MLX
English
nima_joint_schema
tool-routing
decision-model
frozen-backbone
joint-schema
custom-code
research
Instructions to use neimand/nima with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- MLX
How to use neimand/nima with MLX:
# Download the model from the Hub pip install huggingface_hub[hf_xet] hf download neimand/nima --local-dir nima
- Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- LM Studio
- Atomic Chat
Download architecture.json from neimand/nima: direct link, hf CLI and curl.
- Browser
- Download file 479 Bytes
-
https://huggingface.co/neimand/nima/resolve/main/architecture.json
- Command line
-
hf download hf://neimand/nima/architecture.json
-
curl -L -o architecture.json https://huggingface.co/neimand/nima/resolve/main/architecture.json
479 Bytes
| { | |
| "architecture_id": "joint-schema-evidence-v1", | |
| "model": "mlx-community/Qwen3-1.7B-4bit", | |
| "hidden_size": 2048, | |
| "head_dims": 256, | |
| "attention_heads": 4, | |
| "ffn_dims": 512, | |
| "joint_layers": 1, | |
| "max_tokens": 1536, | |
| "candidate_pooling": "mean of final-normalized states within each complete schema span", | |
| "evidence": "all final-normalized states in context and request spans only", | |
| "backbone": "frozen", | |
| "candidate_position_embeddings": false, | |
| "dropout": 0.0 | |
| } | |