3

Dockerfile

@@ -1,13 +1,12 @@
 FROM python:3.9
 
-RUN wget -q zephyr-7b-beta.Q4_K_S.gguf https://huggingface.co/TheBloke/zephyr-7B-beta-GGUF/resolve/main/zephyr-7b-beta.Q4_K_S.gguf
+#RUN wget -q zephyr-7b-beta.Q4_K_S.gguf https://huggingface.co/TheBloke/zephyr-7B-beta-GGUF/resolve/main/zephyr-7b-beta.Q4_K_S.gguf
 
 COPY requirements.txt ./requirements.txt
 
 RUN python -m pip install -U pip && \
     python -m pip install -r requirements.txt && \
     python -m pip cache purge
-#RUN wget -O /code/zephyr-7b-beta.Q4_K_S.gguf https://huggingface.co/TheBloke/zephyr-7B-beta-GGUF/resolve/main/zephyr-7b-beta.Q4_K_S.gguf
 
 COPY ./main.py /code/main.py
 WORKDIR /code

main.py

@@ -1,7 +1,9 @@
+import os
+
 from ctransformers import AutoModelForCausalLM
 from fastapi import FastAPI
 from pydantic import BaseModel
-
+import requests
 #
 # llm = AutoModelForCausalLM.from_pretrained("zephyr-7b-beta.Q4_K_S.gguf",
 # model_type='mistral',
@@ -15,6 +17,16 @@ class validation(BaseModel):
 #Fast API
 
 app = FastAPI()
+file_name = "zephyr-7b-beta.Q4_K_S.gguf"
+
+if not os.path.exists(file_name):
+    print("Downloading model...")
+    url = "https://huggingface.co/TheBloke/zephyr-7B-beta-GGUF/resolve/main/zephyr-7b-beta.Q4_K_S.gguf"
+    response = requests.get(url)
+    print(response.status_code)
+    with open(file_name, 'wb') as file:
+        file.write(response.content)
+
 
 @app.post("/llm_on_cpu")
 async def stream(item: validation):

requirements.txt

@@ -4,4 +4,5 @@ pydantic
 uvicorn
 requests
 python-dotenv
-ctransformers
+ctransformers
+requests