first commit

Dockerfile

@@ -35,4 +35,4 @@ RUN pip install --no-cache-dir --upgrade -r /code/requirements.txt
 RUN pip install fastapi uvicorn
 
 
-CMD ["fastapi", "run", "main.py", "--host", "0.0.0.0", "--port", "7860"]
+CMD ["python", "main.py"]

main.py

@@ -1,3 +1,4 @@
+import uvicorn
 from transformers import AutoTokenizer, LlamaForCausalLM
 from fastapi import FastAPI
 from pydantic import BaseModel
@@ -23,3 +24,6 @@ async def stream(item: validation):
 
     var = tokenizer.batch_decode(generate_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0]
     return var
+
+if __name__ == "__main__":
+    uvicorn.run(app, host="0.0.0.0", port=7860)