Upload config.json with huggingface_hub
Browse files- config.json +42 -0
config.json
ADDED
|
@@ -0,0 +1,42 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"model_config": {
|
| 3 |
+
"n_layer": 2,
|
| 4 |
+
"d_model": 3072,
|
| 5 |
+
"n_ctx": 512,
|
| 6 |
+
"d_head": 16,
|
| 7 |
+
"d_mlp": 12288,
|
| 8 |
+
"vocab_size": 4096,
|
| 9 |
+
"use_rms_norm": true,
|
| 10 |
+
"tie_embeddings": false,
|
| 11 |
+
"use_positional_embeddings": false,
|
| 12 |
+
"use_bigram_table": false,
|
| 13 |
+
"use_attention_sinks": true,
|
| 14 |
+
"activation": "gelu",
|
| 15 |
+
"dropout": 0,
|
| 16 |
+
"use_bias": true,
|
| 17 |
+
"use_flash_attention": true,
|
| 18 |
+
"gradient_checkpointing": false
|
| 19 |
+
},
|
| 20 |
+
"sparsity_config": {
|
| 21 |
+
"enable_weight_sparsity": true,
|
| 22 |
+
"target_l0_fraction": 0.01,
|
| 23 |
+
"sparsity_anneal_start_fraction": 0.5,
|
| 24 |
+
"sparsity_anneal_end_fraction": 0.75,
|
| 25 |
+
"anneal_type": "linear",
|
| 26 |
+
"min_weights_per_neuron": 4,
|
| 27 |
+
"enable_activation_sparsity": true,
|
| 28 |
+
"activation_topk_fraction": 0.2,
|
| 29 |
+
"activation_sparsity_locations": "attn_q,attn_k,attn_v,attn_out,mlp_neuron,mlp_out",
|
| 30 |
+
"activation_topk_mode": "per_token",
|
| 31 |
+
"mlp_neuron_pre_activation": true,
|
| 32 |
+
"activation_topk_annealing": "linear",
|
| 33 |
+
"activation_topk_anneal_start_fraction": 0.01,
|
| 34 |
+
"activation_topk_anneal_end_fraction": 0.33
|
| 35 |
+
},
|
| 36 |
+
"training_config": {
|
| 37 |
+
"total_tokens": 1000000000,
|
| 38 |
+
"batch_size": 32,
|
| 39 |
+
"dataset_name": "jacobcd52/simplestories-tokenized",
|
| 40 |
+
"tokenizer_name": "SimpleStories/SimpleStories-1.25M"
|
| 41 |
+
}
|
| 42 |
+
}
|