Add dockerfile, requirements and code style

.dockerignore

@@ -0,0 +1,4 @@
+*
+!src
+!transformers
+!requirements.txt

.gitmodules

@@ -0,0 +1,3 @@
+[submodule "transformers"]
+path = transformers
+url = https://github.com/bigcode-project/transformers.git

Dockerfile

@@ -0,0 +1,16 @@
+FROM nvcr.io/nvidia/pytorch:22.11-py3
+
+ARG USER=1000
+ARG USERNAME=user
+
+WORKDIR /app
+ENV PYTHONPATH=/app
+
+RUN useradd -m -u $USER -s /bin/bash $USERNAME \
+ && chown $USERNAME /app
+
+COPY --chown=$USERNAME ./requirements.txt ./
+COPY --chown=$USERNAME transformers/ ./transformers
+RUN pip install -r requirements.txt
+
+COPY --chown=$USERNAME src/ ./src

Makefile

@@ -1,3 +1,9 @@
+check_dirs := src scripts
+
+style:
+black --preview $(check_dirs)
+isort $(check_dirs)
+
 batch_size := 1
 
 install-mqa-transformers:

pyproject.toml

@@ -0,0 +1,3 @@
+[tool.black]
+line-length = 119
+target-version = ['py35']

requirements.txt

@@ -0,0 +1,8 @@
+accelerate==0.15.0
+bitsandbytes
+deepspeed==0.7.7
+./transformers
+
+# TODO: Dev only
+isort>=5.5.4
+black~=22.0

src/pipelines/pipeline.py

@@ -3,6 +3,7 @@
 from typing import List, Tuple, Union
 
 import torch
+
 from transformers import AutoTokenizer, BloomConfig, BloomForCausalLM, GPT2Config, GPT2LMHeadModel