37 lines
1.1 KiB
Markdown
37 lines
1.1 KiB
Markdown
---
|
|
license: apache-2.0
|
|
language:
|
|
- en
|
|
library_name: mlx
|
|
pipeline_tag: text-generation
|
|
base_model: Qwen/Qwen3-1.7B
|
|
base_model_relation: finetune
|
|
datasets:
|
|
- jokernifty/docs-instruct-nextjs-20260601-0306
|
|
tags:
|
|
- mlx
|
|
- qwen2
|
|
- instruction-tuning
|
|
- downftuner
|
|
- synthetic-data
|
|
---
|
|
|
|
# qwen3-1.7b-docs-20260601-0306
|
|
|
|
LoRA fine-tune of [`Qwen/Qwen3-1.7B`](https://huggingface.co/Qwen/Qwen3-1.7B)
|
|
on a freshly-synthesized mixed-Wikipedia instruction dataset.
|
|
|
|
- **Training**: MLX-LM LoRA, 800 iterations, 16 layers, LR 1.0e-5.
|
|
- **Dataset**: [jokernifty/docs-instruct-nextjs-20260601-0306](https://huggingface.co/datasets/jokernifty/docs-instruct-nextjs-20260601-0306)
|
|
- **Pipeline**: [DownFTuner](https://github.com/jokernifty/downftuner)
|
|
|
|
## Usage
|
|
|
|
```python
|
|
from mlx_lm import load, generate
|
|
model, tokenizer = load("jokernifty/qwen3-1.7b-docs-nextjs-20260601-0306")
|
|
print(generate(model, tokenizer, prompt="Explain the Krebs cycle.", max_tokens=200))
|
|
```
|
|
|
|
For `transformers`, convert from MLX format with `mlx_lm.convert --hf-path jokernifty/qwen3-1.7b-docs-nextjs-20260601-0306` or load via `mlx_lm`.
|