Fix: eager attention (FlashAttention2 not available)

app.py

@@ -42,7 +42,7 @@ model = AutoModelForCausalLM.from_pretrained(
     device_map="auto",
     trust_remote_code=True,
     torch_dtype=torch.bfloat16,
-    attn_implementation="flash_attention_2",
+    attn_implementation="eager",
 )
 print("✅ Modell geladen!")