README.md

---
library_name: transformers
tags:
- code

---


# Bud Code Millenials 3B

Welcome to our Code Model repository! Our model is specifically fine-tuned for code generation tasks. Bud Millenial Code Gen open-source models are currently the State of the Art (SOTA) for code generation, beating all the existing models of all sizes. We have achieved a HumanEval value of 80.48 @ Pass 1, beating proprietary models like Gemini Ultra, Claude, GPT-3.5 etc. by a large margin, and on par with GPT-4 (HumanEval ~ 82. Ref. WizardCoder). Our proprietary model (Bud Code Jr) beats GPT-4 as well with a HumanEval value of 88.2 & a context size of 168K, we will be releasing an API for Researchers, Enterprises, and potential Partners by January 2024 end. If interested, please reach out to jithinvg@bud.studio

### News 🔥🔥🔥

- [2024/01/09] We released **Code Millenials 3B** , which achieves the **56.09 pass@1** on the [HumanEval Benchmarks](https://github.com/openai/human-eval).
- [2024/01/09] We released **Code Millenials 1B** , which achieves the **51.82 pass@1** on the [HumanEval Benchmarks](https://github.com/openai/human-eval).
- [2024/01/03] We released **Code Millenials 34B** , which achieves the **80.48 pass@1** on the [HumanEval Benchmarks](https://github.com/openai/human-eval).
- [2024/01/02] We released **Code Millenials 13B** , which achieves the **76.21 pass@1** on the [HumanEval Benchmarks](https://github.com/openai/human-eval).


### HumanEval

<p align="center" width="100%">
<a ><img src="https://raw.githubusercontent.com/BudEcosystem/code-millenials/main/assets/result.png" alt="CodeMillenials" style="width: 100%; min-width: 300px; display: block; margin: auto;"></a>
</p>

<p align="center" width="100%">
<a ><img src="https://raw.githubusercontent.com/BudEcosystem/code-millenials/main/assets/result-3b.png" alt="CodeMillenials" style="width: 100%; min-width: 300px; display: block; margin: auto;"></a>
</p>

For the millenial models, the eval script in the github repo is used for the above result.

Note: The humaneval values of other models are taken from the official repos of [WizardCoder](https://github.com/nlpxucan/WizardLM), [DeepseekCoder](https://github.com/deepseek-ai/deepseek-coder), [Gemini](https://deepmind.google/technologies/gemini/#capabilities) etc. 


### Models

|   Model | Checkpoint  | HumanEval (+) | MBPP (+) |
|---------|-------------|---------------|----------|
|Code Millenials 34B | <a href="https://huggingface.co/budecosystem/code-millenials-34b" target="_blank">HF Link</a> | 80.48 (75) | 74.68 (62.9) |
|Code Millenials 13B | <a href="https://huggingface.co/budecosystem/code-millenials-13b" target="_blank">HF Link</a> | 76.21 (69.5) | 70.17 (57.6) |
|Code Millenials 3B | <a href="https://huggingface.co/budecosystem/code-millenials-3b" target="_blank">HF Link</a> | 56.09 (52.43) | 55.13 (47.11) |
|Code Millenials 1B | <a href="https://huggingface.co/budecosystem/code-millenials-1b" target="_blank">HF Link</a> | 51.82 (48.17) | 53.13 (44.61) |




### 🚀 Quick Start

Inference code  using the pre-trained model from the Hugging Face model hub

```python
import torch
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("budecosystem/code-millenials-3b")
model = AutoModelForCausalLM.from_pretrained("budecosystem/code-millenials-3b")

template = """A chat between a curious user and an artificial intelligence assistant. The assistant gives helpful, detailed, and polite answers to the user's questions.
### Instruction: {instruction} ### Response:"""

instruction = <Your code instruction here>

prompt = template.format(instruction=instruction)

inputs = tokenizer(prompt, return_tensors="pt")
sample = model.generate(**inputs, max_length=128)
print(tokenizer.decode(sample[0]))

```


## Training details

The model is trained of 8 A100 80GB for approximately 6hrs. 

| Hyperparameters              | Value  |
| :----------------------------| :-----: |
| per_device_train_batch_size  | 3      |
| gradient_accumulation_steps  | 1      |
| epoch | 3 |
| steps | 26289 |
| learning_rate                | 2e-5   |
| lr schedular type | cosine |
| warmup ratio | 0.15 |
| optimizer                    | adamw  |
| fp16                         | True   |
| GPU                          | 8 A100 80GB |

### Important Note

- **Bias, Risks, and Limitations:** Model may sometimes make errors, produce misleading contents, or struggle to manage tasks that are not related to coding.