Logic
|
61522d9ae5
|
feat(lewm): enable gpu parallel rollout validation
|
2026-04-23 15:11:41 +08:00 |
|
Logic
|
4cd33258d2
|
fix: support dual decoder relaunch and width-adaptive lewm tokens
|
2026-04-17 20:11:54 +08:00 |
|
Logic
|
d8066823e2
|
chore: inherit lewm dataset settings from agent config
|
2026-04-17 19:33:08 +08:00 |
|
Logic
|
395f5a1645
|
feat: add held-out validation and dual-decoder lewm imf
|
2026-04-17 19:26:56 +08:00 |
|
Logic
|
74f4963613
|
feat: add lewm-conditioned imf training and sigreg loss
|
2026-04-17 18:46:02 +08:00 |
|
Logic
|
ff7c9c1f2a
|
feat: add vision transfer backbones and IMF variants
|
2026-04-09 14:02:24 +08:00 |
|
Logic
|
d51b3ecafa
|
docs: finalize phase1 report and launch phase2 attnres suite
|
2026-04-05 00:39:04 +08:00 |
|
Logic
|
2033169840
|
feat: add full attnres vision backbone
|
2026-04-05 00:07:59 +08:00 |
|
Logic
|
a78006808a
|
fix: stabilize headless rollout and summarize phase1 grid
|
2026-04-04 23:47:15 +08:00 |
|
Logic
|
0586a6e6c7
|
feat: add rollout trajectory image artifacts and swanlab logging
|
2026-04-03 09:39:16 +08:00 |
|
Logic
|
48f0eb8dd0
|
fix: defer MuJoCo env import until headless GL is configured
|
2026-04-02 08:26:39 +08:00 |
|
Logic
|
3a17744dcf
|
fix: support headless MuJoCo image capture during rollout
|
2026-04-02 08:25:01 +08:00 |
|
Logic
|
0514f86c36
|
fix: support headless rollout on remote training hosts
|
2026-04-02 08:15:54 +08:00 |
|
Logic
|
dffd92f82d
|
fix: sanitize problematic LD_PRELOAD for cuDNN
|
2026-04-01 23:50:57 +08:00 |
|
Logic
|
c2000b5533
|
feat: add IMF AttnRes policy training path
|
2026-04-01 23:35:31 +08:00 |
|
Logic
|
8d6060224a
|
docs: refine IMF migration spec defaults
|
2026-04-01 22:55:50 +08:00 |
|
Logic
|
8a8193fe7e
|
docs: add IMF-AttnRes migration implementation plan
|
2026-04-01 22:52:39 +08:00 |
|
Logic
|
1a92c5e8a6
|
docs: add IMF-AttnRes migration design spec
|
2026-04-01 22:49:10 +08:00 |
|
Logic
|
b76bcd8b37
|
chore: ignore local git worktrees
|
2026-04-01 22:46:02 +08:00 |
|
Logic
|
2f9b99e0c4
|
feat(vis): add raw action trajectory viewer
|
2026-04-01 22:27:22 +08:00 |
|
Logic
|
d5d5b53f71
|
feat(data): stream sim episodes with raw ee actions
|
2026-03-31 15:44:53 +08:00 |
|
Logic
|
d84bc6876e
|
feat(vla): align transformer training stack and rollout validation
|
2026-03-31 15:39:20 +08:00 |
|
Logic
|
424c265823
|
feat(eval): export rollout video timing and ee trajectory
|
2026-03-31 15:34:28 +08:00 |
|
Logic
|
cb79e00546
|
docs: add VLA training headless swanlab design spec
|
2026-03-30 18:50:12 +08:00 |
|
gouhanke
|
23088e5e33
|
feat: 架构引入DiT
|
2026-03-06 11:31:37 +08:00 |
|
gouhanke
|
ca1716c67f
|
chore: 导入gr00t
|
2026-03-06 11:31:37 +08:00 |
|
JiajunLI
|
642d41dd8f
|
feat: 添加resume机制
|
2026-03-06 11:19:30 +08:00 |
|
gouhanke
|
7d39933a5b
|
feat: 缓存worker内的句柄
|
2026-03-04 10:49:41 +08:00 |
|
gouhanke
|
8bcad5844e
|
fix: 修复VLA设备与损失计算逻辑，并优化Transformer默认训练参数
|
2026-03-03 17:56:12 +08:00 |
|
gouhanke
|
cdb887c9bf
|
feat: 添加transformer头
|
2026-02-28 19:07:27 +08:00 |
|
gouhanke
|
abb4f501e3
|
chore: 删除unet里的local_cond(未使用)
|
2026-02-28 10:42:16 +08:00 |
|
gouhanke
|
1d33db0ef0
|
chore: 缩小物块的大小
|
2026-02-27 18:23:30 +08:00 |
|
gouhanke
|
f27e397f98
|
chore: 修改了采数时的一些参数
|
2026-02-26 17:09:40 +08:00 |
|
gouhanke
|
4e0add4e1d
|
debug: 修复episode首帧图像不正确的问题；修复前2个episode帧重复的问题
|
2026-02-26 16:17:54 +08:00 |
|
gouhanke
|
40c40695dd
|
chore: 添加测试文件
- check_all_episodes.py：检查各个episode是否有重复帧。
- check_specific_frames.py：检查前几帧是否位于正确初始位置。
- generate_dataset_videos.py：根据hdf5生成视频
|
2026-02-26 13:59:47 +08:00 |
|
gouhanke
|
3deeffb9fe
|
chore:改变了一些参数配置
|
2026-02-26 13:56:03 +08:00 |
|
gouhanke
|
0b05c01024
|
feat: 推理时输出action
|
2026-02-12 19:54:11 +08:00 |
|
gouhanke
|
926a78eb66
|
feat: 添加finetune
|
2026-02-12 19:31:44 +08:00 |
|
gouhanke
|
efbe4b6ac9
|
Revert "Merge branch 'dev' of gitlab.com:leeeezd0016-group/gouhanke-vla into dev"
This reverts commit acb1467473, reversing
changes made to 624b926e33.
|
2026-02-12 18:31:56 +08:00 |
|
gouhanke
|
acb1467473
|
Merge branch 'dev' of gitlab.com:leeeezd0016-group/gouhanke-vla into dev
|
2026-02-12 18:08:27 +08:00 |
|
gouhanke
|
624b926e33
|
debug: 添加推理时缩放，加大采数以及推理时物块的放置范围
|
2026-02-12 17:14:23 +08:00 |
|
gouhanke
|
926d8cf894
|
chore: 加载时将图像缩放到224*224， resnet禁用crop
|
2026-02-12 15:02:18 +08:00 |
|
gouhanke
|
116ba13fb9
|
chore: 验证归一化是否有效
|
2026-02-12 13:01:13 +08:00 |
|
gouhanke
|
37a47ac2dd
|
debug: 保存stats到ckpt
|
2026-02-12 13:00:43 +08:00 |
|
gouhanke
|
ab971b3f96
|
debug: 归一化
|
2026-02-12 12:23:34 +08:00 |
|
gouhanke
|
83cd55e67b
|
添加pad_loss
|
2026-02-11 20:33:26 +08:00 |
|
gouhanke
|
eeb07cad15
|
feat: 冻结resnet
|
2026-02-11 20:11:25 +08:00 |
|
gouhanke
|
83d11ab640
|
Merge branch 'dev' of gitlab.com:leeeezd0016-group/gouhanke-vla into dev
|
2026-02-11 17:20:21 +08:00 |
|
JiajunLI
|
aba8779671
|
更改默认参数
|
2026-02-11 17:14:32 +08:00 |
|
gouhanke
|
b42c1c68fd
|
debug: 将归一化放在GPU上
|
2026-02-11 17:13:55 +08:00 |
|