Python-Code-33B
Large Language Models (LLMs) are good with code generations. Sometimes LLMs do make mistakes in code generation. How about if they can give detailed explanation along with the code. This is what I have tried over here. The base Llama-2 model was used for training purpose. It is trained on around 23000+ set of codes. Each set having 2 conversations. This data was generated using GPT-3.5, GPT-4 etc. This conversation is in Vicuna/ShareGPT format. Each set, along with code, has detailed explanation. I have released the data.
Training: Entire dataset was trained on Azure 4 x A100 80GB. For 3 epoch, training took 42 hours. DeepSpeed codebase was used for training purpose. This was trained on Llama-1 by Meta.
This is a full fine tuned model. Links for quantized models are given below.
GPTQ GGML & AWQ
GPTQ: Link
GGUF: Link
AWQ: Link
Example Prompt:
This is a conversation with your helpful AI assistant. AI assistant can generate Python Code along with necessary explanation.
Context
You are a helpful AI assistant.
USER: <prompt>
ASSISTANT:
Open LLM Leaderboard Evaluation Results
Detailed results can be found here
Metric | Value |
---|---|
Avg. | 55.06 |
AI2 Reasoning Challenge (25-Shot) | 56.31 |
HellaSwag (10-Shot) | 81.01 |
MMLU (5-Shot) | 54.22 |
TruthfulQA (0-shot) | 44.39 |
Winogrande (5-shot) | 75.22 |
GSM8k (5-shot) | 19.18 |
- Downloads last month
- 701
Model tree for ajibawa-2023/Python-Code-33B
Dataset used to train ajibawa-2023/Python-Code-33B
Evaluation results
- normalized accuracy on AI2 Reasoning Challenge (25-Shot)test set Open LLM Leaderboard56.310
- normalized accuracy on HellaSwag (10-Shot)validation set Open LLM Leaderboard81.010
- accuracy on MMLU (5-Shot)test set Open LLM Leaderboard54.220
- mc2 on TruthfulQA (0-shot)validation set Open LLM Leaderboard44.390
- accuracy on Winogrande (5-shot)validation set Open LLM Leaderboard75.220
- accuracy on GSM8k (5-shot)test set Open LLM Leaderboard19.180