|
|
--- |
|
|
license: apache-2.0 |
|
|
language: |
|
|
- en |
|
|
- zh |
|
|
base_model: YOYO-AI/QwQ-Coder-instruct |
|
|
pipeline_tag: text-generation |
|
|
tags: |
|
|
- merge |
|
|
- mlx |
|
|
- mlx-my-repo |
|
|
new_version: YOYO-AI/QwQ-coder-32B |
|
|
--- |
|
|
|
|
|
# bobig/QwQ-Coder-instruct-mlx-4Bit |
|
|
|
|
|
This is pretty good. QwQ brains and memory + Qwen code instruct |
|
|
|
|
|
Now in delicious MLX, eat it or wear it |
|
|
|
|
|
32k context is solid in QwQ: https://fiction.live/stories/Fiction-liveBench-Mar-14-2025/oQdzQvKHw8JyXbN87 |
|
|
|
|
|
Test Prompt: Write a quick sort in C++ |
|
|
|
|
|
The Model [bobig/QwQ-Coder-instruct-mlx-4Bit](https://huggingface.co/bobig/QwQ-Coder-instruct-mlx-4Bit) was converted to MLX format from [YOYO-AI/QwQ-Coder-instruct](https://huggingface.co/YOYO-AI/QwQ-Coder-instruct) using mlx-lm version **0.21.5**. |
|
|
|
|
|
## Use with mlx |
|
|
|
|
|
```bash |
|
|
pip install mlx-lm |
|
|
``` |
|
|
|
|
|
```python |
|
|
from mlx_lm import load, generate |
|
|
|
|
|
model, tokenizer = load("bobig/QwQ-Coder-instruct-mlx-4Bit") |
|
|
|
|
|
prompt="hello" |
|
|
|
|
|
if hasattr(tokenizer, "apply_chat_template") and tokenizer.chat_template is not None: |
|
|
messages = [{"role": "user", "content": prompt}] |
|
|
prompt = tokenizer.apply_chat_template( |
|
|
messages, tokenize=False, add_generation_prompt=True |
|
|
) |
|
|
|
|
|
response = generate(model, tokenizer, prompt=prompt, verbose=True) |
|
|
``` |
|
|
|
