Upload blueprint.json with huggingface_hub
Browse files- blueprint.json +69 -0
blueprint.json
ADDED
|
@@ -0,0 +1,69 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"model_config": {
|
| 3 |
+
"tokenizer_id": "Qwen/Qwen3.5-2B-Base",
|
| 4 |
+
"hf_repo_id": "Efe2898/500m-dense-base",
|
| 5 |
+
"output_dir": "/kaggle/working/efe-500m-dense-base",
|
| 6 |
+
"vocab_size": null,
|
| 7 |
+
"hidden_size": 1024,
|
| 8 |
+
"intermediate_size": 4096,
|
| 9 |
+
"num_hidden_layers": 24,
|
| 10 |
+
"num_attention_heads": 16,
|
| 11 |
+
"num_key_value_heads": 4,
|
| 12 |
+
"max_position_embeddings": 65536,
|
| 13 |
+
"rope_theta": 1000000.0,
|
| 14 |
+
"rope_scaling": {
|
| 15 |
+
"type": "yarn",
|
| 16 |
+
"factor": 8.0,
|
| 17 |
+
"original_max_position_embeddings": 8192,
|
| 18 |
+
"rope_theta": 1000000.0,
|
| 19 |
+
"rope_type": "yarn"
|
| 20 |
+
},
|
| 21 |
+
"use_sliding_window": true,
|
| 22 |
+
"sliding_window": 4096,
|
| 23 |
+
"max_window_layers": 20,
|
| 24 |
+
"rms_norm_eps": 1e-06,
|
| 25 |
+
"hidden_act": "silu",
|
| 26 |
+
"attention_dropout": 0.0,
|
| 27 |
+
"tie_word_embeddings": true,
|
| 28 |
+
"initializer_range": 0.02,
|
| 29 |
+
"torch_dtype": "bfloat16",
|
| 30 |
+
"push_to_hub": false,
|
| 31 |
+
"hf_token": null
|
| 32 |
+
},
|
| 33 |
+
"training_ref": {
|
| 34 |
+
"mesh_shape": [
|
| 35 |
+
8
|
| 36 |
+
],
|
| 37 |
+
"mesh_axis_names": [
|
| 38 |
+
"data"
|
| 39 |
+
],
|
| 40 |
+
"attn_implementation": "eager",
|
| 41 |
+
"cpt_seq_len": 2048,
|
| 42 |
+
"cpt_global_batch": 256,
|
| 43 |
+
"cpt_grad_accum": 1,
|
| 44 |
+
"cpt_tokens_target": 10000000000,
|
| 45 |
+
"sft_seq_len": 4096,
|
| 46 |
+
"sft_global_batch": 128,
|
| 47 |
+
"sft_grad_accum": 1,
|
| 48 |
+
"optimizer": "Adafactor",
|
| 49 |
+
"lr": 0.0003,
|
| 50 |
+
"weight_decay": 0.01,
|
| 51 |
+
"clip_threshold": 1.0,
|
| 52 |
+
"relative_step": false,
|
| 53 |
+
"scale_parameter": false,
|
| 54 |
+
"warmup_steps": 500,
|
| 55 |
+
"lr_scheduler": "cosine",
|
| 56 |
+
"min_lr_ratio": 0.1,
|
| 57 |
+
"dtype": "bfloat16",
|
| 58 |
+
"gradient_checkpointing": true,
|
| 59 |
+
"log_every_n_steps": 10,
|
| 60 |
+
"save_every_n_steps": 500,
|
| 61 |
+
"eval_every_n_steps": 1000
|
| 62 |
+
},
|
| 63 |
+
"param_count": 618988544,
|
| 64 |
+
"param_count_M": 619.0,
|
| 65 |
+
"mlp_fraction": 0.8275,
|
| 66 |
+
"vocab_size": 248044,
|
| 67 |
+
"status": "untrained_blueprint",
|
| 68 |
+
"next_step": "CPT notebook"
|
| 69 |
+
}
|