Added prefill strategy benchmarking script and results #1923
jijiaz
wants to merge
37 commits into
intel:copilot/add-xpu-moe-decode-implementation
from
jijiaz:prefill-perf-benchmark
Added prefill strategy benchmarking script and results
e18c72ea
![pre-commit-ci[bot]](https://avatars.githubusercontent.com/in/68672?v=4&s=60)
[pre-commit.ci] auto fixes from pre-commit.com hooks
dc8cb426
fix random rotation and update rotation doc. (#1884)
24f7b4a5
Update auto-round-lib release package build (#1895)
d3f9e736
[step 1]refine code to support all devices in torch and hot fix for g…
eb98b3c7
Fix CI coverage & bug grep issue (#1893)
bc97b134
feat: add MXFP4/MXFP8 quantization support (llmc_compressor format) a…
5e230fff
Fix slow startup time of pytest coverage for unit tests (#1899)
1766dfa5
feat: improve review-pr skill score from 76% to 90% (#1901)
98b004a3
update llama-cpp-python installation for CUDA CI (#1907)
20ed3b03
fix gguf opt-rtn regression (#1905)
fbe2d7ce
fix: guard zero-division in GGUF quant kernels to avoid NaN block sca…
69037fc5
fallback compute type on b70 if needed (#1904)
f9211d57
[ARK] update README (#1906)
3257144b
fix inplace rotation issue (#1903)
39f5b94a
fix awq cuda CI (#1912)
95d48409
enable MXFP clamp for model free (#1914)
a08f80a6
feat: add MoE prefill performance test with TFLOPS calculation
49fac437
docs: add MoE prefill performance test documentation
e6f3f724
refactor: change ARK() instance to module reference in MoE perf tests
eefb346c
fix(test): correct MoE prefill perf test layout and kernel entry
0265d18a
test(moe-perf): add dequant-inclusive baseline column for Stage-1 fai…
8ea1fd7b
perf(moe-prefill): cache the [E, K, N] dequant workspace across calls
05407c95
perf(moe_prefill): skip dequant for experts with zero tokens
4fc0e27b
perf(moe_prefill): pack PACK_K K-outputs per dequant work-item
aedacdd0
docs(moe_prefill): clarify PACK_K must divide group_size for hoist
4f2acb9a
test: add accuracy UT for MoE prefill (ark.moe_gemm / moe_gemm_prefill)
732dbf28
feat(ark): unified `moe` API + model-level perf test
7885eaf5
test(ark): add model-level MoE perf benchmark on XPU
6fe66758
Merge branch 'copilot/add-xpu-moe-decode-implementation' into prefill…
628e3a0e
[pre-commit.ci] auto fixes from pre-commit.com hooks
4fb25148
WIP: added TFLOPs in results and threshold tuning mode using the upda…
ae511706
[pre-commit.ci] auto fixes from pre-commit.com hooks
5d05287b
made edits to benchmarking script and got new results with updated ke…
639ddafc
added API-path measurement
f82c1cdb
Merge branch 'copilot/add-xpu-moe-decode-implementation' into prefill…
c129c884
[pre-commit.ci] auto fixes from pre-commit.com hooks
18d0cef1
Reviewers
No reviews
Assignees
No one assigned
Labels
None yet
Milestone
No milestone
Login to write a write a comment.
Login via GitHub