|
|
ef56e5dcdb
|
Revert "tensorRT engines尝试精度没过，暂时先提交代码，后续再继续调试"
This reverts commit e1f8a83648.
|
2026-02-19 20:22:19 +08:00 |
|
|
|
e1f8a83648
|
tensorRT engines尝试精度没过，暂时先提交代码，后续再继续调试
|
2026-02-18 18:22:12 +08:00 |
|
|
|
5e0e21d91b
|
复原sh为原始版本
|
2026-02-18 14:11:55 +08:00 |
|
|
|
d5bec53f61
|
优化后的全部结果
|
2026-02-11 19:21:06 +08:00 |
|
|
|
508b91f5a2
|
延迟 decode，只解码 CLIP 需要的 1 帧
- world model 调用 decode_video=False，跳过 16 帧全量 decode
- 只 decode 最后 1 帧给 CLIP embedding / observation queue
- 存 raw latent，循环结束后统一 batch decode 生成最终视频
- 每轮省 15 次 VAE decode，8 轮共省 120 次
- 跳过中间迭代的 wm tensorboard/mp4 保存
psnr微弱下降
|
2026-02-11 17:07:33 +08:00 |
|
|
|
3101252c25
|
速度变化不明显psnr显著提升
|
2026-02-11 16:38:21 +08:00 |
|
|
|
f386a5810b
|
补充上次提交
|
2026-02-11 16:24:40 +08:00 |
|
|
|
352a79035f
|
主干部分fp16,最敏感psnr=25.21,可以考虑对主干部分太敏感的部分回退fp32
|
2026-02-11 16:23:21 +08:00 |
|
|
|
9a08e27a19
|
KV 融合实现完成。改动总结： 速度微弱提升psnr略微上升
attention.py — 3处改动：
1. __init__ 添加 _kv_fused = False 标志
2.新增 fuse_kv() 方法：将 to_k + to_v → to_kv，同时处理 _ip/_as/_aa 辅助 KV 对
2. bmm_forward 两个分支加_kv_fused 判断，用to_kv().chunk(2, dim=-1) 替代分别调用
|
2026-02-11 12:36:38 +08:00 |
|
|
|
b558856e1e
|
fix bugs
|
2026-02-10 22:35:45 +08:00 |
|
|
|
dcbcb2c377
|
- state_unet 放到一个独立的 CUDA stream 上执行
- action_unet 在默认 stream 上同时执行
- 用 wait_stream 确保两者都完成后再返回
两个 1D UNet 输入完全独立，共享的 hs_a 和 context_action 都是只读的。GPU 利用率只有 ~31%，小张量 kernel 不会打满 GPU，两个 stream 可以真正并行。
|
2026-02-10 21:41:48 +08:00 |
|
|
|
ff43432ef9
|
结果
|
2026-02-10 20:01:25 +08:00 |
|
|
|
afa12ba031
|
每步迭代保存异步
|
2026-02-10 19:54:53 +08:00 |
|
|
|
bf4d66c874
|
跳过模型加载
|
2026-02-10 19:36:17 +08:00 |
|
|
|
9347a4ebe5
|
实现了Context 预计算和缓存功能，提升了采样效率。 psnr不下降
|
2026-02-10 17:47:46 +08:00 |
|
|
|
223a50f9e0
|
添加CrossAttention kv缓存，减少重复计算，提升性能，psnr=25.1201dB
|
2026-02-10 17:35:03 +08:00 |
|
|
|
2a6068f9e4
|
减少了一路视频vae解码
|
2026-02-10 17:13:45 +08:00 |
|
|
|
91a9b0febc
|
DDIM loop 内小张量分配优化，attention mask 缓存到 GPU
|
2026-02-10 16:53:00 +08:00 |
|
|
|
ed637c972b
|
tf32推理
|
2026-02-10 16:39:14 +08:00 |
|
|
|
fffc5a9956
|
init
|
2026-02-08 03:29:15 +00:00 |
|
UniGen-X
|
cbaebc016f
|
Update README.md
|
2025-10-01 10:13:04 +08:00 |
|
hengguo
|
05d2d82236
|
update readme
|
2025-09-23 16:57:54 +08:00 |
|
yuchen-x
|
50e8c3ed55
|
update readme_cn
|
2025-09-23 16:15:59 +08:00 |
|
yuchen-x
|
ddb5848d86
|
Merge branch 'main' of github.com:unitreerobotics/unifolm-world-model-action
|
2025-09-23 16:11:47 +08:00 |
|
yuchen-x
|
cb0cf4a353
|
updata readme_cn
|
2025-09-23 16:11:29 +08:00 |
|
hengguo
|
54f61a4336
|
update readme_cn
|
2025-09-23 16:00:53 +08:00 |
|
yuchen-x
|
118ada7c35
|
update readme
|
2025-09-23 15:22:27 +08:00 |
|
yuchen-x
|
8d5546d322
|
update readme
|
2025-09-23 15:21:12 +08:00 |
|
yuchen-x
|
eccd1680c1
|
update readme
|
2025-09-23 15:19:02 +08:00 |
|
yuchen-x
|
f12b478265
|
upload real-robot deployment code
|
2025-09-23 15:13:22 +08:00 |
|
yuchen-x
|
5dcd1ca503
|
fix a typo on COLUMNS definition
|
2025-09-22 17:33:28 +08:00 |
|
UniGen-X
|
7b4d383611
|
Update README.md
|
2025-09-21 17:34:40 +08:00 |
|
UniGen-X
|
733e228bb8
|
Update README.md
|
2025-09-21 17:19:28 +08:00 |
|
UniGen-X
|
e9c60f6e62
|
Update README_cn.md
|
2025-09-19 10:18:38 +08:00 |
|
UniGen-X
|
2d4d79ab3a
|
Update README.md
|
2025-09-19 10:18:07 +08:00 |
|
UniGen-X
|
1f21fe7fd8
|
Update README.md
|
2025-09-19 10:16:58 +08:00 |
|
UniGen-X
|
a57037ab03
|
Update README_cn.md
|
2025-09-17 11:45:09 +08:00 |
|
UniGen-X
|
be43dfef9d
|
Update README.md
|
2025-09-17 11:44:32 +08:00 |
|
UniGen-X
|
884dcce130
|
Update README_cn.md
|
2025-09-17 10:49:42 +08:00 |
|
UniGen-X
|
29f3101a1f
|
Update README.md
|
2025-09-17 10:49:20 +08:00 |
|
UniGen-X
|
712a289c28
|
Update README_cn.md
|
2025-09-17 10:44:52 +08:00 |
|
UniGen-X
|
c0fad43420
|
Update README.md
|
2025-09-17 10:44:30 +08:00 |
|
UniGen-X
|
3b83374922
|
Merge pull request #8 from hu-po/patch-1
Fix typo in Acknowledgement section of README
|
2025-09-17 10:31:31 +08:00 |
|
UniGen-X
|
6a33fddb99
|
Update README_cn.md
|
2025-09-17 10:29:47 +08:00 |
|
UniGen-X
|
a0010c4036
|
Update README.md
|
2025-09-17 10:27:02 +08:00 |
|
UniGen-X
|
a1f6430c38
|
Update README.md
|
2025-09-17 10:26:51 +08:00 |
|
hu-po
|
ee45fe75e2
|
Fix typo in Acknowledgement section of README
|
2025-09-16 13:22:33 -05:00 |
|
yuchen-x
|
9c9942f1d4
|
upload eval_utils.py file
|
2025-09-16 21:48:15 +08:00 |
|
UniGen-X
|
a457f2f049
|
Update README.md
|
2025-09-16 20:55:49 +08:00 |
|
UniGen-X
|
ea7260a7e0
|
Update README_cn.md
|
2025-09-15 18:28:52 +08:00 |
|
