Run app locally

Dockerfile

@@ -1,5 +1,8 @@
 FROM python:3.9
 
+ENV TQDM_DISABLE=0
+ENV TQDM_FORCE=1
+
 WORKDIR /code
 RUN mkdir -p /code && chmod -R 777 /code
 RUN mkdir -p /.cache && chmod -R 777 /.cache
@@ -13,4 +16,6 @@ COPY . .
 
 RUN python3 -c "from transformers import AutoModelForCausalLM; AutoModelForCausalLM.from_pretrained('ibm-granite/granite-34b-code-instruct-8k', cache_dir='/.cache/huggingface')"
 
-CMD ["uvicorn", "main:app", "--host", "0.0.0.0", "--port", "7860"]
+RUN uvicorn main:app --host 0.0.0.0 --port 7860
+
+# CMD ["uvicorn", "main:app", "--host", "0.0.0.0", "--port", "7860"]

main.py

@@ -18,12 +18,6 @@ import torch
 # torch.mps.empty_cache()
 # torch.set_num_threads(1)
 
-import os
-# os.environ["HF_HOME"] = "/.cache"
-# os.environ["TRANSFORMERS_CACHE"] = "/.cache"
-os.environ["TQDM_DISABLE"] = "0"
-os.environ["TQDM_FORCE"] = "1"
-
 from transformers.utils import logging
 logging.set_verbosity_info()
 logger = logging.get_logger()  # optional: get a logger instance if you want to customize