alikia2x/cvsa
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

improve: config for filter model

Browse Source
This commit is contained in:
alikia2x (寒寒) 2025-03-07 22:50:22 +08:00
parent 638135dde6
commit 1ef18ed20d
Signed by: alikia2x
GPG Key ID: 56209E0CCD8420C6
1 changed files with 4 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
filter/train.py
View File

@ -61,7 +61,7 @@ class_weights = torch.tensor(
device=device
)
model = VideoClassifierV6_1().to(device)
model = VideoClassifierV6_1(num_heads=8).to(device)
checkpoint_name = './filter/checkpoints/best_model_V6.1.pt'
# 初始化tokenizer和embedding模型
@ -75,10 +75,10 @@ os.makedirs('./filter/checkpoints', exist_ok=True)
eval_interval = 20
num_epochs = 20
total_steps = samples_count * num_epochs / batch_size
warmup_rate = 0.1
optimizer = optim.AdamW(model.parameters(), lr=3e-5, weight_decay=1e-3)
warmup_rate = 0.15
optimizer = optim.AdamW(model.parameters(), lr=3e-5, weight_decay=1e-4)
cosine_annealing_scheduler = optim.lr_scheduler.CosineAnnealingLR(optimizer, T_max=total_steps - int(total_steps * warmup_rate))
warmup_scheduler = optim.lr_scheduler.LinearLR(optimizer, start_factor=0.8, end_factor=1.0, total_iters=int(total_steps * warmup_rate))
warmup_scheduler = optim.lr_scheduler.LinearLR(optimizer, start_factor=0.4, end_factor=1.0, total_iters=int(total_steps * warmup_rate))
scheduler = optim.lr_scheduler.SequentialLR(optimizer, schedulers=[warmup_scheduler, cosine_annealing_scheduler], milestones=[int(total_steps * warmup_rate)])
criterion = nn.CrossEntropyLoss(weight=class_weights).to(device)