Cleanup and fix issues with text encoder quants. (#10872)
This commit is contained in:
@@ -18,6 +18,9 @@ def llama_detect(state_dict, prefix=""):
|
||||
if scaled_fp8_key in state_dict:
|
||||
out["llama_scaled_fp8"] = state_dict[scaled_fp8_key].dtype
|
||||
|
||||
if "_quantization_metadata" in state_dict:
|
||||
out["llama_quantization_metadata"] = state_dict["_quantization_metadata"]
|
||||
|
||||
return out
|
||||
|
||||
|
||||
|
||||
Reference in New Issue
Block a user