Hugging Face's logo

Spaces:
craigwu
/
vstar
Runtime error

App Files Community
4
vstar
File size: 2,405 Bytes 
3672502
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
 
#!/bin/bash

MODEL_VERSION=vicuna-v1-3-7b

PROMPT_VERSION=plain
########### DO NOT CHANGE ###########

# pre-train the linear projection
deepspeed llava/train/train_mem.py \
    --deepspeed ./scripts/zero2.json \
    --model_name_or_path lmsys/vicuna-7b-v1.3 \
    --version $PROMPT_VERSION \
    --data_path /path/to/blip_laion_cc_sbu_558k.json \
    --image_folder /path/to/images \
    --vision_tower openai/clip-vit-large-patch14 \
    --mm_projector_type linear \
    --tune_mm_mlp_adapter True \
    --mm_vision_select_layer -2 \
    --mm_use_im_start_end False \
    --mm_use_im_patch_token False \
    --bf16 True \
    --output_dir ./checkpoints/llava-$MODEL_VERSION-linear-pretrain \
    --num_train_epochs 1 \
    --per_device_train_batch_size 32 \
    --per_device_eval_batch_size 4 \
    --gradient_accumulation_steps 1 \
    --evaluation_strategy "no" \
    --save_strategy "steps" \
    --save_steps 24000 \
    --save_total_limit 1 \
    --learning_rate 2e-3 \
    --weight_decay 0. \
    --warmup_ratio 0.03 \
    --lr_scheduler_type "cosine" \
    --logging_steps 1 \
    --tf32 True \
    --model_max_length 2048 \
    --gradient_checkpointing True \
    --dataloader_num_workers 4 \
    --lazy_preprocess True \
    --report_to wandb

# pre-train the re-sampler

deepspeed llava/train/train_mem.py \
    --deepspeed ./scripts/zero2.json \
    --model_name_or_path lmsys/vicuna-7b-v1.3 \
    --version $PROMPT_VERSION \
    --data_path /path/to/blip_laion_cc_sbu_558k.json \
    --image_folder /path/to/images \
    --vision_tower openai/clip-vit-large-patch14 \
    --mm_projector_type perceiver \
    --tune_mm_mlp_adapter True \
    --mm_vision_select_layer -2 \
    --mm_use_im_start_end False \
    --mm_use_im_patch_token False \
    --bf16 True \
    --output_dir ./checkpoints/llava-$MODEL_VERSION-resampler-pretrain \
    --num_train_epochs 5 \
    --per_device_train_batch_size 32 \
    --per_device_eval_batch_size 4 \
    --gradient_accumulation_steps 1 \
    --evaluation_strategy "no" \
    --save_strategy "steps" \
    --save_steps 24000 \
    --save_total_limit 1 \
    --learning_rate 2e-4 \
    --weight_decay 0. \
    --warmup_ratio 0.03 \
    --lr_scheduler_type "cosine" \
    --logging_steps 1 \
    --tf32 True \
    --model_max_length 2048 \
    --gradient_checkpointing True \
    --dataloader_num_workers 4 \
    --lazy_preprocess True \
    --report_to wandb