Upload log/hps.json with huggingface_hub
Browse files- log/hps.json +78 -0
log/hps.json
ADDED
|
@@ -0,0 +1,78 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"run_cfg": {
|
| 3 |
+
"checkpoint": "",
|
| 4 |
+
"output_dir": "./output/hdli/pretrain_mmctx",
|
| 5 |
+
"gradient_accumulation_steps": 1,
|
| 6 |
+
"clip_lr": 5e-07,
|
| 7 |
+
"optim": "adamw",
|
| 8 |
+
"learning_rate": 5e-05,
|
| 9 |
+
"betas": [
|
| 10 |
+
0.9,
|
| 11 |
+
0.98
|
| 12 |
+
],
|
| 13 |
+
"weight_decay": 0.01,
|
| 14 |
+
"grad_norm": 2.0,
|
| 15 |
+
"warmup_ratio": 0.1,
|
| 16 |
+
"resume": false,
|
| 17 |
+
"seed": 50,
|
| 18 |
+
"fp16": true,
|
| 19 |
+
"bf16": false,
|
| 20 |
+
"zero_shot": false,
|
| 21 |
+
"scheduler": "warmup_linear",
|
| 22 |
+
"new_lr": 0,
|
| 23 |
+
"new_params_name": [],
|
| 24 |
+
"valid_freq": 10,
|
| 25 |
+
"dataset_mix_type": "random",
|
| 26 |
+
"remove_before_ckpt": true,
|
| 27 |
+
"first_eval": true,
|
| 28 |
+
"pretrain_dir": "./output/vast/pretrain_vast",
|
| 29 |
+
"num_train_steps": 0,
|
| 30 |
+
"save_best": false,
|
| 31 |
+
"pin_mem": true,
|
| 32 |
+
"vision_resolution": 224,
|
| 33 |
+
"pretrain_concat_num": 1,
|
| 34 |
+
"use_ddp": false,
|
| 35 |
+
"mode": "training",
|
| 36 |
+
"log_steps": 100,
|
| 37 |
+
"default": "./config/mmctx/default_run_cfg.json"
|
| 38 |
+
},
|
| 39 |
+
"model_cfg": {
|
| 40 |
+
"model_type": "mmctx",
|
| 41 |
+
"itm_ratio": 0.1,
|
| 42 |
+
"frozen_vision": false,
|
| 43 |
+
"frozen_audio": false,
|
| 44 |
+
"checkpointing": true,
|
| 45 |
+
"pool_video": false,
|
| 46 |
+
"max_caption_len": 40,
|
| 47 |
+
"max_omni_caption_len": 70,
|
| 48 |
+
"max_subtitle_len": 70,
|
| 49 |
+
"contra_dim": 512,
|
| 50 |
+
"inherit_keys": [
|
| 51 |
+
"vision_encoder_type",
|
| 52 |
+
"audio_encoder_type"
|
| 53 |
+
],
|
| 54 |
+
"frame_embedding_type": "adaptive",
|
| 55 |
+
"vision_resolution": 224,
|
| 56 |
+
"vision_encoder_type": "evaclip01_giant",
|
| 57 |
+
"audio_encoder_type": "beats",
|
| 58 |
+
"audio_melbins": 224,
|
| 59 |
+
"audio_target_length": 224,
|
| 60 |
+
"beam_size": 3,
|
| 61 |
+
"captioner_mode": false,
|
| 62 |
+
"generate_nums": 1,
|
| 63 |
+
"ret_bidirection_evaluation": false,
|
| 64 |
+
"itm_rerank_num": 50,
|
| 65 |
+
"evaluation_type": "evaluation_mm",
|
| 66 |
+
"default": "./config/mmctx/default_model_cfg.json",
|
| 67 |
+
"max_vision_sample_num": 2,
|
| 68 |
+
"max_audio_sample_num": 2,
|
| 69 |
+
"max_depth_sample_num": 1
|
| 70 |
+
},
|
| 71 |
+
"data_cfg": {
|
| 72 |
+
"train": [
|
| 73 |
+
],
|
| 74 |
+
"val": [],
|
| 75 |
+
"concatenated_nums": 1
|
| 76 |
+
},
|
| 77 |
+
"local_rank": 0
|
| 78 |
+
}
|