content %}
|
| 35 |
+
{%- set content = message.content.split('</think>')[-1].lstrip('\n') %}
|
| 36 |
+
{%- set reasoning_content = message.content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
|
| 37 |
+
{%- endif %}
|
| 38 |
+
{%- endif %}
|
| 39 |
+
{%- if loop.index0 > ns.last_query_index %}
|
| 40 |
+
{%- if loop.last or (not loop.last and reasoning_content) %}
|
| 41 |
+
{{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content.strip('\n') + '\n</think>\n\n' + content.lstrip('\n') }}
|
| 42 |
+
{%- else %}
|
| 43 |
+
{{- '<|im_start|>' + message.role + '\n' + content }}
|
| 44 |
+
{%- endif %}
|
| 45 |
+
{%- else %}
|
| 46 |
+
{{- '<|im_start|>' + message.role + '\n' + content }}
|
| 47 |
+
{%- endif %}
|
| 48 |
+
{%- if message.tool_calls %}
|
| 49 |
+
{%- for tool_call in message.tool_calls %}
|
| 50 |
+
{%- if (loop.first and content) or (not loop.first) %}
|
| 51 |
+
{{- '\n' }}
|
| 52 |
+
{%- endif %}
|
| 53 |
+
{%- if tool_call.function %}
|
| 54 |
+
{%- set tool_call = tool_call.function %}
|
| 55 |
+
{%- endif %}
|
| 56 |
+
{{- '<tool_call>\n{"name": "' }}
|
| 57 |
+
{{- tool_call.name }}
|
| 58 |
+
{{- '", "arguments": ' }}
|
| 59 |
+
{%- if tool_call.arguments is string %}
|
| 60 |
+
{{- tool_call.arguments }}
|
| 61 |
+
{%- else %}
|
| 62 |
+
{{- tool_call.arguments | tojson }}
|
| 63 |
+
{%- endif %}
|
| 64 |
+
{{- '}\n</tool_call>' }}
|
| 65 |
+
{%- endfor %}
|
| 66 |
+
{%- endif %}
|
| 67 |
+
{{- '<|im_end|>\n' }}
|
| 68 |
+
{%- elif message.role == "tool" %}
|
| 69 |
+
{%- if loop.first or (messages[loop.index0 - 1].role != "tool") %}
|
| 70 |
+
{{- '<|im_start|>user' }}
|
| 71 |
+
{%- endif %}
|
| 72 |
+
{{- '\n<tool_response>\n' }}
|
| 73 |
+
{{- message.content }}
|
| 74 |
+
{{- '\n</tool_response>' }}
|
| 75 |
+
{%- if loop.last or (messages[loop.index0 + 1].role != "tool") %}
|
| 76 |
+
{{- '<|im_end|>\n' }}
|
| 77 |
+
{%- endif %}
|
| 78 |
+
{%- endif %}
|
| 79 |
+
{%- endfor %}
|
| 80 |
+
{%- if add_generation_prompt %}
|
| 81 |
+
{{- '<|im_start|>assistant\n' }}
|
| 82 |
+
{%- if enable_thinking is defined and enable_thinking is false %}
|
| 83 |
+
{{- '<think>\n\n</think>\n\n' }}
|
| 84 |
+
{%- endif %}
|
| 85 |
+
{%- endif %}
|
runs/s1_numina_q17/asft_lr1e-6/final/config.json
ADDED
|
@@ -0,0 +1,63 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"architectures": [
|
| 3 |
+
"Qwen3ForCausalLM"
|
| 4 |
+
],
|
| 5 |
+
"attention_bias": false,
|
| 6 |
+
"attention_dropout": 0.0,
|
| 7 |
+
"bos_token_id": 151643,
|
| 8 |
+
"dtype": "float32",
|
| 9 |
+
"eos_token_id": 151643,
|
| 10 |
+
"head_dim": 128,
|
| 11 |
+
"hidden_act": "silu",
|
| 12 |
+
"hidden_size": 2048,
|
| 13 |
+
"initializer_range": 0.02,
|
| 14 |
+
"intermediate_size": 6144,
|
| 15 |
+
"layer_types": [
|
| 16 |
+
"full_attention",
|
| 17 |
+
"full_attention",
|
| 18 |
+
"full_attention",
|
| 19 |
+
"full_attention",
|
| 20 |
+
"full_attention",
|
| 21 |
+
"full_attention",
|
| 22 |
+
"full_attention",
|
| 23 |
+
"full_attention",
|
| 24 |
+
"full_attention",
|
| 25 |
+
"full_attention",
|
| 26 |
+
"full_attention",
|
| 27 |
+
"full_attention",
|
| 28 |
+
"full_attention",
|
| 29 |
+
"full_attention",
|
| 30 |
+
"full_attention",
|
| 31 |
+
"full_attention",
|
| 32 |
+
"full_attention",
|
| 33 |
+
"full_attention",
|
| 34 |
+
"full_attention",
|
| 35 |
+
"full_attention",
|
| 36 |
+
"full_attention",
|
| 37 |
+
"full_attention",
|
| 38 |
+
"full_attention",
|
| 39 |
+
"full_attention",
|
| 40 |
+
"full_attention",
|
| 41 |
+
"full_attention",
|
| 42 |
+
"full_attention",
|
| 43 |
+
"full_attention"
|
| 44 |
+
],
|
| 45 |
+
"max_position_embeddings": 32768,
|
| 46 |
+
"max_window_layers": 28,
|
| 47 |
+
"model_type": "qwen3",
|
| 48 |
+
"num_attention_heads": 16,
|
| 49 |
+
"num_hidden_layers": 28,
|
| 50 |
+
"num_key_value_heads": 8,
|
| 51 |
+
"pad_token_id": null,
|
| 52 |
+
"rms_norm_eps": 1e-06,
|
| 53 |
+
"rope_parameters": {
|
| 54 |
+
"rope_theta": 1000000,
|
| 55 |
+
"rope_type": "default"
|
| 56 |
+
},
|
| 57 |
+
"sliding_window": null,
|
| 58 |
+
"tie_word_embeddings": true,
|
| 59 |
+
"transformers_version": "5.20.2",
|
| 60 |
+
"use_cache": false,
|
| 61 |
+
"use_sliding_window": false,
|
| 62 |
+
"vocab_size": 151936
|
| 63 |
+
}
|
runs/s1_numina_q17/asft_lr1e-6/final/generation_config.json
ADDED
|
@@ -0,0 +1,7 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"bos_token_id": 151643,
|
| 3 |
+
"do_sample": false,
|
| 4 |
+
"eos_token_id": 151643,
|
| 5 |
+
"max_new_tokens": 2048,
|
| 6 |
+
"transformers_version": "5.20.2"
|
| 7 |
+
}
|
runs/s1_numina_q17/asft_lr1e-6/final/tokenizer.json
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:be75606093db2094d7cd20f3c2f385c212750648bd6ea4fb2bf507a6a4c55506
|
| 3 |
+
size 11422650
|
runs/s1_numina_q17/asft_lr1e-6/final/tokenizer_config.json
ADDED
|
@@ -0,0 +1,30 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"add_prefix_space": false,
|
| 3 |
+
"backend": "tokenizers",
|
| 4 |
+
"bos_token": null,
|
| 5 |
+
"clean_up_tokenization_spaces": false,
|
| 6 |
+
"eos_token": "<|endoftext|>",
|
| 7 |
+
"errors": "replace",
|
| 8 |
+
"extra_special_tokens": [
|
| 9 |
+
"<|im_start|>",
|
| 10 |
+
"<|im_end|>",
|
| 11 |
+
"<|object_ref_start|>",
|
| 12 |
+
"<|object_ref_end|>",
|
| 13 |
+
"<|box_start|>",
|
| 14 |
+
"<|box_end|>",
|
| 15 |
+
"<|quad_start|>",
|
| 16 |
+
"<|quad_end|>",
|
| 17 |
+
"<|vision_start|>",
|
| 18 |
+
"<|vision_end|>",
|
| 19 |
+
"<|vision_pad|>",
|
| 20 |
+
"<|image_pad|>",
|
| 21 |
+
"<|video_pad|>"
|
| 22 |
+
],
|
| 23 |
+
"is_local": true,
|
| 24 |
+
"local_files_only": true,
|
| 25 |
+
"model_max_length": 131072,
|
| 26 |
+
"pad_token": "<|endoftext|>",
|
| 27 |
+
"split_special_tokens": false,
|
| 28 |
+
"tokenizer_class": "Qwen2Tokenizer",
|
| 29 |
+
"unk_token": null
|
| 30 |
+
}
|
runs/s1_numina_q17/asft_lr1e-6/train.log
ADDED
|
@@ -0,0 +1,121 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
|
| 2 |
+
*****************************************
|
| 3 |
+
Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed.
|
| 4 |
+
*****************************************
|
| 5 |
+
examples 20000 (dropped 0 longer than 4096), steps/epoch 312, total 936, method asft {'lam': 0.05}
|
| 6 |
+
|
| 7 |
+
|
| 8 |
+
|
| 9 |
+
|
| 10 |
+
NCCL version 2.27.7+cuda13.0
|
| 11 |
+
[rank2]:[W1112 09:48:26.164974987 Utils.hpp:137] Warning: Environment variable TORCH_NCCL_TRACE_BUFFER_SIZE is deprecated; use TORCH_FR_BUFFER_SIZE instead (function operator())
|
| 12 |
+
[rank1]:[W1112 09:48:26.171222265 Utils.hpp:137] Warning: Environment variable TORCH_NCCL_TRACE_BUFFER_SIZE is deprecated; use TORCH_FR_BUFFER_SIZE instead (function operator())
|
| 13 |
+
[rank0]:[W1112 09:48:26.187463656 Utils.hpp:137] Warning: Environment variable TORCH_NCCL_TRACE_BUFFER_SIZE is deprecated; use TORCH_FR_BUFFER_SIZE instead (function operator())
|
| 14 |
+
[rank3]:[W1112 09:48:26.192606997 Utils.hpp:137] Warning: Environment variable TORCH_NCCL_TRACE_BUFFER_SIZE is deprecated; use TORCH_FR_BUFFER_SIZE instead (function operator())
|
| 15 |
+
|
| 16 |
+
|
| 17 |
+
|
| 18 |
+
|
| 19 |
+
{'step': 1, 'epoch': 0.003, 'loss': 0.0869249450707501, 'gnorm': 1.6090588569641113, 'lr': 4.347813841927667e-08, 'ntok': 25530, 'sec': 3, 'mem_gb': 35.3, 'nll': 0.522143334740991, 'w': 0.800987196435566, 'kl': 0.0008342181100034732}
|
| 20 |
+
{'step': 10, 'epoch': 0.032, 'loss': 0.08058229042728164, 'gnorm': 1.4724681377410889, 'lr': 2.390630933829573e-07, 'ntok': 26790, 'sec': 20, 'mem_gb': 44.6, 'nll': 0.5296277908034714, 'w': 0.8091858115201568, 'kl': 0.0007906454185266555}
|
| 21 |
+
{'step': 20, 'epoch': 0.064, 'loss': 0.07997513032494942, 'gnorm': 1.4176256656646729, 'lr': 4.560076405307405e-07, 'ntok': 30147, 'sec': 38, 'mem_gb': 44.6, 'nll': 0.4936367787424951, 'w': 0.8158301665588615, 'kl': 0.0009971890958991696}
|
| 22 |
+
{'step': 30, 'epoch': 0.096, 'loss': 0.08176567983525053, 'gnorm': 1.424399733543396, 'lr': 6.722063031189349e-07, 'ntok': 25645, 'sec': 56, 'mem_gb': 44.6, 'nll': 0.48661182126145447, 'w': 0.8130888696627023, 'kl': 0.0026431132208931444}
|
| 23 |
+
{'step': 40, 'epoch': 0.128, 'loss': 0.06888410232565582, 'gnorm': 1.0930588245391846, 'lr': 8.872940037943723e-07, 'ntok': 25388, 'sec': 74, 'mem_gb': 44.6, 'nll': 0.4689494821618481, 'w': 0.8348925458139672, 'kl': 0.013379880869991482}
|
| 24 |
+
{'step': 50, 'epoch': 0.16, 'loss': 0.06528278532983715, 'gnorm': 0.8136954307556152, 'lr': 9.929756110423193e-07, 'ntok': 25361, 'sec': 92, 'mem_gb': 44.8, 'nll': 0.500497503548756, 'w': 0.8364383921178187, 'kl': 0.038796331269438065}
|
| 25 |
+
{'step': 60, 'epoch': 0.192, 'loss': 0.057885607501609045, 'gnorm': 0.48379114270210266, 'lr': 9.898953260211337e-07, 'ntok': 27190, 'sec': 111, 'mem_gb': 44.8, 'nll': 0.5602462560913939, 'w': 0.8449931328153732, 'kl': 0.1083010506217819}
|
| 26 |
+
{'step': 70, 'epoch': 0.224, 'loss': 0.057980429433751604, 'gnorm': 0.40889155864715576, 'lr': 9.862632043093666e-07, 'ntok': 26478, 'sec': 130, 'mem_gb': 44.8, 'nll': 0.6434019221145857, 'w': 0.8368325532045472, 'kl': 0.14300943272775984}
|
| 27 |
+
{'step': 80, 'epoch': 0.256, 'loss': 0.05644751027319387, 'gnorm': 0.3984757959842682, 'lr': 9.820833372667812e-07, 'ntok': 24050, 'sec': 148, 'mem_gb': 44.8, 'nll': 0.601366050545738, 'w': 0.8459873635654885, 'kl': 0.14538362054963616}
|
| 28 |
+
{'step': 90, 'epoch': 0.288, 'loss': 0.05595124556501565, 'gnorm': 0.4150936007499695, 'lr': 9.773604332542728e-07, 'ntok': 23006, 'sec': 166, 'mem_gb': 44.8, 'nll': 0.5862697842410676, 'w': 0.8472166567960532, 'kl': 0.13649018785181039}
|
| 29 |
+
{'step': 100, 'epoch': 0.321, 'loss': 0.05373572209035291, 'gnorm': 0.3355247676372528, 'lr': 9.720998123301922e-07, 'ntok': 27542, 'sec': 184, 'mem_gb': 44.8, 'nll': 0.6485131656878586, 'w': 0.8446146595426984, 'kl': 0.16124865120770096}
|
| 30 |
+
{'step': 110, 'epoch': 0.353, 'loss': 0.05000246007216722, 'gnorm': 0.3394523561000824, 'lr': 9.6630740025763e-07, 'ntok': 25505, 'sec': 202, 'mem_gb': 44.8, 'nll': 0.6202832471574201, 'w': 0.8543257449519702, 'kl': 0.16722621024063908}
|
| 31 |
+
{'step': 120, 'epoch': 0.385, 'loss': 0.05229340318685203, 'gnorm': 0.3367582857608795, 'lr': 9.599897218294121e-07, 'ntok': 26525, 'sec': 220, 'mem_gb': 44.8, 'nll': 0.6078285520735156, 'w': 0.8522327108859566, 'kl': 0.15928374985273328}
|
| 32 |
+
{'step': 130, 'epoch': 0.417, 'loss': 0.052261552241395524, 'gnorm': 0.3843497633934021, 'lr': 9.531538935183249e-07, 'ntok': 21929, 'sec': 239, 'mem_gb': 44.8, 'nll': 0.586822859070751, 'w': 0.8541920801222126, 'kl': 0.16101965433900314}
|
| 33 |
+
{'step': 140, 'epoch': 0.449, 'loss': 0.05329943168031348, 'gnorm': 0.36078527569770813, 'lr': 9.458076154608515e-07, 'ntok': 24212, 'sec': 257, 'mem_gb': 44.8, 'nll': 0.6917093183390468, 'w': 0.8399009433596977, 'kl': 0.17149953245033456}
|
| 34 |
+
{'step': 150, 'epoch': 0.481, 'loss': 0.05103838665420117, 'gnorm': 0.33362311124801636, 'lr': 9.379591627834439e-07, 'ntok': 28019, 'sec': 275, 'mem_gb': 44.8, 'nll': 0.6397048654484457, 'w': 0.8499069548788322, 'kl': 0.16268391886420822}
|
| 35 |
+
{'step': 160, 'epoch': 0.513, 'loss': 0.05380730929569202, 'gnorm': 0.3240002393722534, 'lr': 9.296173762811084e-07, 'ntok': 25888, 'sec': 294, 'mem_gb': 45.1, 'nll': 0.6629999299868665, 'w': 0.8435787394256026, 'kl': 0.17160945180320034}
|
| 36 |
+
{'step': 170, 'epoch': 0.545, 'loss': 0.04697630039831147, 'gnorm': 0.30346280336380005, 'lr': 9.20791652458797e-07, 'ntok': 28205, 'sec': 312, 'mem_gb': 45.1, 'nll': 0.5712469600358979, 'w': 0.8629914493219287, 'kl': 0.15751580919828045}
|
| 37 |
+
{'step': 180, 'epoch': 0.577, 'loss': 0.05384583456316411, 'gnorm': 0.3401344418525696, 'lr': 9.114919329468282e-07, 'ntok': 24563, 'sec': 330, 'mem_gb': 45.1, 'nll': 0.6612473537434352, 'w': 0.8424495334700566, 'kl': 0.17159053537078126}
|
| 38 |
+
{'step': 190, 'epoch': 0.609, 'loss': 0.05167358711094722, 'gnorm': 0.31569167971611023, 'lr': 9.017286933022572e-07, 'ntok': 27330, 'sec': 348, 'mem_gb': 45.1, 'nll': 0.6167860838593121, 'w': 0.851679987650933, 'kl': 0.17151789185418953}
|
| 39 |
+
{'step': 200, 'epoch': 0.641, 'loss': 0.04897070264247459, 'gnorm': 0.29769694805145264, 'lr': 8.915129312088112e-07, 'ntok': 28334, 'sec': 366, 'mem_gb': 45.1, 'nll': 0.5988592547822404, 'w': 0.8583541140943742, 'kl': 0.17192924964265546}
|
| 40 |
+
{'step': 210, 'epoch': 0.673, 'loss': 0.05301152955478699, 'gnorm': 0.32760363817214966, 'lr': 8.808561540886796e-07, 'ntok': 26149, 'sec': 384, 'mem_gb': 45.1, 'nll': 0.6409917384508012, 'w': 0.8460346834486978, 'kl': 0.16853015486371373}
|
| 41 |
+
{'step': 220, 'epoch': 0.705, 'loss': 0.04755193125463478, 'gnorm': 0.31830576062202454, 'lr': 8.697703661401185e-07, 'ntok': 26970, 'sec': 402, 'mem_gb': 45.1, 'nll': 0.5791299505237301, 'w': 0.8636239340007416, 'kl': 0.1636248392310901}
|
| 42 |
+
{'step': 230, 'epoch': 0.737, 'loss': 0.049468741827101544, 'gnorm': 0.2874078154563904, 'lr': 8.582680548154658e-07, 'ntok': 29633, 'sec': 420, 'mem_gb': 45.1, 'nll': 0.6231776413753248, 'w': 0.8525292747949921, 'kl': 0.15893611535281613}
|
| 43 |
+
{'step': 240, 'epoch': 0.769, 'loss': 0.04825934622231809, 'gnorm': 0.30775976181030273, 'lr': 8.463621767547997e-07, 'ntok': 24847, 'sec': 439, 'mem_gb': 45.1, 'nll': 0.6281022671248038, 'w': 0.8556012012265062, 'kl': 0.16820769279867792}
|
| 44 |
+
{'step': 250, 'epoch': 0.801, 'loss': 0.05521725267947513, 'gnorm': 0.34809762239456177, 'lr': 8.340661431910887e-07, 'ntok': 25935, 'sec': 456, 'mem_gb': 45.1, 'nll': 0.744153050896472, 'w': 0.831193970503181, 'kl': 0.19444490884663582}
|
| 45 |
+
{'step': 260, 'epoch': 0.833, 'loss': 0.05329764776936809, 'gnorm': 0.3586273193359375, 'lr': 8.213938048432696e-07, 'ntok': 24228, 'sec': 474, 'mem_gb': 45.1, 'nll': 0.6481997682175582, 'w': 0.8453945331847449, 'kl': 0.18013831572070538}
|
| 46 |
+
{'step': 270, 'epoch': 0.865, 'loss': 0.054656157260817836, 'gnorm': 0.3239752948284149, 'lr': 8.083594363142717e-07, 'ntok': 27368, 'sec': 493, 'mem_gb': 45.3, 'nll': 0.7119464623374013, 'w': 0.8354148206847413, 'kl': 0.17907349096686825}
|
| 47 |
+
{'step': 280, 'epoch': 0.897, 'loss': 0.04763668988292369, 'gnorm': 0.31952401995658875, 'lr': 7.949777200115614e-07, 'ntok': 27402, 'sec': 509, 'mem_gb': 45.3, 'nll': 0.5870000222383402, 'w': 0.862361209948179, 'kl': 0.1616552887619973}
|
| 48 |
+
{'step': 290, 'epoch': 0.929, 'loss': 0.05026987975642131, 'gnorm': 0.31777870655059814, 'lr': 7.812637296083234e-07, 'ntok': 27545, 'sec': 528, 'mem_gb': 46.1, 'nll': 0.6125863643583227, 'w': 0.8539318473180251, 'kl': 0.16386553891926847}
|
| 49 |
+
{'step': 300, 'epoch': 0.962, 'loss': 0.0538608408463809, 'gnorm': 0.32860127091407776, 'lr': 7.672329130639005e-07, 'ntok': 24350, 'sec': 546, 'mem_gb': 46.1, 'nll': 0.7007662506416837, 'w': 0.8365554414784394, 'kl': 0.1824446547741273}
|
| 50 |
+
{'step': 310, 'epoch': 0.994, 'loss': 0.050665033813086284, 'gnorm': 0.30839115381240845, 'lr': 7.529010752226276e-07, 'ntok': 28463, 'sec': 565, 'mem_gb': 46.1, 'nll': 0.6223851743930717, 'w': 0.8518909143010224, 'kl': 0.168274142362453}
|
| 51 |
+
{'step': 320, 'epoch': 1.026, 'loss': 0.05191356414179816, 'gnorm': 0.3221592605113983, 'lr': 7.382843600106539e-07, 'ntok': 24906, 'sec': 583, 'mem_gb': 46.1, 'nll': 0.6325459509003855, 'w': 0.8490315774813298, 'kl': 0.1847890450907914}
|
| 52 |
+
{'step': 330, 'epoch': 1.058, 'loss': 0.0501503950503645, 'gnorm': 0.30206742882728577, 'lr': 7.233992322508129e-07, 'ntok': 30453, 'sec': 601, 'mem_gb': 46.1, 'nll': 0.646677231102026, 'w': 0.8508300701080354, 'kl': 0.17341284623025646}
|
| 53 |
+
{'step': 340, 'epoch': 1.09, 'loss': 0.05235137473868815, 'gnorm': 0.32102447748184204, 'lr': 7.0826245911602e-07, 'ntok': 25891, 'sec': 619, 'mem_gb': 46.1, 'nll': 0.6816053961801398, 'w': 0.8431810583310803, 'kl': 0.18000110816137846}
|
| 54 |
+
{'step': 350, 'epoch': 1.122, 'loss': 0.04848927643164347, 'gnorm': 0.30236417055130005, 'lr': 6.928910912420927e-07, 'ntok': 29529, 'sec': 639, 'mem_gb': 46.4, 'nll': 0.6119034995258898, 'w': 0.8569023248332148, 'kl': 0.17227255013080361}
|
| 55 |
+
{'step': 360, 'epoch': 1.154, 'loss': 0.05141210623552853, 'gnorm': 0.353666752576828, 'lr': 6.773024435212677e-07, 'ntok': 26129, 'sec': 657, 'mem_gb': 46.4, 'nll': 0.5846261726673428, 'w': 0.8533821227132687, 'kl': 0.16955737862504305}
|
| 56 |
+
{'step': 370, 'epoch': 1.186, 'loss': 0.050204886497083115, 'gnorm': 0.31693151593208313, 'lr': 6.615140755980489e-07, 'ntok': 26398, 'sec': 675, 'mem_gb': 46.4, 'nll': 0.6306135141630048, 'w': 0.8526338406318661, 'kl': 0.17461391044065838}
|
| 57 |
+
{'step': 380, 'epoch': 1.218, 'loss': 0.05035947529101392, 'gnorm': 0.3220553398132324, 'lr': 6.455437720893564e-07, 'ntok': 26427, 'sec': 692, 'mem_gb': 46.4, 'nll': 0.6154233537196806, 'w': 0.853560634010671, 'kl': 0.17791928886650016}
|
| 58 |
+
{'step': 390, 'epoch': 1.25, 'loss': 0.05556197628234711, 'gnorm': 0.3119874596595764, 'lr': 6.294095225512604e-07, 'ntok': 29253, 'sec': 710, 'mem_gb': 46.4, 'nll': 0.7424591735249376, 'w': 0.8300677761682562, 'kl': 0.1969408719832154}
|
| 59 |
+
{'step': 400, 'epoch': 1.282, 'loss': 0.047770573501179706, 'gnorm': 0.28949132561683655, 'lr': 6.131295012148612e-07, 'ntok': 28291, 'sec': 727, 'mem_gb': 46.4, 'nll': 0.6202788375720194, 'w': 0.8578518372671521, 'kl': 0.17320787957632286}
|
| 60 |
+
{'step': 410, 'epoch': 1.314, 'loss': 0.04741672057913655, 'gnorm': 0.3180288076400757, 'lr': 5.967220465141456e-07, 'ntok': 25456, 'sec': 745, 'mem_gb': 46.4, 'nll': 0.5716895697478002, 'w': 0.865207981418919, 'kl': 0.18157663591098366}
|
| 61 |
+
{'step': 420, 'epoch': 1.346, 'loss': 0.04791891417761225, 'gnorm': 0.28354179859161377, 'lr': 5.802056404288801e-07, 'ntok': 31491, 'sec': 763, 'mem_gb': 46.4, 'nll': 0.6197759131577276, 'w': 0.8577432093693437, 'kl': 0.17339281027218095}
|
| 62 |
+
{'step': 430, 'epoch': 1.378, 'loss': 0.052880049405032645, 'gnorm': 0.3530747890472412, 'lr': 5.635988876658055e-07, 'ntok': 23360, 'sec': 781, 'mem_gb': 46.4, 'nll': 0.6875194392792167, 'w': 0.8424013738762842, 'kl': 0.18579736997003424}
|
| 63 |
+
{'step': 440, 'epoch': 1.41, 'loss': 0.04756425601150042, 'gnorm': 0.3048025071620941, 'lr': 5.469204947015897e-07, 'ntok': 26461, 'sec': 800, 'mem_gb': 46.4, 'nll': 0.6082698579395147, 'w': 0.8587237615027776, 'kl': 0.17589491964589396}
|
| 64 |
+
{'step': 450, 'epoch': 1.442, 'loss': 0.05051267470121887, 'gnorm': 0.3512784540653229, 'lr': 5.30189248711143e-07, 'ntok': 23690, 'sec': 818, 'mem_gb': 46.4, 'nll': 0.6135559571549177, 'w': 0.8545471783980583, 'kl': 0.18364334074240185}
|
| 65 |
+
{'step': 460, 'epoch': 1.474, 'loss': 0.04718658728712885, 'gnorm': 0.2995431125164032, 'lr': 5.134239964050307e-07, 'ntok': 26674, 'sec': 836, 'mem_gb': 46.4, 'nll': 0.6031154298778774, 'w': 0.859581632629152, 'kl': 0.17217220830677812}
|
| 66 |
+
{'step': 470, 'epoch': 1.506, 'loss': 0.04533503751283318, 'gnorm': 0.31976836919784546, 'lr': 4.96643622799822e-07, 'ntok': 25812, 'sec': 854, 'mem_gb': 46.4, 'nll': 0.5483138898574307, 'w': 0.8692032775453278, 'kl': 0.16062470641077795}
|
| 67 |
+
{'step': 480, 'epoch': 1.538, 'loss': 0.055809169418852556, 'gnorm': 0.3286913335323334, 'lr': 4.798670299452926e-07, 'ntok': 26267, 'sec': 872, 'mem_gb': 46.4, 'nll': 0.723256456532912, 'w': 0.8302536809494804, 'kl': 0.19780019323795067}
|
| 68 |
+
{'step': 490, 'epoch': 1.571, 'loss': 0.05206660111865599, 'gnorm': 0.3109912574291229, 'lr': 4.631131156324358e-07, 'ntok': 27723, 'sec': 890, 'mem_gb': 46.4, 'nll': 0.6625270815297767, 'w': 0.8438472229737042, 'kl': 0.18909949404600873}
|
| 69 |
+
{'step': 500, 'epoch': 1.603, 'loss': 0.05224436604356986, 'gnorm': 0.3567167818546295, 'lr': 4.464007521062761e-07, 'ntok': 23316, 'sec': 908, 'mem_gb': 46.4, 'nll': 0.6446064732908732, 'w': 0.8467988050051467, 'kl': 0.181763889000579}
|
| 70 |
+
{'step': 510, 'epoch': 1.635, 'loss': 0.05517786202047105, 'gnorm': 0.35826557874679565, 'lr': 4.2974876480745375e-07, 'ntok': 26165, 'sec': 926, 'mem_gb': 46.4, 'nll': 0.6725950398910758, 'w': 0.8408063909325435, 'kl': 0.19269164362578828}
|
| 71 |
+
{'step': 520, 'epoch': 1.667, 'loss': 0.05113714296743587, 'gnorm': 0.3161810338497162, 'lr': 4.131759111665348e-07, 'ntok': 28536, 'sec': 944, 'mem_gb': 46.4, 'nll': 0.6614536106803687, 'w': 0.8453911561273479, 'kl': 0.18249768384671994}
|
| 72 |
+
{'step': 530, 'epoch': 1.699, 'loss': 0.04901572699192539, 'gnorm': 0.31018465757369995, 'lr': 3.9670085947492994e-07, 'ntok': 28763, 'sec': 962, 'mem_gb': 46.4, 'nll': 0.6607524478062093, 'w': 0.8501003893891458, 'kl': 0.18511120243455134}
|
| 73 |
+
{'step': 540, 'epoch': 1.731, 'loss': 0.04689815412041152, 'gnorm': 0.31918099522590637, 'lr': 3.8034216785622125e-07, 'ntok': 27283, 'sec': 980, 'mem_gb': 46.4, 'nll': 0.5709366008595096, 'w': 0.8635858553174137, 'kl': 0.17243375926057436}
|
| 74 |
+
{'step': 550, 'epoch': 1.763, 'loss': 0.04827463266554939, 'gnorm': 0.30851635336875916, 'lr': 3.641182633615878e-07, 'ntok': 27858, 'sec': 998, 'mem_gb': 46.4, 'nll': 0.6144164714803647, 'w': 0.8573052820329887, 'kl': 0.17368590681985427}
|
| 75 |
+
{'step': 560, 'epoch': 1.795, 'loss': 0.049840639616786975, 'gnorm': 0.3112504780292511, 'lr': 3.4804742121287654e-07, 'ntok': 25455, 'sec': 1016, 'mem_gb': 46.4, 'nll': 0.649645974513848, 'w': 0.8486245641818896, 'kl': 0.17748592798566096}
|
| 76 |
+
{'step': 570, 'epoch': 1.827, 'loss': 0.05076468950788551, 'gnorm': 0.33640819787979126, 'lr': 3.321477442166977e-07, 'ntok': 26750, 'sec': 1033, 'mem_gb': 46.4, 'nll': 0.6193126460280374, 'w': 0.8528666764018692, 'kl': 0.1789958382009346}
|
| 77 |
+
{'step': 580, 'epoch': 1.859, 'loss': 0.050401771445079804, 'gnorm': 0.33302420377731323, 'lr': 3.164371423727362e-07, 'ntok': 26691, 'sec': 1051, 'mem_gb': 46.4, 'nll': 0.6269588326823649, 'w': 0.8529638091959837, 'kl': 0.18716106977188005}
|
| 78 |
+
{'step': 590, 'epoch': 1.891, 'loss': 0.05060414886809673, 'gnorm': 0.32977306842803955, 'lr': 3.009333126992495e-07, 'ntok': 27324, 'sec': 1070, 'mem_gb': 46.4, 'nll': 0.6735292662814741, 'w': 0.8457974073845338, 'kl': 0.18451216335684745}
|
| 79 |
+
{'step': 600, 'epoch': 1.923, 'loss': 0.04712421878703309, 'gnorm': 0.3226361572742462, 'lr': 2.856537192984728e-07, 'ntok': 26370, 'sec': 1088, 'mem_gb': 46.4, 'nll': 0.577346786713121, 'w': 0.8647313175483504, 'kl': 0.17066209233978005}
|
| 80 |
+
{'step': 610, 'epoch': 1.955, 'loss': 0.04613272859661001, 'gnorm': 0.32778024673461914, 'lr': 2.7061557368439294e-07, 'ntok': 24093, 'sec': 1107, 'mem_gb': 46.4, 'nll': 0.5467654797295896, 'w': 0.8692978121238534, 'kl': 0.17092595943323788}
|
| 81 |
+
{'step': 620, 'epoch': 1.987, 'loss': 0.04808759509953167, 'gnorm': 0.31468096375465393, 'lr': 2.5583581539504463e-07, 'ntok': 26317, 'sec': 1125, 'mem_gb': 46.4, 'nll': 0.5808343050594673, 'w': 0.8617662923443021, 'kl': 0.17691995445553255}
|
| 82 |
+
{'step': 630, 'epoch': 2.019, 'loss': 0.04799929403155717, 'gnorm': 0.3139543831348419, 'lr': 2.4133109291117155e-07, 'ntok': 26349, 'sec': 1142, 'mem_gb': 46.4, 'nll': 0.6181769973480967, 'w': 0.8570101094918213, 'kl': 0.1788186043587992}
|
| 83 |
+
{'step': 640, 'epoch': 2.051, 'loss': 0.04990893935383671, 'gnorm': 0.3011155426502228, 'lr': 2.2711774490274766e-07, 'ntok': 28602, 'sec': 1161, 'mem_gb': 46.4, 'nll': 0.6682338374501783, 'w': 0.8490826827013845, 'kl': 0.19104834438478951}
|
| 84 |
+
{'step': 650, 'epoch': 2.083, 'loss': 0.05021418551923986, 'gnorm': 0.30439266562461853, 'lr': 2.1321178182447709e-07, 'ntok': 29613, 'sec': 1181, 'mem_gb': 46.4, 'nll': 0.6714440501215683, 'w': 0.8463072726716308, 'kl': 0.18460264693187958}
|
| 85 |
+
{'step': 660, 'epoch': 2.115, 'loss': 0.0488305688501001, 'gnorm': 0.32635927200317383, 'lr': 1.9962886788101047e-07, 'ntok': 25974, 'sec': 1198, 'mem_gb': 46.4, 'nll': 0.6088243065782128, 'w': 0.858461376039501, 'kl': 0.18368319491366367}
|
| 86 |
+
{'step': 670, 'epoch': 2.147, 'loss': 0.049494598842379674, 'gnorm': 0.30959847569465637, 'lr': 1.8638430338219196e-07, 'ntok': 26636, 'sec': 1216, 'mem_gb': 46.4, 'nll': 0.6001432653410798, 'w': 0.8563276234466511, 'kl': 0.17649370873183848}
|
| 87 |
+
{'step': 680, 'epoch': 2.179, 'loss': 0.05042850708011978, 'gnorm': 0.2825760543346405, 'lr': 1.7349300750820757e-07, 'ntok': 29303, 'sec': 1234, 'mem_gb': 46.4, 'nll': 0.6907709828131932, 'w': 0.8432413063508856, 'kl': 0.18440709662299934}
|
| 88 |
+
{'step': 690, 'epoch': 2.212, 'loss': 0.0510461594438734, 'gnorm': 0.3245044946670532, 'lr': 1.6096950150405452e-07, 'ntok': 23957, 'sec': 1252, 'mem_gb': 46.4, 'nll': 0.6460164549374922, 'w': 0.8496572309241558, 'kl': 0.18317221838867553}
|
| 89 |
+
{'step': 700, 'epoch': 2.244, 'loss': 0.050083767027436096, 'gnorm': 0.3089928925037384, 'lr': 1.4882789232226122e-07, 'ntok': 26015, 'sec': 1270, 'mem_gb': 46.4, 'nll': 0.6214222113444167, 'w': 0.8557730365894676, 'kl': 0.19243146231741304}
|
| 90 |
+
{'step': 710, 'epoch': 2.276, 'loss': 0.04852002129905057, 'gnorm': 0.3113033175468445, 'lr': 1.3708185673227895e-07, 'ntok': 26516, 'sec': 1287, 'mem_gb': 46.4, 'nll': 0.6026603041559813, 'w': 0.8574636392885427, 'kl': 0.18161587579197466}
|
| 91 |
+
{'step': 720, 'epoch': 2.308, 'loss': 0.05041260471914175, 'gnorm': 0.3338736891746521, 'lr': 1.257446259144494e-07, 'ntok': 25925, 'sec': 1304, 'mem_gb': 46.4, 'nll': 0.618582600048216, 'w': 0.8544342152844745, 'kl': 0.18761170669599808}
|
| 92 |
+
{'step': 730, 'epoch': 2.34, 'loss': 0.044147906906061817, 'gnorm': 0.3160042464733124, 'lr': 1.1482897055590273e-07, 'ntok': 27234, 'sec': 1322, 'mem_gb': 46.4, 'nll': 0.553458746706782, 'w': 0.8720829345900345, 'kl': 0.16986095190065725}
|
| 93 |
+
{'step': 740, 'epoch': 2.372, 'loss': 0.0512855089570963, 'gnorm': 0.3355576694011688, 'lr': 1.0434718646516916e-07, 'ntok': 23241, 'sec': 1341, 'mem_gb': 46.4, 'nll': 0.6352280015651219, 'w': 0.8504784950357127, 'kl': 0.19512833092648768}
|
| 94 |
+
{'step': 750, 'epoch': 2.404, 'loss': 0.04649140241151247, 'gnorm': 0.2873701751232147, 'lr': 9.431108072171346e-08, 'ntok': 28189, 'sec': 1358, 'mem_gb': 46.4, 'nll': 0.5847832515564582, 'w': 0.8640494325135691, 'kl': 0.17122810399224875}
|
| 95 |
+
{'step': 760, 'epoch': 2.436, 'loss': 0.044327822533050044, 'gnorm': 0.3073582351207733, 'lr': 8.473195837599417e-08, 'ntok': 26475, 'sec': 1376, 'mem_gb': 46.4, 'nll': 0.5268348677998111, 'w': 0.875561850802644, 'kl': 0.17416570762511804}
|
| 96 |
+
{'step': 770, 'epoch': 2.468, 'loss': 0.05095954585663412, 'gnorm': 0.31848329305648804, 'lr': 7.562060971502383e-08, 'ntok': 27449, 'sec': 1393, 'mem_gb': 46.4, 'nll': 0.6471805524563736, 'w': 0.8497348207129586, 'kl': 0.196801732018926}
|
| 97 |
+
{'step': 780, 'epoch': 2.5, 'loss': 0.04686473732223054, 'gnorm': 0.30504345893859863, 'lr': 6.698729810778064e-08, 'ntok': 29225, 'sec': 1411, 'mem_gb': 46.4, 'nll': 0.5894706346236099, 'w': 0.8631965087681779, 'kl': 0.1790974524165954}
|
| 98 |
+
{'step': 790, 'epoch': 2.532, 'loss': 0.04562258359479352, 'gnorm': 0.2976017892360687, 'lr': 5.884174844416101e-08, 'ntok': 26952, 'sec': 1428, 'mem_gb': 46.4, 'nll': 0.567325853600846, 'w': 0.8671373529793708, 'kl': 0.16856919172858228}
|
| 99 |
+
{'step': 800, 'epoch': 2.564, 'loss': 0.04906316343984692, 'gnorm': 0.3100118637084961, 'lr': 5.119313618049309e-08, 'ntok': 26098, 'sec': 1446, 'mem_gb': 46.4, 'nll': 0.6500578947620508, 'w': 0.8509594545080083, 'kl': 0.18181743003534753}
|
| 100 |
+
{'step': 810, 'epoch': 2.596, 'loss': 0.04928981630681915, 'gnorm': 0.29870638251304626, 'lr': 4.4050077003954964e-08, 'ntok': 29386, 'sec': 1464, 'mem_gb': 46.4, 'nll': 0.6395087291780099, 'w': 0.8527245648438032, 'kl': 0.18232457765815355}
|
| 101 |
+
{'step': 820, 'epoch': 2.628, 'loss': 0.05328753872842097, 'gnorm': 0.28887492418289185, 'lr': 3.742061712753824e-08, 'ntok': 31570, 'sec': 1483, 'mem_gb': 46.4, 'nll': 0.7193405774865379, 'w': 0.8362966349580297, 'kl': 0.1974462442092968}
|
| 102 |
+
{'step': 830, 'epoch': 2.66, 'loss': 0.051159540717493485, 'gnorm': 0.32722121477127075, 'lr': 3.131222422648744e-08, 'ntok': 26860, 'sec': 1501, 'mem_gb': 46.4, 'nll': 0.6613184539045048, 'w': 0.850070097263589, 'kl': 0.20324923358386077}
|
| 103 |
+
{'step': 840, 'epoch': 2.692, 'loss': 0.05098399770984398, 'gnorm': 0.3159128427505493, 'lr': 2.5731779026427257e-08, 'ntok': 24548, 'sec': 1518, 'mem_gb': 46.4, 'nll': 0.6278217198193335, 'w': 0.8508806407344793, 'kl': 0.19251588246751264}
|
| 104 |
+
{'step': 850, 'epoch': 2.724, 'loss': 0.04748403877242131, 'gnorm': 0.3189385235309601, 'lr': 2.0685567552652716e-08, 'ntok': 25478, 'sec': 1535, 'mem_gb': 46.4, 'nll': 0.5987504844866159, 'w': 0.8619831018427663, 'kl': 0.18410161769467775}
|
| 105 |
+
{'step': 860, 'epoch': 2.756, 'loss': 0.04993932771836775, 'gnorm': 0.3439858853816986, 'lr': 1.6179274049310966e-08, 'ntok': 27551, 'sec': 1553, 'mem_gb': 46.4, 'nll': 0.6173341031178542, 'w': 0.8563541817901347, 'kl': 0.19269657720001088}
|
| 106 |
+
{'step': 870, 'epoch': 2.788, 'loss': 0.05266914694338725, 'gnorm': 0.3981289565563202, 'lr': 1.2217974576453072e-08, 'ntok': 21020, 'sec': 1570, 'mem_gb': 46.4, 'nll': 0.6516020828377735, 'w': 0.8473905060656518, 'kl': 0.20314750925457897}
|
| 107 |
+
{'step': 880, 'epoch': 2.821, 'loss': 0.04919102352689518, 'gnorm': 0.3187265992164612, 'lr': 8.806131292167617e-09, 'ntok': 24470, 'sec': 1588, 'mem_gb': 46.4, 'nll': 0.619145374118819, 'w': 0.8554301185124642, 'kl': 0.18948726677436656}
|
| 108 |
+
{'step': 890, 'epoch': 2.853, 'loss': 0.045164997122001525, 'gnorm': 0.2998771667480469, 'lr': 5.9475874262360785e-09, 'ntok': 26847, 'sec': 1605, 'mem_gb': 46.4, 'nll': 0.5404718171490297, 'w': 0.8732152917458189, 'kl': 0.17658303011044066}
|
| 109 |
+
{'step': 900, 'epoch': 2.885, 'loss': 0.049914230414842754, 'gnorm': 0.3363229036331177, 'lr': 3.6455629509730135e-09, 'ntok': 25072, 'sec': 1622, 'mem_gb': 46.4, 'nll': 0.6090219153213744, 'w': 0.8559369328832961, 'kl': 0.18822326854907864}
|
| 110 |
+
{'step': 910, 'epoch': 2.917, 'loss': 0.04483612430089964, 'gnorm': 0.3002648651599884, 'lr': 1.9026509541272273e-09, 'ntok': 29234, 'sec': 1640, 'mem_gb': 46.4, 'nll': 0.5360507029486214, 'w': 0.8717496242602791, 'kl': 0.17051392896007217}
|
| 111 |
+
{'step': 920, 'epoch': 2.949, 'loss': 0.04540506270087236, 'gnorm': 0.30061855912208557, 'lr': 7.208147179291191e-10, 'ntok': 27225, 'sec': 1658, 'mem_gb': 46.4, 'nll': 0.5621433798209367, 'w': 0.8674721648301194, 'kl': 0.17242431416437098}
|
| 112 |
+
{'step': 930, 'epoch': 2.981, 'loss': 0.050496972112595864, 'gnorm': 0.32800087332725525, 'lr': 1.0138550757493591e-10, 'ntok': 25687, 'sec': 1678, 'mem_gb': 46.4, 'nll': 0.6469950297086074, 'w': 0.8490367513236268, 'kl': 0.18669727376080314}
|
| 113 |
+
|
| 114 |
+
saved /fsx_0/user/lyw/projects/arl_data/runs/s1_numina_q17/asft_lr1e-6/final
|
| 115 |
+
/usr/local/lib/python3.12/site-packages/torch/distributed/distributed_c10d.py:4876: UserWarning: barrier(): using the device under current context. You can specify `device_id` in `init_process_group` to mute this warning.
|
| 116 |
+
warnings.warn( # warn only once
|
| 117 |
+
[W1112 10:16:47.043240455 AllocatorConfig.cpp:28] Warning: PYTORCH_CUDA_ALLOC_CONF is deprecated, use PYTORCH_ALLOC_CONF instead (function operator())
|
| 118 |
+
[W1112 10:16:47.043404095 AllocatorConfig.cpp:28] Warning: PYTORCH_CUDA_ALLOC_CONF is deprecated, use PYTORCH_ALLOC_CONF instead (function operator())
|
| 119 |
+
[W1112 10:16:47.043422516 AllocatorConfig.cpp:28] Warning: PYTORCH_CUDA_ALLOC_CONF is deprecated, use PYTORCH_ALLOC_CONF instead (function operator())
|
| 120 |
+
[W1112 10:16:47.045207864 AllocatorConfig.cpp:28] Warning: PYTORCH_CUDA_ALLOC_CONF is deprecated, use PYTORCH_ALLOC_CONF instead (function operator())
|
| 121 |
+
[W1112 10:16:50.543919216 AllocatorConfig.cpp:28] Warning: PYTORCH_CUDA_ALLOC_CONF is deprecated, use PYTORCH_ALLOC_CONF instead (function operator())
|
runs/s1_numina_q17/asft_lr1e-6/train.sbatch
ADDED
|
@@ -0,0 +1,12 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
#!/bin/bash
|
| 2 |
+
#SBATCH --job-name=arl-s1_numina_q17-asft_lr1e-6
|
| 3 |
+
#SBATCH --gres=gpu:4
|
| 4 |
+
#SBATCH --cpus-per-task=32
|
| 5 |
+
#SBATCH --mem=600G
|
| 6 |
+
#SBATCH --time=3:00:00
|
| 7 |
+
#SBATCH --requeue
|
| 8 |
+
#SBATCH --output=/fsx_0/user/lyw/projects/arl_data/runs/s1_numina_q17/asft_lr1e-6/train.log
|
| 9 |
+
export HF_HUB_OFFLINE=1 TRANSFORMERS_OFFLINE=1 PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True
|
| 10 |
+
cd /fsx_0/user/lyw/projects/arl_data/runs/s1_numina_q17/code-9c038dc
|
| 11 |
+
[ -e /fsx_0/user/lyw/projects/arl_data/runs/s1_numina_q17/asft_lr1e-6/final/config.json ] && exit 0
|
| 12 |
+
PYTHONPATH=/fsx_0/user/lyw/projects/arl_data/runs/s1_numina_q17/code-9c038dc:/fsx_0/user/lyw/projects/envs/arl-py312:/fsx_0/user/lyw/projects/envs/awe-py312:/fsx_0/user/lyw/projects/envs/arl-eval-py312 python3 -s -m torch.distributed.run --nproc_per_node 4 --master_port $((29500 + RANDOM % 500)) -m arl.train --model /fsx_0/user/lyw/projects/arl_data/models/Qwen3-1.7B-Base --data /fsx_0/user/lyw/projects/arl_data/sft/v1/numina_cot_20k.jsonl --out /fsx_0/user/lyw/projects/arl_data/runs/s1_numina_q17/asft_lr1e-6 --method asft --method_args '{}' --lr 1e-6 --epochs 3 --seqs_per_step 64 --max_len 4096 --micro_tokens 16384 --template plain --fisher_cache_dir /fsx_0/user/lyw/projects/arl_data/fisher
|
runs/s1_numina_q17/asft_lr1e-6/train_log.jsonl
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
runs/s1_numina_q17/asft_lr2e-5/args.json
ADDED
|
@@ -0,0 +1,30 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"model": "/fsx_0/user/lyw/projects/arl_data/models/Qwen3-1.7B-Base",
|
| 3 |
+
"data": "/fsx_0/user/lyw/projects/arl_data/sft/v1/numina_cot_20k.jsonl",
|
| 4 |
+
"out": "/fsx_0/user/lyw/projects/arl_data/runs/s1_numina_q17/asft_lr2e-5",
|
| 5 |
+
"method": "asft",
|
| 6 |
+
"method_args": "{}",
|
| 7 |
+
"template": "plain",
|
| 8 |
+
"max_len": 4096,
|
| 9 |
+
"limit": 0,
|
| 10 |
+
"lr": 2e-05,
|
| 11 |
+
"epochs": 3.0,
|
| 12 |
+
"seqs_per_step": 64,
|
| 13 |
+
"micro_tokens": 16384,
|
| 14 |
+
"chunk": 2048,
|
| 15 |
+
"warmup": 0.05,
|
| 16 |
+
"clip": 1.0,
|
| 17 |
+
"seed": 0,
|
| 18 |
+
"ckpt_every_min": 40,
|
| 19 |
+
"stop_at": 0,
|
| 20 |
+
"save_at": "",
|
| 21 |
+
"zero1": false,
|
| 22 |
+
"no_save": false,
|
| 23 |
+
"fisher": "",
|
| 24 |
+
"fisher_gamma": 1.0,
|
| 25 |
+
"fisher_n": 1024,
|
| 26 |
+
"fisher_cache_dir": "",
|
| 27 |
+
"method_cfg": {
|
| 28 |
+
"lam": 0.05
|
| 29 |
+
}
|
| 30 |
+
}
|
runs/s1_numina_q17/asft_lr2e-5/eval.log
ADDED
|
@@ -0,0 +1,85 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
(MainProcess pid=547852) INFO 11-14 07:27:25 [api_utils.py:286] non-default args: {'dtype': 'bfloat16', 'max_model_len': 6144, 'gpu_memory_utilization': 0.85, 'disable_log_stats': True, 'model': '/fsx_0/user/lyw/projects/arl_data/runs/s1_numina_q17/asft_lr2e-5/final'}
|
| 2 |
+
(MainProcess pid=547852) INFO 11-14 07:27:25 [model.py:733] Resolved architecture: Qwen3ForCausalLM
|
| 3 |
+
(MainProcess pid=547852) INFO 11-14 07:27:25 [model.py:2495] Downcasting torch.float32 to torch.bfloat16.
|
| 4 |
+
(MainProcess pid=547852) INFO 11-14 07:27:25 [model.py:2152] Using max model len 6144
|
| 5 |
+
(MainProcess pid=547852) INFO 11-14 07:27:25 [scheduler.py:307] Chunked prefill is enabled with max_num_batched_tokens=16384.
|
| 6 |
+
(MainProcess pid=547852) INFO 11-14 07:27:25 [kernel.py:467] Final IR op priority after setting platform defaults: IrOpPriorityConfig(rms_norm=['native'], fused_add_rms_norm=['native'], gelu_and_mul_sparse=['triton', 'native'])
|
| 7 |
+
(EngineCore pid=552081) INFO 11-14 07:27:28 [core.py:133] Initializing a V1 LLM engine (v0.32.0) with config: model='/fsx_0/user/lyw/projects/arl_data/runs/s1_numina_q17/asft_lr2e-5/final', speculative_config=None, tokenizer='/fsx_0/user/lyw/projects/arl_data/runs/s1_numina_q17/asft_lr2e-5/final', skip_tokenizer_init=False, tokenizer_mode=auto, revision=None, tokenizer_revision=None, trust_remote_code=False, dtype=torch.bfloat16, max_seq_len=6144, download_dir=None, load_format=auto, tensor_parallel_size=1, pipeline_parallel_size=1, data_parallel_size=1, decode_context_parallel_size=1, dcp_comm_backend=ag_rs, disable_custom_all_reduce=False, quantization=None, quantization_config=None, enforce_eager=False, aux_output_config=AuxOutputConfig(enable_return_routed_experts=False, max_bytes=None), kv_cache_dtype=auto, device_config=cuda, structured_outputs_config=StructuredOutputsConfig(backend='auto', disable_any_whitespace=False, disable_additional_properties=False, reasoning_parser='', reasoning_parser_plugin='', enable_in_reasoning=False), observability_config=ObservabilityConfig(show_hidden_metrics_for_version=None, otlp_traces_endpoint=None, collect_detailed_traces=None, per_request_spec_decode_metrics='none', kv_cache_metrics=False, kv_cache_metrics_sample=0.01, custom_histogram_buckets=None, cudagraph_metrics=False, enable_layerwise_nvtx_tracing=False, enable_mfu_metrics=False, enable_mm_processor_stats=False, enable_logging_iteration_details=False, jit_monitor_mode='warn', jit_monitor_verbose=False), seed=0, served_model_name=/fsx_0/user/lyw/projects/arl_data/runs/s1_numina_q17/asft_lr2e-5/final, enable_prefix_caching=True, enable_chunked_prefill=True, pooler_config=None, compilation_config={'mode': <CompilationMode.VLLM_COMPILE: 3>, 'debug_dump_path': None, 'cache_dir': '', 'compile_cache_save_format': 'binary', 'backend': 'inductor', 'custom_ops': ['none'], 'ir_enable_torch_wrap': True, 'splitting_ops': ['vllm::unified_attention_with_output', 'vllm::unified_mla_attention_with_output', 'vllm::mamba_mixer2', 'vllm::mamba_mixer', 'vllm::short_conv', 'vllm::qwen4_exp_ple_short_conv', 'vllm::qwen4_exp_qsa_with_output', 'vllm::qwen4_exp_qsa_prepare_with_output', 'vllm::linear_attention', 'vllm::qwen_gdn_attention_core', 'vllm::qwen_gdn_attention_core_fused_norm_packed', 'vllm::gdn_attention_core_xpu', 'vllm::olmo_hybrid_gdn_full_forward', 'vllm::sparse_attn_indexer', 'vllm::rocm_aiter_sparse_attn_indexer', 'vllm::deepseek_v4_attention', 'vllm::hpc_rope_norm_forward', 'vllm::unified_kv_cache_update', 'vllm::unified_mla_kv_cache_update'], 'compile_mm_encoder': False, 'cudagraph_mm_encoder': False, 'encoder_cudagraph_token_budgets': [], 'encoder_cudagraph_max_vision_items_per_batch': 0, 'encoder_cudagraph_max_frames_per_batch': None, 'cudagraph_decoder_replay': True, 'decoder_replay_cudagraph_capture_sizes': [], 'decoder_replay_trim_threshold': 768, 'compile_sizes': [], 'compile_ranges_endpoints': [16384], 'inductor_compile_config': {'enable_auto_functionalized_v2': False, 'combo_kernels': True, 'benchmark_combo_kernel': True}, 'inductor_passes': {}, 'cudagraph_mode': <CUDAGraphMode.FULL_AND_PIECEWISE: (2, 1)>, 'cudagraph_num_of_warmups': 1, 'cudagraph_capture_sizes': [1, 2, 4, 8, 16, 24, 32, 40, 48, 56, 64, 72, 80, 88, 96, 104, 112, 120, 128, 136, 144, 152, 160, 168, 176, 184, 192, 200, 208, 216, 224, 232, 240, 248, 256, 272, 288, 304, 320, 336, 352, 368, 384, 400, 416, 432, 448, 464, 480, 496, 512], 'cudagraph_copy_inputs': False, 'cudagraph_specialize_lora': True, 'use_inductor_graph_partition': False, 'pass_config': {'fuse_norm_quant': False, 'fuse_act_quant': False, 'fuse_attn_quant': False, 'fuse_allreduce_rms': False, 'enable_qk_norm_rope_fusion': False, 'fuse_rope_kvcache_cat_mla': False, 'fuse_act_padding': False, 'fuse_qk_norm_rope_kvcache': False}, 'max_cudagraph_capture_size': 512, 'dynamic_shapes_config': {'type': <DynamicShapesType.BACKED: 'backed'>, 'evaluate_guards': False, 'assume_32_bit_indexing': False}, 'local_cache_dir': None, 'fast_moe_cold_start': False, 'static_all_moe_layers': []}, kernel_config=KernelConfig(ir_op_priority=IrOpPriorityConfig(rms_norm=['native'], fused_add_rms_norm=['native'], gelu_and_mul_sparse=['triton', 'native']), enable_flashinfer_autotune=True, enable_rocm_segmented_attn_autotune=False, enable_cutedsl_warmup=True, enable_jit_warmup=True, moe_backend='auto', sparse_indexer_topk_backend='auto', linear_backend='auto', linear_backend_per_quant=None)
|
| 8 |
+
(EngineCore pid=552081) INFO 11-14 07:27:32 [parallel_state.py:1906] world_size=1 rank=0 local_rank=0 distributed_init_method=file:///fsx_0/user/lyw/projects/jiaruiliu/adaptive-reasoning-loss/tmp/vllm_dist_9ca288f42b8d42fbbc526f1cf1b4c514 backend=nccl
|
| 9 |
+
[rank0]:[W1114 07:27:33.580338892 Utils.hpp:137] Warning: Environment variable TORCH_NCCL_TRACE_BUFFER_SIZE is deprecated; use TORCH_FR_BUFFER_SIZE instead (function operator())
|
| 10 |
+
(EngineCore pid=552081) INFO 11-14 07:27:33 [parallel_state.py:2350] rank 0 in world size 1 is assigned as DP rank 0, PP rank 0, PCP rank 0, TP rank 0, ETP rank 0, EP rank N/A, EPLB rank N/A
|
| 11 |
+
(EngineCore pid=552081) INFO 11-14 07:27:33 [gpu_worker.py:518] Using V2 Model Runner
|
| 12 |
+
(EngineCore pid=552081) WARNING 11-14 07:27:33 [flashinfer.py:143] FlashInfer's precompiled kernels are missing or do not match flashinfer-python 0.7.0.post1 (flashinfer-cubin 0.7.0.post1, flashinfer-jit-cache not installed). FlashInfer then downloads and compiles kernels at startup, which can take several minutes, or fails to import mismatched ones. Run `vllm download-kernels` in this Python environment to install them.
|
| 13 |
+
(EngineCore pid=552081) INFO 11-14 07:27:34 [model_runner.py:392] Loading model from scratch...
|
| 14 |
+
(EngineCore pid=552081) INFO 11-14 07:27:38 [cuda.py:506] Using FLASH_ATTN attention backend out of potential backends: ['FLASH_ATTN', 'FLASHINFER', 'TRITON_ATTN', 'FLEX_ATTENTION'].
|
| 15 |
+
(EngineCore pid=552081) INFO 11-14 07:27:38 [flash_attn.py:1117] Using FlashAttention version 3
|
| 16 |
+
(EngineCore pid=552081) INFO 11-14 07:27:39 [base_loader.py:82] Initializing model took 5.206213 seconds
|
| 17 |
+
(EngineCore pid=552081) INFO 11-14 07:27:39 [weight_utils.py:1108] Filesystem type for checkpoints: LUSTRE. Checkpoint size: 3.78 GiB. Available RAM: 1257.47 GiB.
|
| 18 |
+
(EngineCore pid=552081) INFO 11-14 07:27:39 [weight_utils.py:1070] Prefetching checkpoint files into page cache started (in background, num_threads=8, block_size=16777216 bytes)
|
| 19 |
+
|
| 20 |
+
(EngineCore pid=552081) INFO 11-14 07:27:42 [weight_utils.py:1042] Prefetching checkpoint files: 10% (1/1)
|
| 21 |
+
(EngineCore pid=552081) INFO 11-14 07:27:42 [weight_utils.py:1065] Prefetching checkpoint files into page cache finished in 2.91s
|
| 22 |
+
|
| 23 |
+
|
| 24 |
+
|
| 25 |
+
(EngineCore pid=552081) INFO 11-14 07:27:42 [default_loader.py:501] Loading weights took 3.42 seconds
|
| 26 |
+
(EngineCore pid=552081) INFO 11-14 07:27:45 [model_runner.py:424] Model loading took 3.21 GiB memory and 11.402768 seconds
|
| 27 |
+
(EngineCore pid=552081) INFO 11-14 07:27:45 [topk_topp_sampler.py:124] Using FlashInfer for top-p & top-k sampling.
|
| 28 |
+
(EngineCore pid=552081) INFO 11-14 07:27:45 [torch_utils.py:295] Reducing Torch threads from 8 to 1 for serving. Set OMP_NUM_THREADS in the external environment to override.
|
| 29 |
+
(EngineCore pid=552081) INFO 11-14 07:27:45 [utils.py:347] Using LBNHC KV cache layout.
|
| 30 |
+
(EngineCore pid=552081) INFO 11-14 07:27:54 [caching.py:342] reconstructed serializable fn from standalone compile artifacts. num_artifacts=3 num_submods=29
|
| 31 |
+
(EngineCore pid=552081) INFO 11-14 07:27:54 [decorators.py:319] [backbone] Directly load AOT compilation from path /home/lyw/.cache/vllm/torch_compile_cache/torch_aot_compile/2809924cc999adfdedb34e1a3bbbade72742e10f29292501e7555c8474ef5651/rank_0_0/model
|
| 32 |
+
(EngineCore pid=552081) INFO 11-14 07:27:54 [monitor.py:62] [backbone] torch.compile took 1.05 s in total
|
| 33 |
+
(EngineCore pid=552081) INFO 11-14 07:27:54 [monitor.py:91] [backbone] Initial profiling/warmup run took 0.31 s
|
| 34 |
+
|
| 35 |
+
|
| 36 |
+
(EngineCore pid=552081) INFO 11-14 07:28:00 [model_runner.py:1127] Graph capturing finished in 4 secs, took 0.41 GiB
|
| 37 |
+
(EngineCore pid=552081) INFO 11-14 07:28:01 [gpu_worker.py:759] Available KV cache memory: 61.82 GiB
|
| 38 |
+
(EngineCore pid=552081) INFO 11-14 07:28:01 [gpu_worker.py:774] CUDA graph memory profiling is enabled (default since v0.21.0). The current --gpu-memory-utilization=0.8500 is equivalent to --gpu-memory-utilization=0.8427 without CUDA graph memory profiling. To maintain the same effective KV cache size as before, increase --gpu-memory-utilization to 0.8573. To disable, set VLLM_MEMORY_PROFILER_ESTIMATE_CUDAGRAPHS=0.
|
| 39 |
+
(EngineCore pid=552081) INFO 11-14 07:28:01 [kv_cache_utils.py:2327] GPU KV cache size: 578,752 tokens, Maximum concurrency for 6,144 tokens per request: 94.20x
|
| 40 |
+
(EngineCore pid=552081) INFO 11-14 07:28:01 [kernel_warmup.py:176] JIT kernel warmup starting.
|
| 41 |
+
(EngineCore pid=552081) INFO 11-14 07:28:01 [kernel_warmup.py:189] JIT kernel warmup finished in 0.00s.
|
| 42 |
+
(EngineCore pid=552081) INFO 11-14 07:28:01 [kernel.py:467] Final IR op priority after setting platform defaults: IrOpPriorityConfig(rms_norm=['native'], fused_add_rms_norm=['native'], gelu_and_mul_sparse=['triton', 'native'])
|
| 43 |
+
(EngineCore pid=552081) INFO 11-14 07:28:01 [kernel_warmup.py:460] Using FlashInfer autotune cache file: /home/lyw/.cache/vllm/flashinfer_autotune_cache/0.7.0.post1/90a/964c865b329908340a1b6c2d0c5735257169c1f8df4702b5c9cc66707262a058/autotune_configs_dp0_rank0.json
|
| 44 |
+
2026-11-14 07:28:01,572 - INFO - autotuner.py:1046 - flashinfer.jit: [Autotuner]: Autotuning process starts ...
|
| 45 |
+
(EngineCore pid=552081) INFO 11-14 07:28:01 [kernel_warmup.py:367] Running FlashInfer autotune with 16384 tokens and token buckets (1, 2, 4, 8, 16, 32, 64, 128, 256, 512, 768, 1024, 1280, 1536, 1792, 2048, 2560, 3072, 3584, 4096, 8192, 16384).
|
| 46 |
+
2026-11-14 07:28:01,620 - INFO - autotuner.py:1069 - flashinfer.jit: [Autotuner]: Autotuning process ends
|
| 47 |
+
|
| 48 |
+
|
| 49 |
+
(EngineCore pid=552081) INFO 11-14 07:28:09 [model_runner.py:1127] Graph capturing finished in 5 secs, took 0.27 GiB
|
| 50 |
+
(EngineCore pid=552081) INFO 11-14 07:28:09 [gpu_worker.py:971] CUDA graph pool memory: 0.27 GiB (actual), 0.58 GiB (estimated), difference: 0.3 GiB (110.7%).
|
| 51 |
+
(EngineCore pid=552081) INFO 11-14 07:28:09 [gpu_worker.py:1034] Free memory on device (78.66/79.18 GiB) on startup. Desired GPU memory utilization is (0.85, 67.3 GiB). Actual usage is 4.16 GiB for consumed memory (weights + non-torch), 1.33 GiB for peak activation, and 0.27 GiB for CUDAGraph memory. Replace gpu_memory_utilization config with `--kv-cache-memory=65925157991` (61.4 GiB) to fit into requested memory, or `--kv-cache-memory=78125379072` (72.76 GiB) to fully utilize gpu memory. Current kv cache memory in use is 61.82 GiB.
|
| 52 |
+
(EngineCore pid=552081) INFO 11-14 07:28:14 [jit_monitor.py:84] Kernel JIT monitor activated; monitored JIT compilations during inference will use mode=warn.
|
| 53 |
+
(EngineCore pid=552081) INFO 11-14 07:28:15 [core.py:395] init engine (profile, create kv cache, warmup model) took 30.07 s (compilation: 1.05 s)
|
| 54 |
+
(EngineCore pid=552081) INFO 11-14 07:28:15 [kernel.py:467] Final IR op priority after setting platform defaults: IrOpPriorityConfig(rms_norm=['native'], fused_add_rms_norm=['native'], gelu_and_mul_sparse=['triton', 'native'])
|
| 55 |
+
(MainProcess pid=547852) INFO 11-14 07:28:16 [hf.py:648] Detected the chat template content format to be 'string'. You can set `--chat-template-content-format` to override this.
|
| 56 |
+
|
| 57 |
+
(EngineCore pid=552081) WARNING 11-14 07:28:16 [jit_monitor.py:140] Triton kernel JIT compilation during inference: _topp_sb_step_kernel. This causes a latency spike; consider extending warmup to cover this shape/config.
|
| 58 |
+
(EngineCore pid=552081) WARNING 11-14 07:28:16 [jit_monitor.py:140] Triton kernel JIT compilation during inference: _topp_sb_mask_kernel. This causes a latency spike; consider extending warmup to cover this shape/config.
|
| 59 |
+
|
| 60 |
+
|
| 61 |
+
|
| 62 |
+
|
| 63 |
+
wrote /fsx_0/user/lyw/projects/arl_data/runs/s1_numina_q17/asft_lr2e-5/eval/gsm8k.jsonl
|
| 64 |
+
(MainProcess pid=547852) INFO 11-14 07:28:44 [utils.py:628] [shutdown] Process manager: send sigterm to process EngineCore
|
| 65 |
+
(EngineCore pid=552081) INFO 11-14 07:28:44 [core.py:1454] [shutdown] EngineCore: trigger received signal=SIGTERM
|
| 66 |
+
(EngineCore pid=552081) INFO 11-14 07:28:44 [core.py:1635] [shutdown] EngineCore: start mode=abort timeout=0s
|
| 67 |
+
(EngineCore pid=552081) INFO 11-14 07:28:44 [core.py:1666] [shutdown] EngineCore: request processing complete; starting resource teardown
|
| 68 |
+
(EngineCore pid=552081) INFO 11-14 07:28:44 [core.py:1499] [shutdown] EngineCore: exiting busy loop
|
| 69 |
+
Timeout during comparison
|
| 70 |
+
aime24 {"pass1": 0.0125, "n": 16, "kind": "math", "mean_tokens": 1256.2208333333333, "truncated": 0.17916666666666667, "pass4": 0.04357142857142857, "pass8": 0.07222222222222222, "pass16": 0.1}
|
| 71 |
+
Timeout during comparison
|
| 72 |
+
Timeout during comparison
|
| 73 |
+
Timeout during comparison
|
| 74 |
+
aime25 {"pass1": 0.016666666666666666, "n": 16, "kind": "math", "mean_tokens": 1092.6708333333333, "truncated": 0.13541666666666666, "pass4": 0.05857142857142857, "pass8": 0.09777777777777777, "pass16": 0.13333333333333333}
|
| 75 |
+
amc23 {"pass1": 0.2796875, "n": 16, "kind": "math", "mean_tokens": 866.209375, "truncated": 0.0890625, "pass4": 0.4681043956043956, "pass8": 0.578568376068376, "pass16": 0.7}
|
| 76 |
+
gsm8k {"pass1": 0.7460197119029568, "n": 2, "kind": "math", "mean_tokens": 215.50151630022745, "truncated": 0.001516300227445034}
|
| 77 |
+
Timeout during comparison
|
| 78 |
+
Timeout during comparison
|
| 79 |
+
math500 {"pass1": 0.541, "n": 4, "kind": "math", "mean_tokens": 518.339, "truncated": 0.044, "pass4": 0.706}
|
| 80 |
+
Timeout during comparison
|
| 81 |
+
Timeout during comparison
|
| 82 |
+
Timeout during comparison
|
| 83 |
+
Timeout during comparison
|
| 84 |
+
minerva {"pass1": 0.16636029411764705, "n": 4, "kind": "math", "mean_tokens": 707.5, "truncated": 0.06066176470588235, "pass4": 0.3014705882352941}
|
| 85 |
+
olympiadbench {"pass1": 0.21624629080118693, "n": 4, "kind": "math", "mean_tokens": 919.7603857566766, "truncated": 0.08123145400593472, "pass4": 0.37833827893175076}
|
runs/s1_numina_q17/asft_lr2e-5/eval.sbatch
ADDED
|
@@ -0,0 +1,13 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
#!/bin/bash
|
| 2 |
+
#SBATCH --job-name=arl-s1_numina_q17-asft_lr2e-5-eval
|
| 3 |
+
#SBATCH --gres=gpu:1
|
| 4 |
+
#SBATCH --cpus-per-task=8
|
| 5 |
+
#SBATCH --mem=150G
|
| 6 |
+
#SBATCH --time=3:00:00
|
| 7 |
+
#SBATCH --requeue
|
| 8 |
+
#SBATCH --output=/fsx_0/user/lyw/projects/arl_data/runs/s1_numina_q17/asft_lr2e-5/eval.log
|
| 9 |
+
export HF_HUB_OFFLINE=1 TRANSFORMERS_OFFLINE=1 ARL_EVAL_SETS=/fsx_0/user/lyw/projects/arl_data/sft/v1/eval_sets NLTK_DATA=/fsx_0/user/lyw/projects/envs/nltk_data
|
| 10 |
+
cd /fsx_0/user/lyw/projects/arl_data/runs/s1_numina_q17/code-9c038dc
|
| 11 |
+
[ -e /fsx_0/user/lyw/projects/arl_data/runs/s1_numina_q17/asft_lr2e-5/final/config.json ] || { echo "no final model"; exit 1; }
|
| 12 |
+
PYTHONPATH=/fsx_0/user/lyw/projects/arl_data/runs/s1_numina_q17/code-9c038dc:/fsx_0/user/lyw/projects/envs/shim:/fsx_0/user/lyw/projects/envs/vllm32-py312 python3 -s -m arl.eval.generate --model /fsx_0/user/lyw/projects/arl_data/runs/s1_numina_q17/asft_lr2e-5/final --out /fsx_0/user/lyw/projects/arl_data/runs/s1_numina_q17/asft_lr2e-5/eval --bench math500:4,minerva:4,olympiadbench:4,amc23:16,aime24:16,aime25:16,gsm8k:2 --max_tokens 4096 --template plain
|
| 13 |
+
PYTHONPATH=/fsx_0/user/lyw/projects/arl_data/runs/s1_numina_q17/code-9c038dc:/fsx_0/user/lyw/projects/envs/arl-py312:/fsx_0/user/lyw/projects/envs/awe-py312:/fsx_0/user/lyw/projects/envs/arl-eval-py312 python3 -s -m arl.eval.grade --gen_dir /fsx_0/user/lyw/projects/arl_data/runs/s1_numina_q17/asft_lr2e-5/eval
|
runs/s1_numina_q17/asft_lr2e-5/eval/aime24.jsonl
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
runs/s1_numina_q17/asft_lr2e-5/eval/aime25.jsonl
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
runs/s1_numina_q17/asft_lr2e-5/eval/amc23.jsonl
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
runs/s1_numina_q17/asft_lr2e-5/eval/gsm8k.jsonl
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
runs/s1_numina_q17/asft_lr2e-5/eval/math500.jsonl
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
runs/s1_numina_q17/asft_lr2e-5/eval/minerva.jsonl
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
runs/s1_numina_q17/asft_lr2e-5/eval/olympiadbench.jsonl
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|