Add mlc/Dockerfile with instructions inside it.

mlc/Dockerfile

@@ -0,0 +1,58 @@
+# From the parent directory (main directory of this repo) run:
+#
+# docker build --build-arg USERID=$(id -u) -t local/mlc-bench mlc
+#
+# If not already using and having a $HOME/.cache/huggingface/ then:
+#
+# mkdir $HOME/.cache/huggingface/
+# docker run --rm -it -v$HOME/.cache/huggingface/:/home/user/.cache/huggingface/:Z local/mlc-bench \
+#   huggingface-cli login
+# Answer n to: Add token as git credential? (Y/n) n
+#
+# docker run --rm -it -v$HOME/.cache/huggingface/:/home/user/.cache/huggingface/:Z local/mlc-bench \
+#   huggingface-cli download meta-llama/Llama-2-7b-hf
+#
+# mkdir $HOME/.cache/mlc_llm/
+# docker run --rm -it -v$HOME/.cache/huggingface/:/home/user/.cache/huggingface/:Z \
+#   -v$HOME/.cache/mlc_llm/:/home/user/.cache/mlc_llm/:Z -v$(pwd):/home/user/llama-inference \
+#   --gpus all local/mlc-bench \
+#   sh -c 'cd /home/user/llama-inference/mlc && mlc_llm convert_weight --quantization q4f16_1 \
+#   -o Llama-2-7b-hf-q4f16_1 ~/.cache/huggingface/hub/models--meta-llama--Llama-2-7b-hf/snapshots/*'
+#
+# docker run --rm -it -v$HOME/.cache/huggingface/:/home/user/.cache/huggingface/:Z \
+#   -v$HOME/.cache/mlc_llm/:/home/user/.cache/mlc_llm/:Z -v$(pwd):/home/user/llama-inference \
+#   --gpus all local/mlc-bench \
+#   sh -c 'cd /home/user/llama-inference/mlc && mlc_llm gen_config --quantization q4f16_1 \
+#   --conv-template LM -o Llama-2-7b-hf-q4f16_1 \
+#   ~/.cache/huggingface/hub/models--meta-llama--Llama-2-7b-hf/snapshots/*'
+#
+# docker run --rm -it -v$HOME/.cache/huggingface/:/home/user/.cache/huggingface/:Z \
+#   -v$HOME/.cache/mlc_llm/:/home/user/.cache/mlc_llm/:Z -v$(pwd):/home/user/llama-inference \
+#   --gpus all local/mlc-bench \
+#   sh -c 'cd /home/user/llama-inference/mlc && python3 mlc.py'
+#
+# If using Podman with CDI substitute
+#   --gpus all
+# for
+#   --device nvidia.com/gpu=all --security-opt=label=disable
+
+# Select an available version from
+# https://gitlab.com/nvidia/container-images/cuda/blob/master/doc/supported-tags.md:
+# 2024-04-02 CUDNN9 not supported yet:
+#FROM nvcr.io/nvidia/cuda:12.3.2-cudnn9-devel-rockylinux9
+FROM nvcr.io/nvidia/cuda:12.2.2-cudnn8-devel-rockylinux9
+
+RUN dnf install -y python3-pip && dnf clean all && rm -rf /var/cache/dnf/*
+
+# Don't know why libcuda.so.1 isn't on another place or configured in ld.so.conf.d,
+# /usr/local/lib64/ doesn't work neither:
+RUN ln -s /usr/local/cuda-12.*/compat/libcuda.so.1 /usr/lib64/
+
+# Select a matching version from https://llm.mlc.ai/docs/install/mlc_llm.html#install-mlc-packages:
+RUN pip install --no-cache-dir --pre -f https://mlc.ai/wheels mlc-llm-nightly-cu122 \
+    mlc-ai-nightly-cu122 transformers pandas
+
+ARG USERID=1000
+RUN adduser -u $USERID user
+USER user
+

mlc/mlc.py

@@ -3,16 +3,16 @@
 from transformers import AutoTokenizer
 import time
 import sys
-sys.path.append('../../common/')
+sys.path.append('../common/')
 from questions import questions
 import pandas as pd
 
 
 
-tokenizer = AutoTokenizer.from_pretrained("meta-llama/Llama-2-7b-chat-hf")
+tokenizer = AutoTokenizer.from_pretrained("meta-llama/Llama-2-7b-hf")
 
 cfg = ChatConfig(max_gen_len=200)
-cm = ChatModule(model="Llama-2-7b-chat-hf-q4f16_1", chat_config=cfg)
+cm = ChatModule(model="Llama-2-7b-hf-q4f16_1", chat_config=cfg)
 
 def tok_count(prompt:str):
     inputs = tokenizer(prompt)
@@ -27,7 +27,7 @@ def predict(prompt:str):
             'time': request_time,
             'question': prompt,
             'answer': output,
-            'note': 'mlc chat 7b q4f16_1'}
+            'note': 'mlc 7b q4f16_1'}
 
 if __name__ == '__main__':
     counter = 1