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

[CI/Test] improve robustness of test (vllm_runner) (#5357)

Browse Source 
[CI/Test] improve robustness of test by replacing del with context manager (vllm_runner) (#5357)
This commit is contained in:
youkaichao
2024-06-08 01:59:20 -07:00
committed by GitHub
parent 9fb900f90c
commit 8ea5e44a43
28 changed files with 431 additions and 470 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
tests/quantization/test_fp8.py
View File

@@ -16,9 +16,9 @@ capability = capability[0] * 10 + capability[1]
capability < QUANTIZATION_METHODS["fp8"].get_min_capability(),
reason="FP8 is not supported on this GPU type.")
def test_load_fp16_model(vllm_runner) -> None:
llm = vllm_runner("facebook/opt-125m", quantization="fp8")
with vllm_runner("facebook/opt-125m", quantization="fp8") as llm:
model = llm.model.llm_engine.model_executor.driver_worker.model_runner.model
fc1 = model.model.decoder.layers[0].fc1
assert isinstance(fc1.quant_method, Fp8LinearMethod)
assert fc1.weight.dtype == torch.float8_e4m3fn
model = llm.model.llm_engine.model_executor.driver_worker.model_runner.model # noqa: E501
fc1 = model.model.decoder.layers[0].fc1
assert isinstance(fc1.quant_method, Fp8LinearMethod)
assert fc1.weight.dtype == torch.float8_e4m3fn