Instructions to use SceneWorks/hemispheres-step1 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- MLX
How to use SceneWorks/hemispheres-step1 with MLX:
# Download the model from the Hub pip install huggingface_hub[hf_xet] hf download SceneWorks/hemispheres-step1 --local-dir hemispheres-step1
- Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- LM Studio
- Atomic Chat
Upload context-a, context-multi-bios, context-multi-bios-20k, context-multi-bios-20k-s1, dense-a, dense-a-k100, dense-a-to-b, latent-a, latent-a-all, latent-multi, latent-multi-nohop, lookup-a
790f3b4 verified Download latent-multi/config.json from SceneWorks/hemispheres-step1: direct link, hf CLI and curl.
- Browser
- Download file 1.22 kB
-
https://huggingface.co/SceneWorks/hemispheres-step1/resolve/main/latent-multi/config.json
- Command line
-
hf download hf://SceneWorks/hemispheres-step1/latent-multi/config.json
-
curl -L -o config.json https://huggingface.co/SceneWorks/hemispheres-step1/resolve/main/latent-multi/config.json
1.22 kB
| { | |
| "arm": "latent", | |
| "data": "data/world-a,data/pool/w3-s0,data/pool/w3-s1,data/pool/w3-s2,data/pool/w4-s0,data/pool/w4-s1,data/pool/w4-s2,data/pool/w5-s0,data/pool/w5-s1,data/pool/w5-s2,data/pool/w6-s0,data/pool/w6-s1,data/pool/w6-s2,data/pool/w7-s0,data/pool/w7-s1,data/pool/w7-s2", | |
| "out": "runs/latent-multi", | |
| "edits": 0, | |
| "mix": [ | |
| [ | |
| "bios", | |
| 0.5 | |
| ], | |
| [ | |
| "qa", | |
| 0.5 | |
| ] | |
| ], | |
| "init": null, | |
| "init_checkpoint": "final", | |
| "size": "small", | |
| "dtype": "float32", | |
| "steps": 10000, | |
| "batch_size": 64, | |
| "seq_len": 256, | |
| "lr": 0.001, | |
| "min_lr_frac": 0.1, | |
| "warmup": 500, | |
| "weight_decay": 0.1, | |
| "grad_clip": 1.0, | |
| "log_every": 100, | |
| "eval_every": 2000, | |
| "eval_sets": "test_id,test_ood,test_1hop_ood", | |
| "eval_n": 200, | |
| "save_every": 2000, | |
| "seed": 0, | |
| "n_reads": 3, | |
| "top_k": 4, | |
| "hop_weight": 0.5, | |
| "max_supervision": 1024, | |
| "supervise": "all", | |
| "model_type": "latent", | |
| "model": { | |
| "vocab_size": 576, | |
| "n_layer": 8, | |
| "n_head": 8, | |
| "d_model": 512, | |
| "mlp_ratio": 4, | |
| "rope_base": 10000.0, | |
| "n_reads": 3, | |
| "top_k": 4, | |
| "key_dim": 128, | |
| "retrieval_chunk": 2048, | |
| "retrieval_group": 256 | |
| }, | |
| "params": 29157915, | |
| "vocab": 544 | |
| } |