You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
<strong>Anni</strong> is a high-performance code assistant built upon the <strong>Qwen3 14B</strong> architecture. Fine-tuned on the <strong>OpenCodeReasoning-2</strong> dataset, Anni is engineered to excel in deep algorithmic reasoning, competitive programming logic, and the implementation of complex, high-efficiency data structures.
21
+
</p>
4
22
5
23
---
6
24
@@ -16,25 +34,93 @@ A Qwen-based code assistant fine-tuned for reasoning and solving data structures
16
34
17
35
---
18
36
19
-
### License
37
+
## 💻 Usage
38
+
39
+
**Get started immediately** using the provided Google Colab notebooks:
40
+
41
+
***(Recommended) GGUF Inference :** Open the [Colab Notebook](https://colab.research.google.com/drive/16RKUtphbI1rAds_lLwPGk2cRhf9CDJDo?usp=sharing) to run standard inference.
42
+
43
+
***vLLM Serving:** Open the [Colab Notebook](https://colab.research.google.com/drive/1lXYtLT729qcxJPc56TllgwiGEsjIiW0Q?usp=sharing) to run inference using the vLLM server.
44
+
45
+
---
46
+
47
+
## 🛠️ Development Setup
48
+
49
+
### Prerequisites
50
+
51
+
1.**Python Dependencies:**
52
+
```bash
53
+
pip install -r requirements.txt
54
+
```
55
+
2. **System Tools:**
56
+
Ensure `tmux` is installed on your system (required for training scripts).
57
+
58
+
### Configuration
59
+
60
+
1. **Environment Variables:**
61
+
Rename the example environment file and add your API tokens (WandB, HuggingFace, ModelScope).
62
+
```bash
63
+
mv config/example.env config/.env
64
+
# Edit config/.env with your keys
65
+
```
66
+
67
+
2. **Training Config:**
68
+
Edit [config/config.yaml](config/config.yaml) to adjust hyperparameters.
69
+
**Note:* Specify the `LOCAL_STORAGE_PATH`in [src/train.py](src/train.py) before starting training.
70
+
71
+
### Running Training
72
+
73
+
To start the training process, run the shell script:
74
+
75
+
```bash
76
+
./scripts/train.sh
77
+
```
78
+
79
+
---
80
+
81
+
## 📂 Project Structure
82
+
83
+
### Source Code (`src/`)
84
+
| File | Description |
85
+
|------|-------------|
86
+
| [`preprocess.py`](src/preprocess.py) | Downloads the [OpenCodeReasoning-2 dataset](https://huggingface.co/datasets/nvidia/OpenCodeReasoning-2) and preprocesses it for training. |
87
+
| [`train.py`](src/train.py) | Downloads the base model and fine-tunes it on the preprocessed dataset. |
88
+
| [`save.py`](src/save.py) | Loads the fine-tuned LoRA adapters and saves the model as merged 16-bit and GGUF formats. |
89
+
| [`upload.py`](src/upload.py) | Uploads the merged model to Hugging Face and ModelScope. |
90
+
91
+
### Scripts (`scripts/`)
92
+
| File | Description |
93
+
|------|-------------|
94
+
| [`train.sh`](scripts/train.sh) | Runs the training script with specified parameters. |
95
+
| [`eval.sh`](scripts/eval.sh) | Evaluates the model on the LiveCodeBench dataset. |
96
+
| [`serve.sh`](scripts/serve.sh) | Serves the model using the vLLM server. |
97
+
| [`terminate_train.sh`](scripts/terminate_train.sh) | Terminates the training process. |
98
+
99
+
### Frontend (`web/`)
100
+
The frontend code forAnni is availablein the `web` directory.
This repository’s **model and its training code** are released under the **MIT License**.
22
108
All other elements, such as **frontend code, project name and logo**, are **trademarks** of the developer and owner of this repository (**Hans**) and **may not be used without explicit permission**.
23
109
24
110
---
25
111
26
-
###Training Dataset Notice
112
+
## Training Dataset Notice
27
113
28
114
The training dataset includes openly licensed sources under **CC-BY-4.0**, which **permits commercial use with attribution**.
0 commit comments