Add qwen2 fp8 support (#210)

Signed-off-by: changwang <changwang@habana.ai>
Co-authored-by: changwang <changwang@habana.ai>
This commit is contained in:
Wang, Chang 2024-08-26 17:02:58 +08:00 committed by GitHub
parent e33db1877c
commit 55d60a103c
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194

View File

@ -691,9 +691,9 @@ class CausalLM(Model):
"return_dict": True,
}
if model.config.model_type in ["llama", "mistral", "starcoder2"]:
if model.config.model_type in ["llama", "mistral", "starcoder2", "qwen2"]:
if model.config.model_type in ["llama", "mistral"]:
if model.config.model_type in ["llama", "mistral", "qwen2"]:
kwargs["attn_softmax_bf16"] = True
kwargs["trim_logits"] = True