---
library_name: transformers
tags:
- code
datasets:
- Leon-Leee/wizardlm_evol_instruct_v2_196K_backuped
- m-a-p/Code-Feedback
- openbmb/UltraInteract_sft
- ise-uiuc/Magicoder-Evol-Instruct-110K
- flytech/python-codes-25k
metrics:
- code_eval
pipeline_tag: text-generation
license: other
license name: deepseek
---
## AIGCodeGeek-DS-6.7B

### Introduction
AIGCodeGeek-DS-6.7B is our first released version of a Code-LLM family with competitive performance on public and private benchmarks.

### Model Details
#### Model Description
- Developed by: [Leon Li](https://huggingface.co/Leon-Leee)
- License: [DeepSeek](https://github.com/deepseek-ai/DeepSeek-Coder/blob/main/LICENSE-MODEL)
- Fine-tuned from [deepseek-ai/deepseek-coder-6.7b-base](https://huggingface.co/deepseek-ai/deepseek-coder-6.7b-base) with full parameters

### Training data
A mixture of samples from high-quality open-source (read *Acknowledgements*) and our private datasets. 
We have made contamination detection as Magicoder/Bigcode did (https://github.com/ise-uiuc/magicoder/blob/main/src/magicoder/decontamination/find_substrings.py).

### Evaluation
results to be added.

### Requirements
It should work with the same requirements as DeepSeek-Coder-6.7B or the following packages:

```torch>=2.0
tokenizers>=0.14.0
transformers>=4.35.0
accelerate
sympy>=1.12
pebble 
timeout-decorator 
attrdict
```


### QuickStart

```python
import torch
from transformers import AutoTokenizer, AutoModelForCausalLM
tokenizer = AutoTokenizer.from_pretrained("aigcode/AIGCodeGeek-DS-6.7B", trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained("aigcode/AIGCodeGeek-DS-6.7B", trust_remote_code=True, torch_dtype=torch.bfloat16).cuda()
messages=[
    { 'role': 'user', 'content': "write a merge sort algorithm in python."}
]
inputs = tokenizer.apply_chat_template(messages, add_generation_prompt=True, return_tensors="pt").to(model.device)
# tokenizer.eos_token_id is the id of <|EOT|> token
outputs = model.generate(inputs, max_new_tokens=512, do_sample=False, top_k=50, top_p=0.95, num_return_sequences=1, eos_token_id=tokenizer.eos_token_id)
print(tokenizer.decode(outputs[0][len(inputs[0]):], skip_special_tokens=True))
```


### Acknowledgements
We gain a lot of knowledge and resources from the open-source community:
- [DeepSeekCoder](https://huggingface.co/deepseek-ai): impressive model series and insightful tech reports
- [WizardCoder](https://github.com/nlpxucan/WizardLM/tree/main/WizardCoder): Evol Instruct and public datasets
  - We used a ([Leon-Leee/wizardlm_evol_instruct_v2_196K_backuped](https://huggingface.co/datasets/Leon-Leee/wizardlm_evol_instruct_v2_196K_backuped)) since this original has been deleted.
- [Magicoder](https://github.com/ise-uiuc/magicoder/): OSS-Instruct, [Magicoder-Evol-Instruct-110K](https://huggingface.co/datasets/ise-uiuc/Magicoder-Evol-Instruct-110K) from theblackcat102/evol-codealpaca-v1(https://huggingface.co/datasets/theblackcat102/evol-codealpaca-v1)
- [Eurus](https://github.com/OpenBMB/Eurus): creative datasets for reasoning, [openbmb/UltraInteract_sft](https://huggingface.co/datasets/openbmb/UltraInteract_sft)
- [OpenCoderInterpreter](https://opencodeinterpreter.github.io/): well-designed system and datasets [m-a-p/Code-Feedback](https://huggingface.co/datasets/m-a-p/Code-Feedback)
- [flytech/python-codes-25k](https://huggingface.co/datasets/flytech/python-codes-25k): diversity
- [LLaMA-Factory](https://github.com/hiyouga/LLaMA-Factory): easily used to finetune base models