Views
No views yet
1pip install huggingface_hub
2huggingface-cli download denkiwakame/Qwen3.5-2B-FFT-VLA0-2-0724 --local-dir ./Qwen3.5-2B-FFT-VLA0-2-07241import pickle
2import torch
3from transformers import Qwen2_5_VLForConditionalGeneration, Qwen2_5_VLProcessor
4
5ckpt_dir = "./Qwen3.5-2B-FFT-VLA0-2-0724"
6
7model = Qwen2_5_VLForConditionalGeneration.from_pretrained(
8 f"{ckpt_dir}/model_final", torch_dtype=torch.bfloat16, device_map="auto",
9)
10processor = Qwen2_5_VLProcessor.from_pretrained(f"{ckpt_dir}/model_final")
11
12# Load dataset stats (required for action denormalization)
13with open(f"{ckpt_dir}/dataset_stats.pkl", "rb") as f:
14 dataset_stats = pickle.load(f)1from rv_train.train import get_pretrained_model
2
3model, cfg = get_pretrained_model("./Qwen3.5-2B-FFT-VLA0-2-0724", device=0)
4model.eval()dataset_stats.pkl1import pickle
2
3with open("dataset_stats.pkl", "rb") as f:
4 stats = pickle.load(f)
5# stats contains mean/std for action dimensionsQwen/Qwen3.5-2B1DATALOADER:
2 ROBOVERSE:
3 cfg_opts: IMAGE.crop_img:0.875:IMAGE.img_size:224:IMAGE.cam_list:('3p1','3p2')
4 cfg_path: libs/RoboVerse/roboverse/configs/img_libero_aug.yaml
5 batch_size: 16
6 num_workers: 8
7EXP:
8 AMP: true
9 DATASET: roboverse
10 EXP_ID: vla0_qwen3_5_2b_fft
11 LOSS: {}
12 LR_SCHED: none
13 MODEL: qwen
14 OPTIMIZER: adamw
15 SEED: 0
16EXP_EXTRA:
17 no_test: true
18 no_track: true
19 no_val: true
20 save_ckp: 2
21 save_last_ckpt: true
22 test_eval_freq: 1
23 val_eval_freq: 1
24LR_SCHED:
25 lr_clip: 1.0e-08
26 lr_decay_factor: 0.5
27 lr_patience: 4
28MODEL:
29 QWEN:
30 action_mask_aug_per: 0.4
31 action_type: original
32 add_vision_id: false
33 attention_dropout: 0.0
34 enable_thinking: false
35 grad_checkpoint: false
36 history: 1
37 horizon: 8
38 lora_config: default
39 lora_rank: 8
40 num_bins_actions: 1000
41 num_cam: 2
42 original_action_dim: 7
43 qwen_model_id: Qwen/Qwen3.5-2B
44 reasoning: false
45 rgb_img_size:
46 - 224
47 - 224
48 rgb_input: true
49 tiled_rgb_imgs: true
50 use_flash_attention_2: true
51 use_lora: false
52 use_qlora: false
53TRAIN:
54 clip_grad_norm: 0.0
55 l2: 1.0e-10
56 lr: 5.0e-06
57 num_epochs: 100
58 num_iters: 18000
59 save_iter_ckp: 17000
60WANDB:
61 enable: true
62 entity: ''
63 example_log_interval: 2000
64 mode: online
65 project: vla0
66 resume_id: ''
67 run_name: ''
68 tags: ''
69
70| File | Description |
|---|---|
model_final/model-*.safetensors | Full model weights |
model_final/config.json | Model configuration |
model_final/tokenizer.json | Tokenizer |
dataset_stats.pkl | Action normalization statistics (required for inference) |
config.yaml | Training configuration |