tzhouamandClaude Opus 5
9264ff81ab
The packaging commit shipped the code and a summary of the numbers. It left out
everything the numbers are measured *against*, which is most of what makes them
checkable.
Added:
baselines/ llama.cpp runs (thread sweeps, NUMA sweeps, the
interleaved A/B passes behind the headline decode
figure) and the quality references: each build
against its own bf16 over 12 greedy prompts.
benchmarks/.../results/ 102 artifacts backing the published figures, each
with a provenance block naming the model,
quantization, thread set and build SHAs.
docs/engines/ the engine analyses -- llama.cpp, ExecuTorch, MNN,
ncnn, vLLM, vLLM-Omni -- and the cross-engine and
device-simulation comparisons.
docs/analysis/ the full study documents the summaries distil.
docs/edge/comparisons.md where this fork sits against llama.cpp and why:
decode is llama.cpp's, prefill is ours by ~2x, and
since neither engine exceeds 39% of the measured
222 GB/s ceiling the decode gap is the execution
model rather than the kernels.
docs/edge/environment.md pins and machine, and the correction below.
On "the modified vllm": there isn't one, and the doc says so plainly because it
is the obvious assumption. The vLLM checkout is clean upstream (0 commits
ahead, 0 uncommitted) and the installed wheel is stock 0.28.0+cpu. This fork
changes vLLM's CPU *behaviour* through vLLM's own extension points at import
time -- three CPU patches in vllm_omni/patch.py, the quantization-config
registry, the IR op registry, platform hooks -- specifically so the changes
survive a wheel upgrade rather than being lost on it. One of those patches is
why "KV sizing is not a memory lever" looked true for months: CpuPlatform
overwrote kv_cache_memory_bytes unconditionally, so the setting had no effect.
There is no patch series to apply; pin the version and let patch.py do the rest.
.gitignore gains explicit exceptions: upstream's blanket `*.json` and `results/`
rules were silently eating the measurement artifacts, which in this fork are the
substance rather than build output. Binary captures stay ignored.
No vLLM or llama.cpp source is redistributed -- only output produced with them.
NOTICE records both.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01L1ax5wYLA6XbqEL1i8L9gZ
10 lines
1.8 KiB
JSON
10 lines
1.8 KiB
JSON
{"tag": "vllm-bf16", "model": "XHToken/Spark-X2.5-1.7B", "window": 512, "chunks": 143, "scored_tokens": 73073, "nll": 3.7763726822277026, "ppl": 43.657394933805286}
|
|
{"tag": "vllm-int4-g64", "model": "/data/zhoutaichang/embedding_infer/models/spark2_5-int4-g64", "window": 512, "chunks": 143, "scored_tokens": 73073, "nll": 3.871370641182274, "ppl": 48.00814294379576}
|
|
{"tag": "vllm-int4-g128", "model": "/data/zhoutaichang/embedding_infer/models/spark2_5-int4-g128", "window": 512, "chunks": 143, "scored_tokens": 73073, "nll": 3.949247471796542, "ppl": 51.89629870830851}
|
|
{"tag": "vllm-int4-g32", "model": "/data/zhoutaichang/embedding_infer/models/spark2_5-int4-g32", "window": 512, "chunks": 143, "scored_tokens": 73073, "nll": 4.041093142384169, "ppl": 56.88849605431311}
|
|
{"tag": "vllm-int4-g64-fpembed", "model": "/data/zhoutaichang/embedding_infer/models/spark2_5-int4-g64-fpembed", "window": 512, "chunks": 143, "scored_tokens": 73073, "nll": 3.884205416004657, "ppl": 48.62828784511256}
|
|
{"tag": "vllm-int4-g32-rtn", "model": "/data/zhoutaichang/embedding_infer/models/spark2_5-int4-g32-rtn", "window": 512, "chunks": 143, "scored_tokens": 73073, "nll": 4.039075346914289, "ppl": 56.7738224375464}
|
|
{"tag": "vllm-int4-g64-rtn", "model": "/data/zhoutaichang/embedding_infer/models/spark2_5-int4-g64-rtn", "window": 512, "chunks": 143, "scored_tokens": 73073, "nll": 4.089284296531948, "ppl": 59.69715095314776}
|
|
{"tag": "expanded-g32", "model": "/data/zhoutaichang/embedding_infer/models/spark2_5-int4-g32-expanded", "window": 512, "chunks": 143, "scored_tokens": 73073, "nll": 4.039537154311983, "ppl": 56.80004706364053}
|
|
{"tag": "expanded-g64", "model": "/data/zhoutaichang/embedding_infer/models/spark2_5-int4-g64-expanded", "window": 512, "chunks": 143, "scored_tokens": 73073, "nll": 3.872951761505493, "ppl": 48.084109634697604}
|