metadata
license: apache-2.0
datasets:
- Leon-Leee/wizardlm_evol_instruct_v2_196K_backuped
- m-a-p/Code-Feedback
- openbmb/UltraInteract_sft
- ise-uiuc/Magicoder-Evol-Instruct-110K
language:
- en
metrics:
- code_eval
library_name: transformers
tags:
- code
AIGCodeGeek-DS-6.7B
Introduction
AIGCodeGeek-DS-6.7B is the first version of our Code-LLM family with competitive performance on benchmarks such as HumanEval(+) and MBPP(+).
It gains a lot of insights from the open-source community and we deeply appreciate all of these great works.
We are preparing for the tech report, so stay tuned for more details:)
Model Details
Model Description
- Developed by: Leon Li
- License: DeepSeek
- Fine-tuned from deepseek-ai/deepseek-coder-6.7b-base with full parameters
Training data
A mixture of both
- samples from several high-quality open-source datasets (read Acknowledgements),
- our private datasets (already decontaminated with benchmarks).
Evaluation
To check out our evaluation results: EvalPlus
Requirements
It should work with the same requirements as DeepSeek-Coder-6.7B or the following packages:
tokenizers>=0.14.0
transformers>=4.35.0
accelerate
sympy>=1.12
pebble
timeout-decorator
attrdict
QuickStart
TBD
Limits
Acknowledgements
- WizardCoder: WizardLM-Evol-Instruct V2 datasets
- We used a back-up(Leon-Leee/wizardlm_evol_instruct_v2_196K_backuped) since this dataset has been deleted.
- Magicoder: Magicoder-Evol-Instruct-110K from theblackcat102/evol-codealpaca-v1(https://huggingface.co/datasets/theblackcat102/evol-codealpaca-v1)
- Eurus: reasoning enhancement dataset of openbmb/UltraInteract_sft
- OpenCoderInterpreter: m-a-p/Code-Feedback