Spaces:

microsoft
/

phi-4-mini

Running

nguyenbh commited on Feb 25

Commit

88c6033

verified ·

1 Parent(s): bb1b031

update app.py

Files changed (1) hide show

app.py CHANGED Viewed

@@ -19,7 +19,7 @@ tokenizer = AutoTokenizer.from_pretrained(
 model = AutoModelForCausalLM.from_pretrained(
     model_path,
     device_map="auto",
-    attn_implementation="flash_attention_2",
     torch_dtype="auto",
     token=hf_token,
     trust_remote_code=True

 model = AutoModelForCausalLM.from_pretrained(
     model_path,
     device_map="auto",
+    attn_implementation="eager",  # "flash_attention_2",
     torch_dtype="auto",
     token=hf_token,
     trust_remote_code=True