EngineX-Hygon/sglang
5
0
Fork 0
Code Issues Pull Requests Actions 7 Projects Releases Wiki Activity

Support compressed tensors fp8w8a8 (#4743)

Browse Source
This commit is contained in:
Xiaoyu Zhang
2025-03-27 04:21:25 +08:00
committed by GitHub
parent 45fdf1f7f3
commit 04e3ff6975
30 changed files with 2386 additions and 113 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
test/srt/test_nightly_gsm8k_eval.py
View File

@@ -37,9 +37,6 @@ MODEL_SCORE_THRESHOLDS = {
"neuralmagic/Mixtral-8x7B-Instruct-v0.1-FP8": 0.65,
"neuralmagic/Qwen2-72B-Instruct-FP8": 0.94,
"neuralmagic/Qwen2-57B-A14B-Instruct-FP8": 0.82,
"hugging-quants/Meta-Llama-3.1-8B-Instruct-AWQ-INT4": 0.84,
"hugging-quants/Meta-Llama-3.1-8B-Instruct-GPTQ-INT4": 0.83,
"hugging-quants/Mixtral-8x7B-Instruct-v0.1-AWQ-INT4": 0.62,
}
@@ -138,7 +135,6 @@ class TestNightlyGsm8KEval(CustomTestCase):
(parse_models(DEFAULT_MODEL_NAME_FOR_NIGHTLY_EVAL_TP2), False, True),
(parse_models(DEFAULT_MODEL_NAME_FOR_NIGHTLY_EVAL_FP8_TP1), True, False),
(parse_models(DEFAULT_MODEL_NAME_FOR_NIGHTLY_EVAL_FP8_TP2), True, True),
(parse_models(DEFAULT_MODEL_NAME_FOR_NIGHTLY_EVAL_QUANT_TP1), False, False),
]
cls.base_url = DEFAULT_URL_FOR_TEST