salma-remyx
commited on
Commit
•
f60e53d
1
Parent(s):
d0d3451
adding SpaceMinitron-4B weights
Browse files- .gitattributes +1 -0
- checkpoints/latest-checkpoint.pt +3 -0
- config.json +59 -0
- config.yaml +52 -0
.gitattributes
CHANGED
@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
|
|
33 |
*.zip filter=lfs diff=lfs merge=lfs -text
|
34 |
*.zst filter=lfs diff=lfs merge=lfs -text
|
35 |
*tfevents* filter=lfs diff=lfs merge=lfs -text
|
|
|
|
33 |
*.zip filter=lfs diff=lfs merge=lfs -text
|
34 |
*.zst filter=lfs diff=lfs merge=lfs -text
|
35 |
*tfevents* filter=lfs diff=lfs merge=lfs -text
|
36 |
+
checkpoints/latest-checkpoint.pt filter=lfs diff=lfs merge=lfs -text
|
checkpoints/latest-checkpoint.pt
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:f8c8ceddb893ef8d95b43e92eeffaffe335a56de36356a10ce0591fb5e239387
|
3 |
+
size 16984255468
|
config.json
ADDED
@@ -0,0 +1,59 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"dataset": {
|
3 |
+
"align_stage_components": [
|
4 |
+
"/home/ubuntu/prismatic-vlms/data/download/llava-v1.5-instruct/llava_v1_5_lrv_mix1008k_spacellava.json",
|
5 |
+
"/home/ubuntu/prismatic-vlms/data/download/llava-v1.5-instruct"
|
6 |
+
],
|
7 |
+
"dataset_id": "spacellava-llava-lrv",
|
8 |
+
"dataset_root_dir": "/home/ubuntu/prismatic-vlms/data/download/llava-v1.5-instruct",
|
9 |
+
"finetune_stage_components": [
|
10 |
+
"/home/ubuntu/prismatic-vlms/data/download/llava-v1.5-instruct/llava_v1_5_lrv_mix1008k_spacellava.json",
|
11 |
+
"/home/ubuntu/prismatic-vlms/data/download/llava-v1.5-instruct"
|
12 |
+
],
|
13 |
+
"type": "spacellava-llava-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": 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": 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": "minitron-4b-pure",
|
42 |
+
"llm_max_length": 2048,
|
43 |
+
"model_id": "minitron+4b+dinosiglip",
|
44 |
+
"reduce_in_full_precision": false,
|
45 |
+
"type": "minitron+4b+dinosiglip",
|
46 |
+
"vision_backbone_id": "dinosiglip-vit-so-224px"
|
47 |
+
},
|
48 |
+
"pretrained_checkpoint": null,
|
49 |
+
"run_id": "spacellava-llava-lrv+minitron+4b+dinosiglip+stage-finetune+x7",
|
50 |
+
"run_root_dir": "runs",
|
51 |
+
"seed": 7,
|
52 |
+
"stage": "finetune",
|
53 |
+
"trackers": [
|
54 |
+
"jsonl",
|
55 |
+
"wandb"
|
56 |
+
],
|
57 |
+
"wandb_entity": "smellslikeml",
|
58 |
+
"wandb_project": "prismatic"
|
59 |
+
}
|
config.yaml
ADDED
@@ -0,0 +1,52 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
dataset:
|
2 |
+
align_stage_components:
|
3 |
+
- /home/ubuntu/prismatic-vlms/data/download/llava-v1.5-instruct/llava_v1_5_lrv_mix1008k_spacellava.json
|
4 |
+
- /home/ubuntu/prismatic-vlms/data/download/llava-v1.5-instruct
|
5 |
+
dataset_id: spacellava-llava-lrv
|
6 |
+
dataset_root_dir: /home/ubuntu/prismatic-vlms/data/download/llava-v1.5-instruct
|
7 |
+
finetune_stage_components:
|
8 |
+
- /home/ubuntu/prismatic-vlms/data/download/llava-v1.5-instruct/llava_v1_5_lrv_mix1008k_spacellava.json
|
9 |
+
- /home/ubuntu/prismatic-vlms/data/download/llava-v1.5-instruct
|
10 |
+
type: spacellava-llava-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: 1
|
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: minitron-4b-pure
|
38 |
+
llm_max_length: 2048
|
39 |
+
model_id: minitron+4b+dinosiglip
|
40 |
+
reduce_in_full_precision: false
|
41 |
+
type: minitron+4b+dinosiglip
|
42 |
+
vision_backbone_id: dinosiglip-vit-so-224px
|
43 |
+
pretrained_checkpoint: null
|
44 |
+
run_id: spacellava-llava-lrv+minitron+4b+dinosiglip+stage-finetune+x7
|
45 |
+
run_root_dir: runs
|
46 |
+
seed: 7
|
47 |
+
stage: finetune
|
48 |
+
trackers:
|
49 |
+
- jsonl
|
50 |
+
- wandb
|
51 |
+
wandb_entity: smellslikeml
|
52 |
+
wandb_project: prismatic
|