| { |
| "model_type": "halo_s", |
| "architectures": [ |
| "HaloSForCausalLM" |
| ], |
| "auto_map": { |
| "AutoConfig": "configuration_halos.HaloSConfig", |
| "AutoModel": "modeling_halos.HaloSHFModel", |
| "AutoModelForCausalLM": "modeling_halos.HaloSForCausalLM" |
| }, |
| "vocab_size": 50257, |
| "hidden_size": 512, |
| "num_layers": 6, |
| "num_heads": 8, |
| "num_kv_heads": 2, |
| "num_globals": 2, |
| "local_window": 64, |
| "dilated_offsets": [1, 2, 4, 8, 16, 32, 64, 128], |
| "num_random": 2, |
| "dropout": 0.1, |
| "use_swiglu": false, |
| "max_seq_len": 1024, |
| "torch_dtype": "float16", |
| "framework": "pyhalos", |
| "architecture": "HALO-S", |
| "tokenizer": "tiktoken-gpt2", |
| "training": { |
| "dataset": "wikitext-103-raw-v1", |
| "max_tokens": 30000000, |
| "epochs": 2, |
| "batch_size": 4, |
| "grad_accumulation": 4, |
| "effective_batch": 32, |
| "precision": "fp16_mixed", |
| "gpus": "2x T4 (DataParallel)", |
| "gradient_checkpointing": true |
| } |
| } |
