ayo/ipex-llm
1
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions 1

[LLM] Fix performance tests (#9596)

Browse source 
* Fix missing key for cpu_embedding

* Remove 512 as it stuck for now

* Small fix
This commit is contained in:
Yuwen Hu 2023-12-05 10:59:28 +08:00 committed by GitHub
parent 8c8a27ded7
commit 1012507a40
6 changed files with 10 additions and 5 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

2
python/llm/test/benchmark/arc-perf-test.yaml
View file

@ -26,4 +26,4 @@ in_out_pairs:
- '2048-256' - '2048-256'
test_api: test_api:
- "transformer_int4_gpu" # on Intel GPU - "transformer_int4_gpu" # on Intel GPU
cpu_embedding: False # whether put embedding to CPU (only avaiable now for gpu win related test_api)

1
python/llm/test/benchmark/arc-perf-transformers-434.yaml
View file

@ -12,3 +12,4 @@ in_out_pairs:
- '2048-256' - '2048-256'
test_api: test_api:
- "transformer_int4_gpu" # on Intel GPU - "transformer_int4_gpu" # on Intel GPU
cpu_embedding: False # whether put embedding to CPU (only avaiable now for gpu win related test_api)

3
python/llm/test/benchmark/core-perf-test.yaml
View file

@ -25,4 +25,5 @@ test_api:
# - "transformer_int4_gpu" # on Intel GPU # - "transformer_int4_gpu" # on Intel GPU
# - "optimize_model_gpu" # on Intel GPU # - "optimize_model_gpu" # on Intel GPU
# - "deepspeed_transformer_int4_cpu" # on Intel SPR Server # - "deepspeed_transformer_int4_cpu" # on Intel SPR Server
# - "transformer_int4_gpu_win" # on Intel GPU for Windows (catch GPU peak memory)
cpu_embedding: False # whether put embedding to CPU (only avaiable now for gpu win related test_api)

3
python/llm/test/benchmark/cpu-perf-test.yaml
View file

@ -16,3 +16,6 @@ test_api:
# - "ipex_fp16_gpu" # on Intel GPU # - "ipex_fp16_gpu" # on Intel GPU
# - "transformer_int4_gpu" # on Intel GPU # - "transformer_int4_gpu" # on Intel GPU
# - "optimize_model_gpu" # on Intel GPU # - "optimize_model_gpu" # on Intel GPU
# - "deepspeed_transformer_int4_cpu" # on Intel SPR Server
# - "transformer_int4_gpu_win" # on Intel GPU for Windows (catch GPU peak memory)
cpu_embedding: False # whether put embedding to CPU (only avaiable now for gpu win related test_api)

2
python/llm/test/benchmark/igpu-perf-test-434.yaml
View file

@ -7,7 +7,7 @@ num_beams: 1 # default to greedy search
low_bit: 'sym_int4' # default to use 'sym_int4' (i.e. symmetric int4) low_bit: 'sym_int4' # default to use 'sym_int4' (i.e. symmetric int4)
in_out_pairs: in_out_pairs:
- '32-32' - '32-32'
- '512-64' # - '512-64'
# - '1024-128' # - '1024-128'
test_api: test_api:
# - "transformer_int4" # - "transformer_int4"

2
python/llm/test/benchmark/igpu-perf-test.yaml
View file

@ -18,7 +18,7 @@ num_beams: 1 # default to greedy search
low_bit: 'sym_int4' # default to use 'sym_int4' (i.e. symmetric int4) low_bit: 'sym_int4' # default to use 'sym_int4' (i.e. symmetric int4)
in_out_pairs: in_out_pairs:
- '32-32' - '32-32'
- '512-64' # - '512-64'
# - '1024-128' # - '1024-128'
test_api: test_api:
# - "transformer_int4" # - "transformer_int4"