|
|
--- |
|
|
license: mit |
|
|
library_name: transformers |
|
|
datasets: |
|
|
- AI-MO/NuminaMath-CoT |
|
|
- KbsdJames/Omni-MATH |
|
|
- RUC-AIBOX/STILL-3-Preview-RL-Data |
|
|
- hendrycks/competition_math |
|
|
language: |
|
|
- en |
|
|
base_model: agentica-org/DeepScaleR-1.5B-Preview |
|
|
tags: |
|
|
- mlx |
|
|
--- |
|
|
|
|
|
# bobig/DeepScaleR-1.5B-6.5bit |
|
|
|
|
|
This works well as a draft model for speculative decoding in [LMstudio 3.10 beta](https://lmstudio.ai/docs/advanced/speculative-decoding) |
|
|
|
|
|
Try it with: [mlx-community/FuseO1-DeepSeekR1-Qwen2.5-Coder-32B-4.5bit](https://huggingface.co/mlx-community/FuseO1-DeepSeekR1-Qwen2.5-Coder-32B-4.5bit) |
|
|
|
|
|
you should see 30% faster TPS for math/code prompts even with "thinking" slowing down the Specultive Decoding |
|
|
|
|
|
The Model [bobig/DeepScaleR-1.5B-6.5bit](https://huggingface.co/bobig/DeepScaleR-1.5B-6.5bit) was |
|
|
converted to MLX format from [agentica-org/DeepScaleR-1.5B-Preview](https://huggingface.co/agentica-org/DeepScaleR-1.5B-Preview) |
|
|
using mlx-lm version **0.21.4**. |
|
|
|
|
|
## Use with mlx |
|
|
|
|
|
```bash |
|
|
pip install mlx-lm |
|
|
``` |
|
|
|
|
|
```python |
|
|
from mlx_lm import load, generate |
|
|
|
|
|
model, tokenizer = load("bobig/DeepScaleR-1.5B-6.5bit") |
|
|
|
|
|
prompt = "hello" |
|
|
|
|
|
if tokenizer.chat_template is not None: |
|
|
messages = [{"role": "user", "content": prompt}] |
|
|
prompt = tokenizer.apply_chat_template( |
|
|
messages, add_generation_prompt=True |
|
|
) |
|
|
|
|
|
response = generate(model, tokenizer, prompt=prompt, verbose=True) |
|
|
``` |
|
|
|
