Deploy Qwen2.5 Coder 3B Gradio app

README.md

@@ -14,5 +14,5 @@ pinned: false
 A coding assistant using Qwen2.5-Coder-3B-Instruct with Hugging Face ZeroGPU.
 The model is downloaded from Hugging Face on first startup. Inference requests
 use the largest ZeroGPU allocation; xlarge requests consume twice the daily
-GPU quota of the default large size. Transformers and Accelerate are pinned to
-versions compatible with the ZeroGPU runtime dependencies.
+GPU quota of the default large size. Transformers is pinned to a version
+compatible with the ZeroGPU runtime dependencies.

app.py

@@ -20,7 +20,6 @@ tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)
 model = AutoModelForCausalLM.from_pretrained(
     MODEL_ID,
     torch_dtype=torch.float16,
-    low_cpu_mem_usage=True,
     attn_implementation="sdpa",
 )
 model.to("cuda")

requirements.txt

@@ -1,2 +1 @@
 transformers==4.51.3
-accelerate==1.7.0