nas / LAVIS-main /lavis /projects /instructblip /classification_modelnet40_vicuna7b.yaml
yuccaaa's picture
Add files using upload-large-folder tool
31ec239 verified
# Copyright (c) 2022, salesforce.com, inc.
# All rights reserved.
# SPDX-License-Identifier: BSD-3-Clause
# For full license text, see the LICENSE file in the repo root or https://opensource.org/licenses/BSD-3-Clause
model:
arch: blip2_vicuna_instruct
model_type: vicuna7b
load_pretrained: True
prompt: "describe the 3d model."
format_candidates_prompt: " a 3d model of a {}"
datasets:
modelnet40_cls: # name of the dataset builder
data_type: [pc, images]
vis_processor:
train:
name: "clip_image_train"
image_size: 224
eval:
name: "clip_image_eval"
image_size: 224
pc_processor:
train:
name: "ulip_pc"
eval:
name: "ulip_pc"
text_processor:
train:
name: "blip_caption"
eval:
name: "blip_caption"
build_info:
# Be careful not to append minus sign (-) before split to avoid itemizing
annotations:
train:
url:
- /export/home/ULIP/data/modelnet40_normal_resampled/modelnet40_shape_names.txt
- /export/home/ULIP/data/modelnet40_normal_resampled/modelnet40_train_8192pts_fps.dat
- /export/home/ULIP/data/modelnet40_normal_resampled/modelnet40_train.txt
storage:
- /export/home/ULIP/data/modelnet40_normal_resampled/modelnet40_shape_names.txt
- /export/home/ULIP/data/modelnet40_normal_resampled/modelnet40_train_8192pts_fps.dat
- /export/home/ULIP/data/modelnet40_normal_resampled/modelnet40_train.txt
val:
url:
- /export/home/ULIP/data/modelnet40_normal_resampled/modelnet40_shape_names.txt
- /export/home/ULIP/data/modelnet40_normal_resampled/modelnet40_test_8192pts_fps.dat
- /export/home/ULIP/data/modelnet40_normal_resampled/modelnet40_test.txt
storage:
- /export/home/ULIP/data/modelnet40_normal_resampled/modelnet40_shape_names.txt
- /export/home/ULIP/data/modelnet40_normal_resampled/modelnet40_test_8192pts_fps.dat
- /export/home/ULIP/data/modelnet40_normal_resampled/modelnet40_test.txt
pc:
storage: /export/home/ULIP/data/modelnet40_normal_resampled
images:
storage: /export/einstein-vision/3d_vision/3d_object_datasets/modelnet_images8192
run:
task: multimodal_classification
# optimizer
lr_sched: "linear_warmup_cosine_lr"
init_lr: 1e-5
min_lr: 0
warmup_lr: 1e-8
warmup_steps: 1000
weight_decay: 0.05
max_epoch: 1
batch_size_train: 16
batch_size_eval: 1
num_workers: 8
accum_grad_iters: 1
prompt: "describe the 3d model."
max_len: 3
min_len: 1
num_beams: 5
seed: 42
output_dir: "output/instructblip/modelent_classification_vicuna7b/"
amp: True
resume_ckpt_path: null
evaluate: True
# train_splits: ["train"]
valid_splits: ["val"]
device: "cuda"
world_size: 1
dist_url: "env://"
distributed: True
save_freq: -1 # save epoch every xxx epochs -1 only save last and best.
val_freq: 1