biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[CI] Prune tests/models/decoder_only/language/* tests (#9940)

Browse Source 
Signed-off-by: mgoin <michael@neuralmagic.com>
This commit is contained in:
Michael Goin
2024-11-05 16:02:23 -05:00
committed by GitHub
parent b9c64c0ca7
commit 02462465ea
9 changed files with 70 additions and 270 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
tests/models/decoder_only/language/test_fp8.py
View File

@@ -21,11 +21,11 @@ os.environ["TOKENIZERS_PARALLELISM"] = "true"
"kv_cache_dtype,base_model,test_model,scale_path",
[
# Test FP8 checkpoint w. fp8_e4m3 kv-cache scaling factors.
("fp8_e4m3", "meta-llama/Meta-Llama-3-8B-Instruct",
"nm-testing/Meta-Llama-3-8B-Instruct-FP8-KV", None),
("fp8_e4m3", "meta-llama/Llama-3.2-1B-Instruct",
"nm-testing/Llama-3.2-1B-Instruct-FP8-KV", None),
# Test FP16 checkpoint w. fp8_e5m2 kv-cache.
("fp8_e5m2", "meta-llama/Meta-Llama-3-8B-Instruct",
"meta-llama/Meta-Llama-3-8B-Instruct", None),
("fp8_e5m2", "meta-llama/Llama-3.2-1B-Instruct",
"meta-llama/Llama-3.2-1B-Instruct", None),
# Test FP16 checkpoint w. fp8_e4m3 kv-cache scaling factors in json.
("fp8_e4m3", "meta-llama/Llama-2-7b-chat-hf",
"meta-llama/Llama-2-7b-chat-hf",
@@ -33,7 +33,7 @@ os.environ["TOKENIZERS_PARALLELISM"] = "true"
])
# Due to low-precision numerical divergence, we only test logprob of 4 tokens
@pytest.mark.parametrize("max_tokens", [4])
@pytest.mark.parametrize("enforce_eager", [False, True])
@pytest.mark.parametrize("enforce_eager", [True])
@pytest.mark.parametrize("backend", ["FLASH_ATTN", "XFORMERS", "FLASHINFER"])
# NOTE: Increasing this in this suite will fail CI because we currently cannot
# reset distributed env properly. Use a value > 1 just when you test.