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

[Hardware][Intel] Support compressed-tensor W8A8 for CPU backend (#7257)

Browse Source
This commit is contained in:
Li, Jiang
2024-09-12 00:46:46 +08:00
committed by GitHub
parent 3b7fea770f
commit 0b952af458
18 changed files with 686 additions and 43 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
tests/quantization/test_compressed_tensors.py
View File

@ -56,7 +56,7 @@ def test_compressed_tensors_w8a8_static_setup(vllm_runner, model_args):
assert qkv_proj.weight_scale.dtype is torch.float32
assert qkv_proj.input_scale.dtype is torch.float32
output = llm.generate_greedy("Hello my name is", max_tokens=20)
output = llm.generate_greedy(["Hello my name is"], max_tokens=20)
assert output
@ -85,7 +85,7 @@ def test_compressed_tensors_w8a8_dynanmic_per_token(vllm_runner, model_args):
assert qkv_proj.scheme.strategy == strategy
assert qkv_proj.weight.dtype is torch.int8
output = llm.generate_greedy("Hello my name is", max_tokens=20)
output = llm.generate_greedy(["Hello my name is"], max_tokens=20)
assert output