Instructions to use ekryski/mamba2-130m-4bit with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- MLX
How to use ekryski/mamba2-130m-4bit with MLX:
# Download the model from the Hub pip install huggingface_hub[hf_xet] huggingface-cli download --local-dir mamba2-130m-4bit ekryski/mamba2-130m-4bit
- Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- LM Studio
- Atomic Chat
Download config.json from ekryski/mamba2-130m-4bit: direct link, hf CLI and curl.
- Browser
- Download file 983 Bytes
-
https://huggingface.co/ekryski/mamba2-130m-4bit/resolve/main/config.json
- Command line
-
hf download hf://ekryski/mamba2-130m-4bit/config.json
-
curl -L -o config.json https://huggingface.co/ekryski/mamba2-130m-4bit/resolve/main/config.json
983 Bytes
| { | |
| "bos_token_id" : 0, | |
| "chunk_size" : 256, | |
| "conv_kernel" : 4, | |
| "eos_token_id" : 0, | |
| "expand" : 2, | |
| "head_dim" : 64, | |
| "hidden_act" : "silu", | |
| "hidden_size" : 768, | |
| "initializer_range" : 0.10000000000000001, | |
| "layer_norm_epsilon" : 1.0000000000000001e-05, | |
| "model_type" : "mamba2", | |
| "n_groups" : 1, | |
| "num_heads" : 24, | |
| "num_hidden_layers" : 24, | |
| "pad_token_id" : 0, | |
| "quantization" : { | |
| "bits" : 4, | |
| "group_size" : 64, | |
| "mode" : "affine" | |
| }, | |
| "quantization_config" : { | |
| "bits" : 4, | |
| "group_size" : 64, | |
| "mode" : "affine" | |
| }, | |
| "rescale_prenorm_residual" : 0, | |
| "residual_in_fp32" : 1, | |
| "rms_norm" : 1, | |
| "state_size" : 128, | |
| "tie_word_embeddings" : 1, | |
| "time_step_floor" : 0.0001, | |
| "time_step_limit" : [ | |
| 0, | |
| 1e+308 | |
| ], | |
| "time_step_max" : 0.10000000000000001, | |
| "time_step_min" : 0.001, | |
| "time_step_rank" : 256, | |
| "transformers_version" : "4.45.0", | |
| "use_bias" : 0, | |
| "use_cache" : 1, | |
| "use_conv_bias" : 1, | |
| "vocab_size" : 50288 | |
| } |