Edit model card

WizardCoder: Empowering Code Large Language Models with Evol-Instruct

🏠 Home Page

πŸ€— HF Repo β€’πŸ± Github Repo β€’ 🐦 Twitter

πŸ“ƒ [WizardLM] β€’ πŸ“ƒ [WizardCoder] β€’ πŸ“ƒ [WizardMath]

πŸ‘‹ Join our Discord


[2024/01/04] πŸ”₯ We released WizardCoder-33B-V1.1 trained from deepseek-coder-33b-base, the SOTA OSS Code LLM on EvalPlus Leaderboard, achieves 79.9 pass@1 on HumanEval, 73.2 pass@1 on HumanEval-Plus, 78.9 pass@1 on MBPP, and 66.9 pass@1 on MBPP-Plus.

[2024/01/04] πŸ”₯ WizardCoder-33B-V1.1 outperforms ChatGPT 3.5, Gemini Pro, and DeepSeek-Coder-33B-instruct on HumanEval and HumanEval-Plus pass@1.

[2024/01/04] πŸ”₯ WizardCoder-33B-V1.1 is comparable with ChatGPT 3.5, and surpasses Gemini Pro on MBPP and MBPP-Plus pass@1.

Model Checkpoint Paper HumanEval HumanEval+ MBPP MBPP+ License
GPT-4-Turbo (Nov 2023) - - 85.4 81.7 83.0 70.7 -
GPT-4 (May 2023) - - 88.4 76.8 - - -
GPT-3.5-Turbo (Nov 2023) - - 72.6 65.9 81.7 69.4 -
Gemini Pro - - 63.4 55.5 72.9 57.9 -
DeepSeek-Coder-33B-instruct - - 78.7 72.6 78.7 66.7 -
WizardCoder-33B-V1.1 πŸ€— HF Link πŸ“ƒ [WizardCoder] 79.9 73.2 78.9 66.9 MSFTResearch
WizardCoder-Python-34B-V1.0 πŸ€— HF Link πŸ“ƒ [WizardCoder] 73.2 64.6 73.2 59.9 Llama2
WizardCoder-15B-V1.0 πŸ€— HF Link πŸ“ƒ [WizardCoder] 59.8 52.4 -- -- OpenRAIL-M
WizardCoder-Python-13B-V1.0 πŸ€— HF Link πŸ“ƒ [WizardCoder] 64.0 -- -- -- Llama2
WizardCoder-Python-7B-V1.0 πŸ€— HF Link πŸ“ƒ [WizardCoder] 55.5 -- -- -- Llama2
WizardCoder-3B-V1.0 πŸ€— HF Link πŸ“ƒ [WizardCoder] 34.8 -- -- -- OpenRAIL-M
WizardCoder-1B-V1.0 πŸ€— HF Link πŸ“ƒ [WizardCoder] 23.8 -- -- -- OpenRAIL-M


  • πŸ”₯ [08/11/2023] We release WizardMath Models.
  • πŸ”₯ Our WizardMath-70B-V1.0 model slightly outperforms some closed-source LLMs on the GSM8K, including ChatGPT 3.5, Claude Instant 1 and PaLM 2 540B.
  • πŸ”₯ Our WizardMath-70B-V1.0 model achieves 81.6 pass@1 on the GSM8k Benchmarks, which is 24.8 points higher than the SOTA open-source LLM.
  • πŸ”₯ Our WizardMath-70B-V1.0 model achieves 22.7 pass@1 on the MATH Benchmarks, which is 9.2 points higher than the SOTA open-source LLM.
Model Checkpoint Paper GSM8k MATH Online Demo License
WizardMath-70B-V1.0 πŸ€— HF Link πŸ“ƒ [WizardMath] 81.6 22.7 Demo Llama 2
WizardMath-13B-V1.0 πŸ€— HF Link πŸ“ƒ [WizardMath] 63.9 14.0 Demo Llama 2
WizardMath-7B-V1.0 πŸ€— HF Link πŸ“ƒ [WizardMath] 54.9 10.7 Demo Llama 2
Model Checkpoint Paper MT-Bench AlpacaEval WizardEval HumanEval License
WizardLM-13B-V1.2 πŸ€— HF Link 7.06 89.17% 101.4% 36.6 pass@1 Llama 2 License
WizardLM-13B-V1.1 πŸ€— HF Link 6.76 86.32% 99.3% 25.0 pass@1 Non-commercial
WizardLM-30B-V1.0 πŸ€— HF Link 7.01 97.8% 37.8 pass@1 Non-commercial
WizardLM-13B-V1.0 πŸ€— HF Link 6.35 75.31% 89.1% 24.0 pass@1 Non-commercial
WizardLM-7B-V1.0 πŸ€— HF Link πŸ“ƒ [WizardLM] 78.0% 19.1 pass@1 Non-commercial

WizardCoder: Empowering Code Large Language Models with Evol-Instruct

To develop our WizardCoder model, we begin by adapting the Evol-Instruct method specifically for coding tasks. This involves tailoring the prompt to the domain of code-related instructions. Subsequently, we fine-tune the Code LLM, StarCoder, utilizing the newly created instruction-following training set.


Comparing WizardCoder with the Closed-Source Models.

πŸ”₯ The following figure shows that our WizardCoder attains the third position in this benchmark, surpassing Claude-Plus (59.8 vs. 53.0) and Bard (59.8 vs. 44.5). Notably, our model exhibits a substantially smaller size compared to these models.


❗Note: In this study, we copy the scores for HumanEval and HumanEval+ from the LLM-Humaneval-Benchmarks. Notably, all the mentioned models generate code solutions for each problem utilizing a single attempt, and the resulting pass rate percentage is reported. Our WizardCoder generates answers using greedy decoding and tests with the same code.

Comparing WizardCoder with the Open-Source Models.

The following table clearly demonstrates that our WizardCoder exhibits a substantial performance advantage over all the open-source models. ❗If you are confused with the different scores of our model (57.3 and 59.8), please check the Notes.

Model HumanEval Pass@1 MBPP Pass@1
CodeGen-16B-Multi 18.3 20.9
CodeGeeX 22.9 24.4
LLaMA-33B 21.7 30.2
LLaMA-65B 23.7 37.7
PaLM-540B 26.2 36.8
PaLM-Coder-540B 36.0 47.0
PaLM 2-S 37.6 50.0
CodeGen-16B-Mono 29.3 35.3
Code-Cushman-001 33.5 45.9
StarCoder-15B 33.6 43.6*
InstructCodeT5+ 35.0 --
WizardLM-30B 1.0 37.8 --
WizardCoder-15B 1.0 57.3 51.8

❗Note: The reproduced result of StarCoder on MBPP.

❗Note: The above table conducts a comprehensive comparison of our WizardCoder with other models on the HumanEval and MBPP benchmarks. We adhere to the approach outlined in previous studies by generating 20 samples for each problem to estimate the pass@1 score and evaluate with the same code. The scores of GPT4 and GPT3.5 reported by OpenAI are 67.0 and 48.1 (maybe these are the early version GPT4&3.5).

Call for Feedbacks

We welcome everyone to use your professional and difficult instructions to evaluate WizardCoder, and show us examples of poor performance and your suggestions in the issue discussion area. We are focusing on improving the Evol-Instruct now and hope to relieve existing weaknesses and issues in the the next version of WizardCoder. After that, we will open the code and pipeline of up-to-date Evol-Instruct algorithm and work with you together to improve it.


  1. Online Demo

  2. Fine-tuning

  3. Inference

  4. Evaluation

  5. Citation

  6. Disclaimer

Online Demo

We will provide our latest models for you to try for as long as possible. If you find a link is not working, please try another one. At the same time, please try as many real-world and challenging code-related problems that you encounter in your work and life as possible. We will continue to evolve our models with your feedbacks.


We fine-tune WizardCoder using the modified code train.py from Llama-X. We fine-tune StarCoder-15B with the following hyperparameters:

Hyperparameter StarCoder-15B
Batch size 512
Learning rate 2e-5
Epochs 3
Max length 2048
Warmup step 30
LR scheduler cosine

To reproduce our fine-tuning of WizardCoder, please follow the following steps:

  1. According to the instructions of Llama-X, install the environment, download the training code, and deploy. (Note: deepspeed==0.9.2 and transformers==4.29.2)
  2. Replace the train.py with the train_wizardcoder.py in our repo (src/train_wizardcoder.py)
  3. Login Huggingface:
huggingface-cli login
  1. Execute the following training command:
deepspeed train_wizardcoder.py \
    --model_name_or_path "bigcode/starcoder" \
    --data_path "/your/path/to/code_instruction_data.json" \
    --output_dir "/your/path/to/ckpt" \
    --num_train_epochs 3 \
    --model_max_length 2048 \
    --per_device_train_batch_size 16 \
    --per_device_eval_batch_size 1 \
    --gradient_accumulation_steps 4 \
    --evaluation_strategy "no" \
    --save_strategy "steps" \
    --save_steps 50 \
    --save_total_limit 2 \
    --learning_rate 2e-5 \
    --warmup_steps 30 \
    --logging_steps 2 \
    --lr_scheduler_type "cosine" \
    --report_to "tensorboard" \
    --gradient_checkpointing True \
    --deepspeed configs/deepspeed_config.json \
    --fp16 True


We provide the decoding script for WizardCoder, which reads a input file and generates corresponding responses for each sample, and finally consolidates them into an output file.

You can specify base_model, input_data_path and output_data_path in src\inference_wizardcoder.py to set the decoding model, path of input file and path of output file.

pip install jsonlines

The decoding command is:

python src\inference_wizardcoder.py \
    --base_model "/your/path/to/ckpt" \
    --input_data_path "/your/path/to/input/data.jsonl" \
    --output_data_path "/your/path/to/output/result.jsonl"

The format of data.jsonl should be:

{"idx": 11, "Instruction": "Write a Python code to count 1 to 10."}
{"idx": 12, "Instruction": "Write a Jave code to sum 1 to 10."}

The prompt for our WizardCoder in src\inference_wizardcoder.py is:

Below is an instruction that describes a task. Write a response that appropriately completes the request.

### Instruction:

### Response:


We provide the evaluation script on HumanEval for WizardCoder.

  1. According to the instructions of HumanEval, install the environment.
  2. Run the following script to generate the answer.


mkdir -p ${output_path}
echo 'Output path: '$output_path
echo 'Model to eval: '$model

# 164 problems, 21 per GPU if GPU=8
for ((i = 0; i < $gpu_num; i++)); do
  start_index=$((i * 21))
  end_index=$(((i + 1) * 21))

  echo 'Running process #' ${i} 'from' $start_index 'to' $end_index 'on GPU' ${gpu}
    CUDA_VISIBLE_DEVICES=$gpu python humaneval_gen.py --model ${model} \
      --start_index ${start_index} --end_index ${end_index} --temperature ${temp} \
      --num_seqs_per_iter ${num_seqs_per_iter} --N ${pred_num} --max_len ${max_len} --output_path ${output_path}
  ) &
  if (($index % $gpu_num == 0)); then wait; fi
  1. Run the post processing code src/process_humaneval.py to collect the code completions from all answer files.

echo 'Output path: '$output_path
python process_humaneval.py --path ${output_path} --out_path ${output_path}.jsonl --add_prompt

evaluate_functional_correctness ${output_path}.jsonl


Please cite the repo if you use the data, method or code in this repo.

  title={WizardCoder: Empowering Code Large Language Models with Evol-Instruct},
  author={Luo, Ziyang and Xu, Can and Zhao, Pu and Sun, Qingfeng and Geng, Xiubo and Hu, Wenxiang and Tao, Chongyang and Ma, Jing and Lin, Qingwei and Jiang, Daxin},
  journal={arXiv preprint arXiv:2306.08568},


WizardCoder model follows the same license as StarCoder. The content produced by any version of WizardCoder is influenced by uncontrollable variables such as randomness, and therefore, the accuracy of the output cannot be guaranteed by this project. This project does not accept any legal liability for the content of the model output, nor does it assume responsibility for any losses incurred due to the use of associated resources and output results.

Downloads last month
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social visibility and check back later, or deploy to Inference Endpoints (dedicated) instead.

Spaces using WizardLMTeam/WizardCoder-15B-V1.0 87

Evaluation results