Instructions to use EleutherAI/sae-llama-3.1-8b-64x with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use EleutherAI/sae-llama-3.1-8b-64x with Transformers:
# pip install -U transformers accelerate # Load model directly from transformers import AutoModel model = AutoModel.from_pretrained("EleutherAI/sae-llama-3.1-8b-64x", device_map="auto") - Notebooks
- Google Colab
- Kaggle
Download config.json from EleutherAI/sae-llama-3.1-8b-64x: direct link, hf CLI and curl.
- Browser
- Download file 708 Bytes
-
https://huggingface.co/EleutherAI/sae-llama-3.1-8b-64x/resolve/main/config.json
- Command line
-
hf download hf://EleutherAI/sae-llama-3.1-8b-64x/config.json
-
curl -L -o config.json https://huggingface.co/EleutherAI/sae-llama-3.1-8b-64x/resolve/main/config.json
708 Bytes
| {"sae": {"expansion_factor": 64, "normalize_decoder": true, "num_latents": 0, "k": 32, "multi_topk": true}, "batch_size": 4, "grad_acc_steps": 2, "micro_acc_steps": 1, "lr": null, "lr_warmup_steps": 1000, "auxk_alpha": 0.0, "dead_feature_threshold": 10000000, "hookpoints": ["layers.29.mlp"], "layers": [], "layer_stride": 1, "distribute_modules": false, "save_every": 1000, "log_to_wandb": false, "run_name": "llama-64x", "wandb_log_frequency": 1, "model": "meta-llama/Meta-Llama-3.1-8B", "dataset": "/home/fslcollab366/sae/rpj-pretokenized.hf", "split": "train", "ctx_len": 2048, "hf_token": null, "load_in_8bit": false, "max_examples": null, "resume": false, "seed": 42, "data_preprocessing_num_proc": 64} |