Instructions to use poolside/Laguna-S-2.1-NVFP4-mlx with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- MLX
How to use poolside/Laguna-S-2.1-NVFP4-mlx with MLX:
# Download the model from the Hub pip install huggingface_hub[hf_xet] huggingface-cli download --local-dir Laguna-S-2.1-NVFP4-mlx poolside/Laguna-S-2.1-NVFP4-mlx
- Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- LM Studio
Rebuild MLX from norot NVFP4 (15-shard nvfp4-pack source)
Browse files- generation_config.json +24 -0
generation_config.json
ADDED
|
@@ -0,0 +1,24 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"bos_token_id": 2,
|
| 3 |
+
"do_sample": true,
|
| 4 |
+
"eos_token_id": [
|
| 5 |
+
2,
|
| 6 |
+
24
|
| 7 |
+
],
|
| 8 |
+
"pad_token_id": 9,
|
| 9 |
+
"temperature": 1.0,
|
| 10 |
+
"top_p": 1.0,
|
| 11 |
+
"speculative_config": {
|
| 12 |
+
"method": "dflash",
|
| 13 |
+
"source": "huggingface",
|
| 14 |
+
"model": "poolside/Laguna-S-2.1-DFlash-NVFP4",
|
| 15 |
+
"num_speculative_tokens": 15
|
| 16 |
+
},
|
| 17 |
+
"tool_call_parser": "poolside_v1",
|
| 18 |
+
"reasoning_parser": "poolside_v1",
|
| 19 |
+
"default_chat_template_kwargs": {
|
| 20 |
+
"enable_thinking": true
|
| 21 |
+
},
|
| 22 |
+
"top_k": 20,
|
| 23 |
+
"min_p": 0.0
|
| 24 |
+
}
|