Hugging Face's logo

Stanford-ILIAD
/
prism-qwen25-extra-dinosiglip-224px-0_5b

Image-Text-to-Text
Transformers
English
robotics
vlm
multimodal
pretraining
Inference Endpoints
Model card Files Community
prism-qwen25-extra-dinosiglip-224px-0_5b
File size: 2,914 Bytes 
ca81022
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
 
_wandb:
    value:
        cli_version: 0.18.5
        m: []
        python_version: 3.10.15
        t:
            "1":
                - 1
                - 2
                - 3
                - 11
                - 41
                - 49
                - 55
                - 63
                - 71
            "2":
                - 1
                - 2
                - 3
                - 11
                - 41
                - 49
                - 55
                - 63
                - 71
            "3":
                - 2
                - 13
                - 16
                - 23
                - 55
                - 61
            "4": 3.10.15
            "5": 0.18.5
            "6": 4.40.1
            "8":
                - 5
            "12": 0.18.5
            "13": linux-x86_64
dataset:
    value:
        align_stage_components:
            - download/llava-laion-cc-sbu-558k/chat.json
            - download/llava-laion-cc-sbu-558k
        dataset_id: llava-v15
        dataset_root_dir: /hai/scratch/belkhale/datasets/prismatic-vlms
        finetune_stage_components:
            - download/llava-v1.5-instruct/llava_v1_5_mix665k.json
            - download/llava-v1.5-instruct
        type: llava-v15
hf_token:
    value: .hf_token
model:
    value:
        align_epochs: 1
        align_global_batch_size: 96
        align_learning_rate: 0.001
        align_lr_scheduler_type: linear-warmup+cosine-decay
        align_max_grad_norm: 1
        align_max_steps: null
        align_per_device_batch_size: 16
        align_save_every_n_steps: 10000
        align_train_strategy: fsdp-shard-grad-op
        align_warmup_ratio: 0.03
        align_weight_decay: 0
        arch_specifier: no-align+fused-gelu-mlp
        enable_gradient_checkpointing: true
        enable_mixed_precision_training: true
        finetune_epochs: 2
        finetune_global_batch_size: 64
        finetune_learning_rate: 2e-05
        finetune_lr_scheduler_type: linear-warmup+cosine-decay
        finetune_max_grad_norm: 1
        finetune_max_steps: null
        finetune_per_device_batch_size: 4
        finetune_save_every_n_steps: 10000
        finetune_train_strategy: fsdp-full-shard
        finetune_warmup_ratio: 0.03
        finetune_weight_decay: 0.1
        image_resize_strategy: resize-naive
        llm_backbone_id: qwen25-0_5b-extra
        llm_max_length: 32768
        model_id: prism-qwen25-extra-dinosiglip-224px+0_5b
        reduce_in_full_precision: false
        type: prism-qwen25-extra-dinosiglip-224px+0_5b
        vision_backbone_id: dinosiglip-vit-so-224px
pretrained_checkpoint:
    value: null
run_id:
    value: prism-qwen25-extra-dinosiglip-224px+0_5b+stage-finetune+x7
run_root_dir:
    value: runs
seed:
    value: 7
stage:
    value: finetune
trackers:
    value:
        - jsonl
        - wandb
wandb_entity:
    value: null
wandb_project:
    value: prismatic