Upload R2R router config
Browse files- config.json +97 -0
config.json
ADDED
|
@@ -0,0 +1,97 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"model": {
|
| 3 |
+
"model_type": "HiddenStatesTokenLMHeadLogitsClassifier",
|
| 4 |
+
"init_args": {
|
| 5 |
+
"hidden_states_size": 1536,
|
| 6 |
+
"logits_size": 100,
|
| 7 |
+
"hidden_dims": [
|
| 8 |
+
1024,
|
| 9 |
+
1024,
|
| 10 |
+
1024,
|
| 11 |
+
1024,
|
| 12 |
+
1024,
|
| 13 |
+
1024
|
| 14 |
+
],
|
| 15 |
+
"expansion_factor": 4,
|
| 16 |
+
"dropout_rate": 0.1,
|
| 17 |
+
"use_position_embedding": false,
|
| 18 |
+
"freeze_lm_head": true,
|
| 19 |
+
"normalize_input": false,
|
| 20 |
+
"pretrained_model_name": "deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B"
|
| 21 |
+
},
|
| 22 |
+
"model_specific_args": {},
|
| 23 |
+
"input_type": [
|
| 24 |
+
"hidden_states",
|
| 25 |
+
"token",
|
| 26 |
+
"logits"
|
| 27 |
+
],
|
| 28 |
+
"output_type": "binary"
|
| 29 |
+
},
|
| 30 |
+
"data": {
|
| 31 |
+
"train": {
|
| 32 |
+
"path": [
|
| 33 |
+
"local:output_qwen_r1/query_dataset_train/LLM_response/SLM_prefill/LLM_continuation_verify/divergent_label_dataset"
|
| 34 |
+
],
|
| 35 |
+
"type": "divergent",
|
| 36 |
+
"input_prefix": "small_"
|
| 37 |
+
},
|
| 38 |
+
"test": {
|
| 39 |
+
"path": [
|
| 40 |
+
"local:output_qwen_r1/query_dataset_validation/LLM_response/SLM_prefill/LLM_continuation_verify/divergent_label_dataset"
|
| 41 |
+
],
|
| 42 |
+
"type": "divergent",
|
| 43 |
+
"input_prefix": "small_"
|
| 44 |
+
}
|
| 45 |
+
},
|
| 46 |
+
"training": {
|
| 47 |
+
"optimizer": {
|
| 48 |
+
"lr": 5e-05,
|
| 49 |
+
"weight_decay": 0.0005
|
| 50 |
+
},
|
| 51 |
+
"params": {
|
| 52 |
+
"num_epochs": 50,
|
| 53 |
+
"batch_size": 1024,
|
| 54 |
+
"patience": 10,
|
| 55 |
+
"device": "cuda"
|
| 56 |
+
},
|
| 57 |
+
"loss": {
|
| 58 |
+
"type": "BCEWithLogitsLoss",
|
| 59 |
+
"recall_factor": 1.0
|
| 60 |
+
},
|
| 61 |
+
"validation": {
|
| 62 |
+
"valid_freq": 2
|
| 63 |
+
},
|
| 64 |
+
"dtype": "float32"
|
| 65 |
+
},
|
| 66 |
+
"optimizing": {
|
| 67 |
+
"type": "threshold",
|
| 68 |
+
"min_recall": 0.95
|
| 69 |
+
},
|
| 70 |
+
"output": {
|
| 71 |
+
"output_dir": "resource/default_router.pt",
|
| 72 |
+
"checkpoint_dir": "output/checkpoint",
|
| 73 |
+
"model_name": null
|
| 74 |
+
},
|
| 75 |
+
"result": {
|
| 76 |
+
"model_path": "resource/default_router.pt/classifier_20250725_071911.pt",
|
| 77 |
+
"results": {
|
| 78 |
+
"threshold": 0.41585858585858587,
|
| 79 |
+
"best_epoch": 4,
|
| 80 |
+
"best_val_loss": 0.7956125675237121,
|
| 81 |
+
"final_metrics": {
|
| 82 |
+
"accuracy": 0.7007202534916438,
|
| 83 |
+
"precision": 0.2664509538449507,
|
| 84 |
+
"recall": 0.9510688780850916,
|
| 85 |
+
"f1": 0.4162777526774294,
|
| 86 |
+
"positive_rate": 0.40050413829284653
|
| 87 |
+
},
|
| 88 |
+
"pre_opt_metrics": {
|
| 89 |
+
"accuracy": 0.6180255418711208,
|
| 90 |
+
"precision": 0.6029766343289514,
|
| 91 |
+
"recall": 0.9264055921282188,
|
| 92 |
+
"f1": 0.7304922422944558,
|
| 93 |
+
"positive_rate": 0.8585155818412245
|
| 94 |
+
}
|
| 95 |
+
}
|
| 96 |
+
}
|
| 97 |
+
}
|