icer/qwen36_01
1
0
Fork 0
Code Issues Pull Requests Actions Projects Releases Wiki Activity

stabilize platform launch and record official-like baseline

Browse Source
This commit is contained in:
the magician
2026-07-15 03:23:39 +08:00
parent f49b8a51ec
commit 3cf62afaaa
4 changed files with 293 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

124
worklogs/remote_results/2026-07-15-official-like-baseline/eager_c1_r8_t256_cap40k.json Normal file
View File

@@ -0,0 +1,124 @@
{
"created_at": "2026-07-14T18:06:30",
"url": "http://127.0.0.1:1111",
"model": "llm",
"dataset": "/root/work/logs/synthetic_cumulative_8_s3_cap40k.jsonl",
"concurrency": 1,
"max_requests": 8,
"max_tokens": 256,
"wall_sec": 326.64670574851334,
"success_rate": 1.0,
"ttft_p90_sec": 9.802654626592993,
"output_tps_p10_per_request": 5.825013286992466,
"aggregate_output_tps": 5.789129254087408,
"aggregate_input_tps_uncached": 178.3302846006587,
"aggregate_cache_tps": 512.0149600674832,
"cache_hit_rate": 0.7416795639891973,
"weighted_token_throughput": 883.1090591866864,
"totals": {
"requests": 8,
"success": 8,
"prompt_tokens": 225499,
"input_tokens_uncached": 58251,
"cached_tokens": 167248,
"completion_tokens": 1891,
"reasoning_tokens": 1759
},
"targets": {
"output_tps_p10_per_request_gte_20": false,
"ttft_p90_lte_5": false,
"cache_hit_rate_gte_50pct": true,
"success_rate_gte_99pct": true,
"weighted_token_throughput_gte_8000": false
},
"results": [
{
"ok": true,
"elapsed_sec": 33.95716667920351,
"ttft_sec": 6.443204963579774,
"prompt_tokens": 7498,
"cached_tokens": 0,
"completion_tokens": 234,
"reasoning_tokens": 148,
"output_tps": 8.504773046446585,
"error": null
},
{
"ok": true,
"elapsed_sec": 34.63446234725416,
"ttft_sec": 14.952531272545457,
"prompt_tokens": 34177,
"cached_tokens": 7488,
"completion_tokens": 121,
"reasoning_tokens": 75,
"output_tps": 6.147770741636481,
"error": null
},
{
"ok": true,
"elapsed_sec": 45.5024864859879,
"ttft_sec": 3.0004746820777655,
"prompt_tokens": 35384,
"cached_tokens": 34160,
"completion_tokens": 256,
"reasoning_tokens": 256,
"output_tps": 6.02324429208427,
"error": null
},
{
"ok": true,
"elapsed_sec": 37.82375564984977,
"ttft_sec": 7.5955646354705095,
"prompt_tokens": 19274,
"cached_tokens": 7472,
"completion_tokens": 256,
"reasoning_tokens": 256,
"output_tps": 8.468915651559275,
"error": null
},
{
"ok": true,
"elapsed_sec": 46.32829406298697,
"ttft_sec": 3.3594408705830574,
"prompt_tokens": 37099,
"cached_tokens": 35376,
"completion_tokens": 256,
"reasoning_tokens": 256,
"output_tps": 5.95780387374304,
"error": null
},
{
"ok": true,
"elapsed_sec": 34.30666992627084,
"ttft_sec": 4.208318674936891,
"prompt_tokens": 14661,
"cached_tokens": 7472,
"completion_tokens": 256,
"reasoning_tokens": 256,
"output_tps": 8.505449280669623,
"error": null
},
{
"ok": true,
"elapsed_sec": 46.81570158340037,
"ttft_sec": 3.0013746675103903,
"prompt_tokens": 38199,
"cached_tokens": 37088,
"completion_tokens": 256,
"reasoning_tokens": 256,
"output_tps": 5.842837674796219,
"error": null
},
{
"ok": true,
"elapsed_sec": 47.27015720307827,
"ttft_sec": 3.0057141836732626,
"prompt_tokens": 39207,
"cached_tokens": 38192,
"completion_tokens": 256,
"reasoning_tokens": 256,
"output_tps": 5.783423048783708,
"error": null
}
]
}

99
worklogs/remote_results/2026-07-15-official-like-baseline/patch_head_lf_20260715.log Normal file
View File

@@ -0,0 +1,99 @@
[patch_ops] linked libcusolver.so -> /usr/local/corex/lib64/libcusolver.so (enables LOOP1_NATIVE)
[patch_model_runner] patched Case-1 prefix_cache_hit fix in: /usr/local/corex/lib64/python3/dist-packages/vllm/worker/model_runner.py
Looking in indexes: https://pypi.tuna.tsinghua.edu.cn/simple
Requirement already satisfied: transformers==4.55.3 in /usr/local/lib/python3.10/site-packages (4.55.3)
Requirement already satisfied: filelock in /usr/local/corex-3.2.3/lib64/python3/dist-packages (from transformers==4.55.3) (3.19.1)
Requirement already satisfied: huggingface-hub<1.0,>=0.34.0 in /usr/local/lib/python3.10/site-packages (from transformers==4.55.3) (0.35.0)
Requirement already satisfied: numpy>=1.17 in /usr/local/lib/python3.10/site-packages (from transformers==4.55.3) (1.26.4)
Requirement already satisfied: packaging>=20.0 in /usr/local/corex-3.2.3/lib64/python3/dist-packages (from transformers==4.55.3) (25.0)
Requirement already satisfied: pyyaml>=5.1 in /usr/local/corex-3.2.3/lib64/python3/dist-packages (from transformers==4.55.3) (6.0.2)
Requirement already satisfied: regex!=2019.12.17 in /usr/local/corex-3.2.3/lib64/python3/dist-packages (from transformers==4.55.3) (2025.9.1)
Requirement already satisfied: requests in /usr/local/corex-3.2.3/lib64/python3/dist-packages (from transformers==4.55.3) (2.32.5)
Requirement already satisfied: tokenizers<0.22,>=0.21 in /usr/local/lib/python3.10/site-packages (from transformers==4.55.3) (0.21.4)
Requirement already satisfied: safetensors>=0.4.3 in /usr/local/lib/python3.10/site-packages (from transformers==4.55.3) (0.6.2)
Requirement already satisfied: tqdm>=4.27 in /usr/local/corex-3.2.3/lib64/python3/dist-packages (from transformers==4.55.3) (4.67.1)
Requirement already satisfied: fsspec>=2023.5.0 in /usr/local/corex-3.2.3/lib64/python3/dist-packages (from huggingface-hub<1.0,>=0.34.0->transformers==4.55.3) (2025.9.0)
Requirement already satisfied: typing-extensions>=3.7.4.3 in /usr/local/corex-3.2.3/lib64/python3/dist-packages (from huggingface-hub<1.0,>=0.34.0->transformers==4.55.3) (4.15.0)
Requirement already satisfied: hf-xet<2.0.0,>=1.1.3 in /usr/local/lib/python3.10/site-packages (from huggingface-hub<1.0,>=0.34.0->transformers==4.55.3) (1.1.10)
Requirement already satisfied: charset_normalizer<4,>=2 in /usr/local/corex-3.2.3/lib64/python3/dist-packages (from requests->transformers==4.55.3) (3.4.3)
Requirement already satisfied: idna<4,>=2.5 in /usr/local/corex-3.2.3/lib64/python3/dist-packages (from requests->transformers==4.55.3) (3.10)
Requirement already satisfied: urllib3<3,>=1.21.1 in /usr/local/corex-3.2.3/lib64/python3/dist-packages (from requests->transformers==4.55.3) (2.5.0)
Requirement already satisfied: certifi>=2017.4.17 in /usr/local/corex-3.2.3/lib64/python3/dist-packages (from requests->transformers==4.55.3) (2025.8.3)
WARNING: Running pip as the 'root' user can result in broken permissions and conflicting behaviour with the system package manager, possibly rendering your system unusable. It is recommended to use a virtual environment instead: https://pip.pypa.io/warnings/venv. Use the --root-user-action option if you know what you are doing and want to suppress this warning.
[notice] A new release of pip is available: 25.2 -> 26.1.2
[notice] To update, run: pip install --upgrade pip
=== Patching /usr/local/lib/python3.10/site-packages/transformers/models/auto/configuration_auto.py ===
[skip] already patched: '("qwen3", "Qwen3Config"),\n ("qwen3_5", "Qwen3_5Config'
[skip] already patched: '("qwen3", "Qwen3Config"),\n ("qwen3_5", "Qwen3_5Config'
[skip] already patched: '("qwen3", "Qwen3"),\n ("qwen3_5", "Qwen3_5"),\n '
[skip] already patched: '("qwen3", "Qwen3"),\n ("qwen3_5", "Qwen3_5"),\n '
=== Patching /usr/local/lib/python3.10/site-packages/transformers/models/__init__.py ===
[skip] already patched: 'from .qwen3 import *\n from .qwen3_5 import *\n from .qw'
=== Verification ===
Qwen3_5Config() smoke-test OK (model_type=qwen3_5)
Qwen3_5MoeConfig() smoke-test OK (model_type=qwen3_5_moe)
num_experts=256, top_k=8, shared=512, layers=40
Done.
/usr/local/corex/lib64/python3/dist-packages/torch/cuda/__init__.py:51: FutureWarning: The pynvml package is deprecated. Please install nvidia-ml-py instead. If you did not install pynvml directly, please report this to the maintainers of the package that installed pynvml for you.
import pynvml # type: ignore[import]
=== Patching /usr/local/corex/lib64/python3/dist-packages/vllm/model_executor/models/registry.py ===
[skip] already patched: ' "Qwen3ForCausalLM": ("qwen3", "Qwen3ForCausalLM"),\n "Qwen3MoeFo'
=== Verification ===
INFO 07-14 17:43:30 importing.py:10] Triton not installed; certain GPU-related functions will not be available.
2026-07-14 17:43:31.766722: I tensorflow/core/util/port.cc:110] oneDNN custom operations are on. You may see slightly different numerical results due to floating-point round-off errors from different computation orders. To turn them off, set the environment variable `TF_ENABLE_ONEDNN_OPTS=0`.
2026-07-14 17:43:31.818158: I tensorflow/core/platform/cpu_feature_guard.cc:182] This TensorFlow binary is optimized to use available CPU instructions in performance-critical operations.
To enable the following instructions: SSE3 SSE4.1 SSE4.2 AVX AVX2 AVX512F AVX512_VNNI AVX512_BF16 AVX_VNNI AMX_TILE AMX_INT8 AMX_BF16 FMA, in other operations, rebuild TensorFlow with the appropriate compiler flags.
WARNING:tensorflow:Deprecation warnings have been disabled. Set TF_ENABLE_DEPRECATION_WARNINGS=1 to re-enable them.
Qwen3_5ForCausalLM found: <class 'qwen3_5.Qwen3_5ForCausalLM'>
Qwen3_5MoeForCausalLM found: <class 'qwen3_5.Qwen3_5MoeForCausalLM'>
Done. Remember to:
1. Set config.json 'architectures': ['Qwen3_5ForCausalLM'] or ['Qwen3_5MoEForCausalLM']
2. Run patch_transformers_qwen3_5.py if not already done
=== patch_xformers_sdpa_seq (sequential, pure-math) ===
Target: /usr/local/corex/lib64/python3/dist-packages/vllm/attention/backends/xformers.py
[skip] _run_sdpa_fallback already present
[warn] dispatch block anchor not found
=== patch_arg_utils (disable chunked-prefill auto-enable) ===
Target: /usr/local/corex/lib64/python3/dist-packages/vllm/engine/arg_utils.py
[skip] chunked-prefill auto-enable already disabled
=== patch_logits_processor (seq_groups=None guard for chunked prefill) ===
Target: /usr/local/corex/lib64/python3/dist-packages/vllm/model_executor/layers/logits_processor.py
[skip] seq_groups=None guard already present
Done.
=== Patching /usr/local/corex/lib/python3/dist-packages/vllm/entrypoints/openai/tool_parsers/__init__.py ===
[skip] already patched: 'from .mistral_tool_parser import MistralToolParser\nfrom .qwen3coder_to'
[skip] already patched: '"MistralToolParser", "Internlm2ToolParser", "Llama3JsonToolParser",\n '
=== Verification ===
Module spec loaded: qwen3coder_tool_parser
(full import requires torch/vllm runtime — skipping exec)
Done. Start vLLM server with:
--tool-call-parser qwen3_coder --enable-auto-tool-choice
[patch_ops] deployed multi_system chat template → /workspace/chat_template_multi_system.jinja
=== patch_ops verification ===
[ok] MOE_NATIVE (qwen3_5.py)
[ok] _native_experts_sorted
[ok] CHUNK_PARALLEL
[ok] LOOP1_NATIVE
[ok] RMSNORM_NATIVE
[ok] PREFIX_FLASH (paged_attn.py)
[ok] _forward_prefix_flash
[ok] libcusolver.so link (LOOP1_NATIVE runtime)
[ok] /workspace/chat_template_multi_system.jinja (multi_system fix)
[ok] torch.linalg.solve_triangular callable
=> all five optimizations deployed and enabled
=== patch complete. Start the server with the computility runner using: ===
../computility-run.yaml
(env + command are defined there; run from outside qwen3_6_scripts/)