Upload adapters/adapter_config.json with huggingface_hub
Browse files- adapters/adapter_config.json +50 -0
adapters/adapter_config.json
ADDED
|
@@ -0,0 +1,50 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"adapter_path": "adapters",
|
| 3 |
+
"batch_size": 8,
|
| 4 |
+
"clear_cache_threshold": 0,
|
| 5 |
+
"config": "lora_config.yaml",
|
| 6 |
+
"data": "data/sft",
|
| 7 |
+
"fine_tune_type": "lora",
|
| 8 |
+
"grad_accumulation_steps": 1,
|
| 9 |
+
"grad_checkpoint": true,
|
| 10 |
+
"iters": 2000,
|
| 11 |
+
"learning_rate": 0.0001,
|
| 12 |
+
"lora_parameters": {
|
| 13 |
+
"keys": [
|
| 14 |
+
"self_attn.q_proj",
|
| 15 |
+
"self_attn.k_proj",
|
| 16 |
+
"self_attn.v_proj",
|
| 17 |
+
"self_attn.o_proj",
|
| 18 |
+
"mlp.gate_proj",
|
| 19 |
+
"mlp.up_proj",
|
| 20 |
+
"mlp.down_proj"
|
| 21 |
+
],
|
| 22 |
+
"rank": 16,
|
| 23 |
+
"scale": 20.0,
|
| 24 |
+
"dropout": 0.0
|
| 25 |
+
},
|
| 26 |
+
"lr_schedule": null,
|
| 27 |
+
"mask_prompt": true,
|
| 28 |
+
"max_seq_length": 2048,
|
| 29 |
+
"model": "WeiboAI/VibeThinker-3B",
|
| 30 |
+
"num_layers": 36,
|
| 31 |
+
"optimizer": "adamw",
|
| 32 |
+
"optimizer_config": {
|
| 33 |
+
"adam": {},
|
| 34 |
+
"adamw": {},
|
| 35 |
+
"muon": {},
|
| 36 |
+
"sgd": {},
|
| 37 |
+
"adafactor": {}
|
| 38 |
+
},
|
| 39 |
+
"project_name": null,
|
| 40 |
+
"report_to": null,
|
| 41 |
+
"resume_adapter_file": null,
|
| 42 |
+
"save_every": 200,
|
| 43 |
+
"seed": 13,
|
| 44 |
+
"steps_per_eval": 200,
|
| 45 |
+
"steps_per_report": 10,
|
| 46 |
+
"test": false,
|
| 47 |
+
"test_batches": 500,
|
| 48 |
+
"train": true,
|
| 49 |
+
"val_batches": 25
|
| 50 |
+
}
|