| { |
| "model_config": { |
| "n_layer": 2, |
| "d_model": 3072, |
| "n_ctx": 512, |
| "d_head": 16, |
| "d_mlp": 12288, |
| "vocab_size": 4096, |
| "use_rms_norm": true, |
| "tie_embeddings": false, |
| "use_positional_embeddings": false, |
| "use_bigram_table": false, |
| "use_attention_sinks": true, |
| "activation": "gelu", |
| "dropout": 0.0, |
| "use_bias": true, |
| "use_flash_attention": true |
| }, |
| "sparsity_config": { |
| "enable_weight_sparsity": true, |
| "target_l0_fraction": 0.05, |
| "sparsity_anneal_start_fraction": 0.01, |
| "sparsity_anneal_end_fraction": 0.5, |
| "anneal_type": "linear", |
| "min_weights_per_neuron": 4, |
| "enable_activation_sparsity": true, |
| "activation_topk_fraction": 0.2, |
| "activation_sparsity_locations": "attn_q,attn_k,attn_v,attn_out,mlp_neuron,mlp_out", |
| "activation_topk_mode": "per_token", |
| "mlp_neuron_pre_activation": true, |
| "activation_topk_annealing": "linear", |
| "activation_topk_anneal_start_fraction": 0.01, |
| "activation_topk_anneal_end_fraction": 0.5 |
| }, |
| "training_config": { |
| "total_tokens": 1000000000, |
| "batch_size": 32, |
| "dataset_name": "jacobcd52/simplestories-tokenized", |
| "tokenizer_name": "SimpleStories/SimpleStories-1.25M" |
| } |
| } |
