skaramcheti
commited on
Commit
•
6c633ae
1
Parent(s):
af11448
Upload folder using huggingface_hub
Browse files
prism-dinosiglip-224px+7b/checkpoints/latest-checkpoint.pt
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:b0281cb5c9e37d08e1a9bde840522518b02bad916b96058ee9bb99e0956e0159
|
3 |
+
size 13620752058
|
prism-dinosiglip-224px+7b/config.json
ADDED
@@ -0,0 +1,59 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"dataset": {
|
3 |
+
"align_stage_components": [
|
4 |
+
"download/llava-laion-cc-sbu-558k/chat.json",
|
5 |
+
"download/llava-laion-cc-sbu-558k"
|
6 |
+
],
|
7 |
+
"dataset_id": "llava-lvis4v-lrv",
|
8 |
+
"dataset_root_dir": "/opt/ml/input/data/training/skaramcheti/datasets/prismatic-vlms",
|
9 |
+
"finetune_stage_components": [
|
10 |
+
"download/llava-v1.5-instruct/llava_v1_5_lvis4v_lrv_mix1231k.json",
|
11 |
+
"download/llava-v1.5-instruct"
|
12 |
+
],
|
13 |
+
"type": "llava-lvis4v-lrv"
|
14 |
+
},
|
15 |
+
"hf_token": ".hf_token",
|
16 |
+
"model": {
|
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+fused-gelu-mlp",
|
28 |
+
"enable_gradient_checkpointing": true,
|
29 |
+
"enable_mixed_precision_training": true,
|
30 |
+
"finetune_epochs": 2,
|
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": 16,
|
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": "prism-dinosiglip-224px+7b",
|
44 |
+
"reduce_in_full_precision": false,
|
45 |
+
"type": "prism-dinosiglip-224px+7b",
|
46 |
+
"vision_backbone_id": "dinosiglip-vit-so-224px"
|
47 |
+
},
|
48 |
+
"pretrained_checkpoint": null,
|
49 |
+
"run_id": "llava-lvis4v-lrv+prism-dinosiglip-224px+7b+stage-finetune+x7",
|
50 |
+
"run_root_dir": "/opt/ml/input/data/training/x-prismatic-vlms/runs",
|
51 |
+
"seed": 7,
|
52 |
+
"stage": "finetune",
|
53 |
+
"trackers": [
|
54 |
+
"jsonl",
|
55 |
+
"wandb"
|
56 |
+
],
|
57 |
+
"wandb_entity": "stanford-voltron",
|
58 |
+
"wandb_project": "onyx-vlms"
|
59 |
+
}
|
prism-dinosiglip-224px+7b/config.yaml
ADDED
@@ -0,0 +1,52 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
dataset:
|
2 |
+
align_stage_components:
|
3 |
+
- download/llava-laion-cc-sbu-558k/chat.json
|
4 |
+
- download/llava-laion-cc-sbu-558k
|
5 |
+
dataset_id: llava-lvis4v-lrv
|
6 |
+
dataset_root_dir: /opt/ml/input/data/training/skaramcheti/datasets/prismatic-vlms
|
7 |
+
finetune_stage_components:
|
8 |
+
- download/llava-v1.5-instruct/llava_v1_5_lvis4v_lrv_mix1231k.json
|
9 |
+
- download/llava-v1.5-instruct
|
10 |
+
type: llava-lvis4v-lrv
|
11 |
+
hf_token: .hf_token
|
12 |
+
model:
|
13 |
+
align_epochs: 1
|
14 |
+
align_global_batch_size: 256
|
15 |
+
align_learning_rate: 0.001
|
16 |
+
align_lr_scheduler_type: linear-warmup+cosine-decay
|
17 |
+
align_max_grad_norm: 1.0
|
18 |
+
align_max_steps: null
|
19 |
+
align_per_device_batch_size: 16
|
20 |
+
align_train_strategy: fsdp-shard-grad-op
|
21 |
+
align_warmup_ratio: 0.03
|
22 |
+
align_weight_decay: 0.0
|
23 |
+
arch_specifier: no-align+fused-gelu-mlp
|
24 |
+
enable_gradient_checkpointing: true
|
25 |
+
enable_mixed_precision_training: true
|
26 |
+
finetune_epochs: 2
|
27 |
+
finetune_global_batch_size: 128
|
28 |
+
finetune_learning_rate: 2.0e-05
|
29 |
+
finetune_lr_scheduler_type: linear-warmup+cosine-decay
|
30 |
+
finetune_max_grad_norm: 1.0
|
31 |
+
finetune_max_steps: null
|
32 |
+
finetune_per_device_batch_size: 16
|
33 |
+
finetune_train_strategy: fsdp-full-shard
|
34 |
+
finetune_warmup_ratio: 0.03
|
35 |
+
finetune_weight_decay: 0.1
|
36 |
+
image_resize_strategy: resize-naive
|
37 |
+
llm_backbone_id: llama2-7b-pure
|
38 |
+
llm_max_length: 2048
|
39 |
+
model_id: prism-dinosiglip-224px+7b
|
40 |
+
reduce_in_full_precision: false
|
41 |
+
type: prism-dinosiglip-224px+7b
|
42 |
+
vision_backbone_id: dinosiglip-vit-so-224px
|
43 |
+
pretrained_checkpoint: null
|
44 |
+
run_id: llava-lvis4v-lrv+prism-dinosiglip-224px+7b+stage-finetune+x7
|
45 |
+
run_root_dir: /opt/ml/input/data/training/x-prismatic-vlms/runs
|
46 |
+
seed: 7
|
47 |
+
stage: finetune
|
48 |
+
trackers:
|
49 |
+
- jsonl
|
50 |
+
- wandb
|
51 |
+
wandb_entity: stanford-voltron
|
52 |
+
wandb_project: onyx-vlms
|
prism-dinosiglip-224px+7b/run-metrics.jsonl
ADDED
@@ -0,0 +1 @@
|
|
|
|
|
1 |
+
{"hparams": {"dataset": {"align_stage_components": ["download/llava-laion-cc-sbu-558k/chat.json", "download/llava-laion-cc-sbu-558k"], "dataset_id": "llava-lvis4v-lrv", "dataset_root_dir": "/opt/ml/input/data/training/skaramcheti/datasets/prismatic-vlms", "finetune_stage_components": ["download/llava-v1.5-instruct/llava_v1_5_lvis4v_lrv_mix1231k.json", "download/llava-v1.5-instruct"], "type": "llava-lvis4v-lrv"}, "hf_token": ".hf_token", "model": {"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-shard-grad-op", "align_warmup_ratio": 0.03, "align_weight_decay": 0.0, "arch_specifier": "no-align+fused-gelu-mlp", "enable_gradient_checkpointing": true, "enable_mixed_precision_training": true, "finetune_epochs": 2, "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": 16, "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-dinosiglip-224px+7b", "reduce_in_full_precision": false, "type": "prism-dinosiglip-224px+7b", "vision_backbone_id": "dinosiglip-vit-so-224px"}, "pretrained_checkpoint": null, "run_id": "llava-lvis4v-lrv+prism-dinosiglip-224px+7b+stage-finetune+x7", "run_root_dir": "/opt/ml/input/data/training/x-prismatic-vlms/runs", "seed": 7, "stage": "finetune", "trackers": ["jsonl", "wandb"], "wandb_entity": "stanford-voltron", "wandb_project": "onyx-vlms"}, "run_id": "llava-lvis4v-lrv+prism-dinosiglip-224px+7b+stage-finetune+x7"}
|