Upload folder using huggingface_hub
Browse files
merv-full/checkpoints/latest-checkpoint.pt
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:f4fb83072ab1b80a96943f9f1ee269739ee05b28a0263f90bedc2e814d1b6d2c
|
3 |
+
size 27190883468
|
merv-full/config.json
ADDED
@@ -0,0 +1,73 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"dataset": {
|
3 |
+
"align_stage_components": [
|
4 |
+
"download/videollava/valley_llavaimage.json",
|
5 |
+
"download/videollava"
|
6 |
+
],
|
7 |
+
"dataset_id": "videollava",
|
8 |
+
"dataset_root_dir": "data",
|
9 |
+
"finetune_stage_components": [
|
10 |
+
"download/videollava/videochatgpt_llavaimage_tune.json",
|
11 |
+
"download/videollava"
|
12 |
+
],
|
13 |
+
"type": "videollava"
|
14 |
+
},
|
15 |
+
"dataset_class": "FinetuneVideoDataset",
|
16 |
+
"hf_token": ".hf_token",
|
17 |
+
"model": {
|
18 |
+
"align_epochs": 1,
|
19 |
+
"align_global_batch_size": 256,
|
20 |
+
"align_learning_rate": 0.001,
|
21 |
+
"align_lr_scheduler_type": "linear-warmup+cosine-decay",
|
22 |
+
"align_max_grad_norm": 1.0,
|
23 |
+
"align_max_steps": null,
|
24 |
+
"align_per_device_batch_size": 16,
|
25 |
+
"align_train_strategy": "fsdp-full-shard",
|
26 |
+
"align_warmup_ratio": 0.03,
|
27 |
+
"align_weight_decay": 0.0,
|
28 |
+
"arch_specifier": "3davg+linear",
|
29 |
+
"enable_gradient_checkpointing": true,
|
30 |
+
"enable_mixed_precision_training": true,
|
31 |
+
"feature_fusion": "cross_attention_avg_lq",
|
32 |
+
"finetune_epochs": 1,
|
33 |
+
"finetune_global_batch_size": 128,
|
34 |
+
"finetune_learning_rate": 2e-05,
|
35 |
+
"finetune_lr_scheduler_type": "linear-warmup+cosine-decay",
|
36 |
+
"finetune_max_grad_norm": 1.0,
|
37 |
+
"finetune_max_steps": null,
|
38 |
+
"finetune_per_device_batch_size": 8,
|
39 |
+
"finetune_train_strategy": "fsdp-full-shard",
|
40 |
+
"finetune_warmup_ratio": 0.03,
|
41 |
+
"finetune_weight_decay": 0.1,
|
42 |
+
"image_resize_strategy": "resize-naive",
|
43 |
+
"llm_backbone_id": "llama2-7b-pure",
|
44 |
+
"llm_max_length": 2048,
|
45 |
+
"model_id": "prism-siglip-controlled+7b",
|
46 |
+
"num_frames": [
|
47 |
+
16,
|
48 |
+
16,
|
49 |
+
32,
|
50 |
+
16
|
51 |
+
],
|
52 |
+
"projector_token_length": 64,
|
53 |
+
"reduce_in_full_precision": false,
|
54 |
+
"type": "merv-full",
|
55 |
+
"video_backbone_ids": [
|
56 |
+
"languagebind-video-noclass",
|
57 |
+
"dinov2-video-all-tokens",
|
58 |
+
"vivit-google-b-all-no-cls-16frames",
|
59 |
+
"siglip-vit-b16-224px-all-no-cls"
|
60 |
+
],
|
61 |
+
"visual_feature_length": 1024
|
62 |
+
},
|
63 |
+
"pretrained_checkpoint": "runs/final-4e-stage1-lr1e-4/checkpoints/latest-checkpoint.pt",
|
64 |
+
"run_id": "final-4e-stage2-(lr1e-4)",
|
65 |
+
"run_root_dir": "runs",
|
66 |
+
"seed": 7,
|
67 |
+
"slurm_id": "751580",
|
68 |
+
"stage": "second_finetune",
|
69 |
+
"trackers": [
|
70 |
+
"jsonl",
|
71 |
+
"wandb"
|
72 |
+
]
|
73 |
+
}
|
merv-full/config.yaml
ADDED
@@ -0,0 +1,64 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
dataset:
|
2 |
+
align_stage_components:
|
3 |
+
- download/videollava/valley_llavaimage.json
|
4 |
+
- download/videollava
|
5 |
+
dataset_id: videollava
|
6 |
+
dataset_root_dir: data
|
7 |
+
finetune_stage_components:
|
8 |
+
- download/videollava/videochatgpt_llavaimage_tune.json
|
9 |
+
- download/videollava
|
10 |
+
type: videollava
|
11 |
+
dataset_class: FinetuneVideoDataset
|
12 |
+
hf_token: .hf_token
|
13 |
+
model:
|
14 |
+
align_epochs: 1
|
15 |
+
align_global_batch_size: 256
|
16 |
+
align_learning_rate: 0.001
|
17 |
+
align_lr_scheduler_type: linear-warmup+cosine-decay
|
18 |
+
align_max_grad_norm: 1.0
|
19 |
+
align_max_steps: null
|
20 |
+
align_per_device_batch_size: 16
|
21 |
+
align_train_strategy: fsdp-full-shard
|
22 |
+
align_warmup_ratio: 0.03
|
23 |
+
align_weight_decay: 0.0
|
24 |
+
arch_specifier: 3davg+linear
|
25 |
+
enable_gradient_checkpointing: true
|
26 |
+
enable_mixed_precision_training: true
|
27 |
+
feature_fusion: cross_attention_avg_lq
|
28 |
+
finetune_epochs: 1
|
29 |
+
finetune_global_batch_size: 128
|
30 |
+
finetune_learning_rate: 2.0e-05
|
31 |
+
finetune_lr_scheduler_type: linear-warmup+cosine-decay
|
32 |
+
finetune_max_grad_norm: 1.0
|
33 |
+
finetune_max_steps: null
|
34 |
+
finetune_per_device_batch_size: 8
|
35 |
+
finetune_train_strategy: fsdp-full-shard
|
36 |
+
finetune_warmup_ratio: 0.03
|
37 |
+
finetune_weight_decay: 0.1
|
38 |
+
image_resize_strategy: resize-naive
|
39 |
+
llm_backbone_id: llama2-7b-pure
|
40 |
+
llm_max_length: 2048
|
41 |
+
model_id: prism-siglip-controlled+7b
|
42 |
+
num_frames:
|
43 |
+
- 16
|
44 |
+
- 16
|
45 |
+
- 32
|
46 |
+
- 16
|
47 |
+
projector_token_length: 64
|
48 |
+
reduce_in_full_precision: false
|
49 |
+
type: merv-full
|
50 |
+
video_backbone_ids:
|
51 |
+
- languagebind-video-noclass
|
52 |
+
- dinov2-video-all-tokens
|
53 |
+
- vivit-google-b-all-no-cls-16frames
|
54 |
+
- siglip-vit-b16-224px-all-no-cls
|
55 |
+
visual_feature_length: 1024
|
56 |
+
pretrained_checkpoint: runs/final-4e-stage1-lr1e-4/checkpoints/latest-checkpoint.pt
|
57 |
+
run_id: final-4e-stage2-(lr1e-4)
|
58 |
+
run_root_dir: runs
|
59 |
+
seed: 7
|
60 |
+
slurm_id: '751580'
|
61 |
+
stage: second_finetune
|
62 |
+
trackers:
|
63 |
+
- jsonl
|
64 |
+
- wandb
|
merv-full/final-4e-stage2-(lr1e-4).jsonl
ADDED
The diff for this file is too large to render.
See raw diff
|
|
merv-full/run-metrics.jsonl
ADDED
@@ -0,0 +1 @@
|
|
|
|
|
1 |
+
{"hparams": {"dataset": {"align_stage_components": ["download/videollava/valley_llavaimage.json", "download/videollava"], "dataset_id": "videoonlyllava", "dataset_root_dir": "data", "finetune_stage_components": ["download/videollava/videochatgpt_llavaimage_tune.json", "download/videollava"], "type": "videoonlyllava"}, "dataset_class": "FinetuneVideoOnlyDataset", "hf_token": ".hf_token", "model": {"adapter": "cross_attention_avg_lq", "align_epochs": 1, "align_global_batch_size": 256, "align_learning_rate": 0.001, "align_lr_scheduler_type": "linear-warmup+cosine-decay", "align_max_grad_norm": 1.0, "align_max_steps": null, "align_per_device_batch_size": 16, "align_train_strategy": "fsdp-full-shard", "align_warmup_ratio": 0.03, "align_weight_decay": 0.0, "arch_specifier": "3dagv+linear", "enable_gradient_checkpointing": true, "enable_mixed_precision_training": true, "finetune_epochs": 1, "finetune_global_batch_size": 128, "finetune_learning_rate": 2e-05, "finetune_lr_scheduler_type": "linear-warmup+cosine-decay", "finetune_max_grad_norm": 1.0, "finetune_max_steps": null, "finetune_per_device_batch_size": 8, "finetune_train_strategy": "fsdp-full-shard", "finetune_warmup_ratio": 0.03, "finetune_weight_decay": 0.1, "image_resize_strategy": "resize-naive", "llm_backbone_id": "llama2-7b-pure", "llm_max_length": 2048, "model_id": "prism-siglip-controlled+7b", "normalize_post_proj": false, "num_frames": [16, 16, 32, 16], "pre_proj_layernorm": false, "projector_token_length": 64, "reduce_in_full_precision": false, "text_embedding_loc": "openai", "type": "prism-siglip-controlled+7b", "video_backbone_id": null, "video_backbone_ids": ["languagebind-video-noclass", "dinov2-video-all-tokens", "vivit-google-b-all-no-cls-16frames", "siglip-vit-b16-224px-all-no-cls"], "vision_backbone_id": "siglip-vit-so400m-384px", "visual_feature_length": 1024}, "pretrained_checkpoint": "runs/final-4e-stage1-lr1e-4/checkpoints/latest-checkpoint.pt", "run_id": "final-4e-stage2-(lr1e-4)", "run_root_dir": "runs", "seed": 7, "slurm_id": "751580", "stage": "second_finetune", "trackers": ["jsonl", "wandb"], "wandb_entity": "tz-ego4d", "wandb_project": "vidlm"}, "run_id": "final-4e-stage2-(lr1e-4)"}
|