up

app-fast.py

@@ -51,7 +51,7 @@ text_encoder = AutoModelForCausalLM.from_pretrained(
     LLAMA_MODEL_NAME,
     torch_dtype=torch.bfloat16,
     low_cpu_mem_usage=True,
-    device_map="auto",
+    device_map=None,
     output_hidden_states=True,
     output_attentions=True,
     quantization_config=quantization_config,
@@ -84,7 +84,10 @@ pipe.transformer = transformer
 
 @spaces.GPU(duration=120)
 def generate_image(
-    prompt: str, resolution: str, seed: int, progress=gr.Progress(track_tqdm=True)
+    prompt: str,
+    resolution: str,
+    seed: int,
+    progress=gr.Progress(track_tqdm=True),  # noqa: ARG001, B008
 ) -> tuple[PIL.Image.Image, int]:
     if seed == -1:
         seed = torch.randint(0, 1_000_000, (1,)).item()