Upload arch_budget_hours_24.json with huggingface_hub
Browse files- arch_budget_hours_24.json +62 -0
arch_budget_hours_24.json
ADDED
|
@@ -0,0 +1,62 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"architectures": [
|
| 3 |
+
"ScriptableMaskedLM"
|
| 4 |
+
],
|
| 5 |
+
"num_transformer_layers": 16,
|
| 6 |
+
"hidden_size": 768,
|
| 7 |
+
"intermed_size": 3072,
|
| 8 |
+
"hidden_dropout_prob": 0.1,
|
| 9 |
+
"norm": "LayerNorm",
|
| 10 |
+
"norm_eps": 1e-12,
|
| 11 |
+
"norm_scheme": "pre",
|
| 12 |
+
"nonlin": "GELUglu",
|
| 13 |
+
"tie_weights": true,
|
| 14 |
+
"sparse_prediction": true,
|
| 15 |
+
"decoder_bias": false,
|
| 16 |
+
"loss": "cross-entropy",
|
| 17 |
+
"z_loss_factor": 0,
|
| 18 |
+
"gradient_checkpointing": false,
|
| 19 |
+
"layer_fusion": true,
|
| 20 |
+
"embedding": {
|
| 21 |
+
"vocab_size": null,
|
| 22 |
+
"pos_embedding": "scaled-sinusoidal",
|
| 23 |
+
"dropout_prob": 0.1,
|
| 24 |
+
"pad_token_id": 0,
|
| 25 |
+
"max_seq_length": 128,
|
| 26 |
+
"embedding_dim": 768,
|
| 27 |
+
"normalization": true
|
| 28 |
+
},
|
| 29 |
+
"attention": {
|
| 30 |
+
"type": "flash-attention-impl",
|
| 31 |
+
"causal_attention": false,
|
| 32 |
+
"num_attention_heads": 12,
|
| 33 |
+
"dropout_prob": 0.1,
|
| 34 |
+
"skip_output_projection": false,
|
| 35 |
+
"qkv_bias": false,
|
| 36 |
+
"rotary_embedding": false,
|
| 37 |
+
"seq_op_in_fp32": false,
|
| 38 |
+
"sequence_op": "torch-softmax",
|
| 39 |
+
"high_level_fusion": false,
|
| 40 |
+
"low_level_fusion": true
|
| 41 |
+
},
|
| 42 |
+
"init": {
|
| 43 |
+
"type": "normal",
|
| 44 |
+
"std": 0.02
|
| 45 |
+
},
|
| 46 |
+
"ffn_layer_frequency": 1,
|
| 47 |
+
"deepnorm_scaling": false,
|
| 48 |
+
"skip_head_transform": true,
|
| 49 |
+
"layer_drop_theta": null,
|
| 50 |
+
"use_bias": false,
|
| 51 |
+
"final_norm": true,
|
| 52 |
+
"recurrent_layers": null,
|
| 53 |
+
"layer_macro_type": "transformer",
|
| 54 |
+
"num_labels": null,
|
| 55 |
+
"classification_head": {
|
| 56 |
+
"pooler": "avg",
|
| 57 |
+
"include_ff_layer": true,
|
| 58 |
+
"head_dim": 1024,
|
| 59 |
+
"nonlin": "Tanh",
|
| 60 |
+
"classifier_dropout": 0.1
|
| 61 |
+
}
|
| 62 |
+
}
|