 Hawk Lee
|
450187c5b3
|
fix(nodes): prevent infinite loop in progress bar calculation
|
2026-01-20 00:27:44 +08:00 |
|
|
Hawk Lee
|
efc42867ff
|
fix(transformer): fix device mismatch in time_embedding
|
2026-01-20 00:23:49 +08:00 |
|
|
Hawk Lee
|
02a1f52b8f
|
feat(nodes): add realtime progress bar
|
2026-01-20 00:23:09 +08:00 |
|
|
Hawk Lee
|
895981804a
|
chore: remove debug prints
|
2026-01-20 00:17:41 +08:00 |
|
|
Hawk Lee
|
66527a866f
|
fix(test): correct clip encoder input shape
|
2026-01-20 00:14:02 +08:00 |
|
|
Hawk Lee
|
2c903784b7
|
chore: add test_official_style.py
|
2026-01-20 00:11:27 +08:00 |
|
|
Hawk Lee
|
31c7b08a6b
|
fix(wan): ensure input tensor is on correct device in patch_embed
|
2026-01-20 00:08:37 +08:00 |
|
|
Hawk Lee
|
53ac7dc881
|
fix(wan): disable autocast for patch_embedding
|
2026-01-20 00:06:04 +08:00 |
|
|
Hawk Lee
|
5254567c32
|
fix(wan): force F.conv3d with explicit float32 weights
|
2026-01-20 00:03:18 +08:00 |
|
|
Hawk Lee
|
f9f1c58cc7
|
fix(test): correct encode_prompt arguments
|
2026-01-20 00:00:49 +08:00 |
|
|
Hawk Lee
|
28d22fb086
|
chore: add final verification script
|
2026-01-19 23:57:55 +08:00 |
|
|
Hawk Lee
|
798d742a0f
|
fix(test): add missing torchaudio import
|
2026-01-19 23:55:41 +08:00 |
|
|
Hawk Lee
|
b6d9a925a2
|
chore: add version tag to test script
|
2026-01-19 23:54:07 +08:00 |
|
|
Hawk Lee
|
8101c25d81
|
fix(test): use explicit pth paths for encoders
|
2026-01-19 23:52:40 +08:00 |
|
|
Hawk Lee
|
caa24eec2a
|
chore: add test_server_fix.py with correct paths
|
2026-01-19 23:51:21 +08:00 |
|
|
Hawk Lee
|
cfdfd564ba
|
fix(test): correct model paths in server verification script
|
2026-01-19 23:49:31 +08:00 |
|
|
Hawk Lee
|
7d42466b38
|
chore: add manual test script for server verification
|
2026-01-19 23:45:52 +08:00 |
|
|
Hawk Lee
|
fa50a5dae2
|
fix(transformer): use in-place float32 conversion for patch_embedding
|
2026-01-19 23:43:44 +08:00 |
|
|
Hawk Lee
|
f1848bb81a
|
fix(transformer): disable autocast for forced float32 conv3d
|
2026-01-19 23:38:14 +08:00 |
|
|
Hawk Lee
|
67cc60515b
|
fix(transformer): force patch_embedding to float32
|
2026-01-19 23:33:20 +08:00 |
|
|
Hawk Lee
|
09292cb631
|
fix(transformer): add .contiguous() to patch_embedding input
|
2026-01-19 23:26:04 +08:00 |
|
|
Hawk Lee
|
ccd416ad2a
|
fix(pipeline): ensure correct device usage in loop
|
2026-01-19 23:22:18 +08:00 |
|
|
Hawk Lee
|
7c14df3fbe
|
fix(pipeline): fix indentation bug in prepare_mask_latents
|
2026-01-19 23:13:29 +08:00 |
|
|
Hawk Lee
|
220e1146b9
|
fix(echomimic): add .contiguous() to vae and move back to GPU
|
2026-01-19 23:10:12 +08:00 |
|
|
Hawk Lee
|
312aa234f0
|
feat(debug): add logging to VAE encoding loop
|
2026-01-19 23:08:33 +08:00 |
|
|
Hawk Lee
|
130ae46f86
|
fix(echomimic): run VAE on CPU to avoid cuda 3D conv error
|
2026-01-19 22:51:47 +08:00 |
|
|
Hawk Lee
|
b1df960d51
|
fix(echomimic): force VAE to float32 to fix slow_conv3d_forward error on cuda
|
2026-01-19 22:19:19 +08:00 |
|
|
Hawk Lee
|
59ff8bd825
|
fix(echomimic): restore missing ip_mask definition
|
2026-01-19 22:05:14 +08:00 |
|
|
Hawk Lee
|
3a15acebc8
|
fix(echomimic): use cpu generator to avoid device mismatch with offloaded text encoder
|
2026-01-19 22:01:00 +08:00 |
|
|
Hawk Lee
|
5728e75829
|
fix(pipeline): add clip_context argument to __call__
|
2026-01-19 21:48:02 +08:00 |
|
|
Hawk Lee
|
97328f2d90
|
fix(echomimic): add missing TF import
|
2026-01-19 21:45:08 +08:00 |
|
|
Hawk Lee
|
10714a2e56
|
fix(segments): add aggressive vram cleanup for nemo model
|
2026-01-19 21:41:30 +08:00 |
|
|
Hawk Lee
|
7bc72eca9e
|
fix(audio): aggressive vram cleanup for vibevoice and audio_enhance nodes
|
2026-01-19 21:39:56 +08:00 |
|
|
Hawk Lee
|
d75303b1b6
|
perf(echomimic): pre-compute CLIP context to save VRAM
|
2026-01-19 21:35:03 +08:00 |
|
|
Hawk Lee
|
317b693a4a
|
perf(echomimic): permit passing pre-computed clip_context to pipeline
|
2026-01-19 21:34:26 +08:00 |
|
|
Hawk Lee
|
c814546fb0
|
perf(echomimic): add post-generation model cleanup (move to CPU)
|
2026-01-19 21:26:01 +08:00 |
|
|
Hawk Lee
|
cd87d26437
|
perf(echomimic): implement manual device management for robust speed optimization
|
2026-01-19 21:25:28 +08:00 |
|
|
Hawk Lee
|
741a95909a
|
fix(echomimic): handle missing text_encoder in pipeline to support memory optimization
|
2026-01-19 21:17:54 +08:00 |
|
|
Hawk Lee
|
aa67ce4b13
|
perf(echomimic): delete text_encoder after use to free 10GB VRAM & remove debug prints
|
2026-01-19 21:11:20 +08:00 |
|
|
Hawk Lee
|
cf0775785d
|
debug: add print statements to pipeline
|
2026-01-19 21:05:57 +08:00 |
|
|
Hawk Lee
|
cd701b0032
|
fix(echomimic): correct batch_size logic in pipeline __call__ method
|
2026-01-19 21:01:11 +08:00 |
|
|
Hawk Lee
|
d51c6c07e9
|
fix(echomimic): correct batch_size logic (take 2)
|
2026-01-19 20:50:46 +08:00 |
|
|
Hawk Lee
|
5f48fd666c
|
fix(echomimic): revert transpose and implement correct flat-list CFG
|
2026-01-19 20:41:19 +08:00 |
|
|
Hawk Lee
|
13708df19c
|
fix(echomimic): transpose T5 embeddings to Time-Major [S,B,D] for transformer
|
2026-01-19 20:31:36 +08:00 |
|
|
Hawk Lee
|
d624a8e06f
|
fix(echomimic): correct I2V context concatenation for 4D T5 layers
|
2026-01-19 20:26:04 +08:00 |
|
|
Hawk Lee
|
dd0a41ae72
|
fix(echomimic): correct CFG embedding concatenation for list inputs
|
2026-01-19 20:20:26 +08:00 |
|
|
Hawk Lee
|
e9e5141a0c
|
fix(echomimic): syntax error in pipeline
|
2026-01-19 20:10:13 +08:00 |
|
|
Hawk Lee
|
adb2e780d9
|
fix(echomimic): handle list-type prompt_embeds in encode_prompt
|
2026-01-19 20:09:55 +08:00 |
|
|
Hawk Lee
|
d928adb501
|
fix(echomimic): add mm import
|
2026-01-19 20:03:15 +08:00 |
|
|
Hawk Lee
|
5aaf22b20a
|
fix(echomimic): handle list embeddings in pipeline and verbose memory cleanup
|
2026-01-19 20:00:52 +08:00 |
|