Upload merv-base/config.json with huggingface_hub
Browse files- merv-base/config.json +71 -0
merv-base/config.json
ADDED
|
@@ -0,0 +1,71 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"dataset": {
|
| 3 |
+
"align_stage_components": [
|
| 4 |
+
"download/videollava/valley_llavaimage.json",
|
| 5 |
+
"download/videollava"
|
| 6 |
+
],
|
| 7 |
+
"dataset_root_dir": "data",
|
| 8 |
+
"finetune_stage_components": [
|
| 9 |
+
"download/videollava/videochatgpt_llavaimage_tune.json",
|
| 10 |
+
"download/videollava"
|
| 11 |
+
],
|
| 12 |
+
"type": "videollava"
|
| 13 |
+
},
|
| 14 |
+
"hf_token": ".hf_token",
|
| 15 |
+
"model": {
|
| 16 |
+
"feature_fusion": "cross_attention_avg_lq",
|
| 17 |
+
"align_epochs": 1,
|
| 18 |
+
"align_global_batch_size": 256,
|
| 19 |
+
"align_learning_rate": 0.001,
|
| 20 |
+
"align_lr_scheduler_type": "linear-warmup+cosine-decay",
|
| 21 |
+
"align_max_grad_norm": 1.0,
|
| 22 |
+
"align_max_steps": null,
|
| 23 |
+
"align_per_device_batch_size": 16,
|
| 24 |
+
"align_train_strategy": "fsdp-shard-grad-op",
|
| 25 |
+
"align_warmup_ratio": 0.03,
|
| 26 |
+
"align_weight_decay": 0.0,
|
| 27 |
+
"arch_specifier": "no-align+3davg+linear",
|
| 28 |
+
"enable_gradient_checkpointing": true,
|
| 29 |
+
"enable_mixed_precision_training": true,
|
| 30 |
+
"finetune_epochs": 1,
|
| 31 |
+
"finetune_global_batch_size": 128,
|
| 32 |
+
"finetune_learning_rate": 2e-05,
|
| 33 |
+
"finetune_lr_scheduler_type": "linear-warmup+cosine-decay",
|
| 34 |
+
"finetune_max_grad_norm": 1.0,
|
| 35 |
+
"finetune_max_steps": null,
|
| 36 |
+
"finetune_per_device_batch_size": 8,
|
| 37 |
+
"finetune_train_strategy": "fsdp-full-shard",
|
| 38 |
+
"finetune_warmup_ratio": 0.03,
|
| 39 |
+
"finetune_weight_decay": 0.1,
|
| 40 |
+
"image_resize_strategy": "resize-naive",
|
| 41 |
+
"llm_backbone_id": "llama2-7b-pure",
|
| 42 |
+
"llm_max_length": 2048,
|
| 43 |
+
"model_id": "merv-base",
|
| 44 |
+
"num_frames": [
|
| 45 |
+
16,
|
| 46 |
+
16,
|
| 47 |
+
32,
|
| 48 |
+
16
|
| 49 |
+
],
|
| 50 |
+
"projector_token_length": 64,
|
| 51 |
+
"reduce_in_full_precision": false,
|
| 52 |
+
"type": "merv-base",
|
| 53 |
+
"video_backbone_ids": [
|
| 54 |
+
"languagebind-video-noclass",
|
| 55 |
+
"dinov2-video-all-tokens",
|
| 56 |
+
"vivit-google-b-all-no-cls-16frames",
|
| 57 |
+
"siglip-vit-b16-224px-all-no-cls"
|
| 58 |
+
],
|
| 59 |
+
"visual_feature_length": 1024
|
| 60 |
+
},
|
| 61 |
+
"pretrained_checkpoint": null,
|
| 62 |
+
"run_id": "merv-base",
|
| 63 |
+
"run_root_dir": "runs",
|
| 64 |
+
"seed": 7,
|
| 65 |
+
"stage": "finetune",
|
| 66 |
+
"trackers": [
|
| 67 |
+
"jsonl",
|
| 68 |
+
"wandb"
|
| 69 |
+
],
|
| 70 |
+
"type": "merv-base"
|
| 71 |
+
}
|