47356cd864
Testing the original code path.
...
Need to check rectified linear flow vs flow matching.
2025-01-17 06:03:45 +00:00
cd66e6e58e
fixed the training pipeline to address new changes from upstream
2025-01-17 01:46:35 +00:00
315431be0a
Recommit the conditioning code.
2025-01-16 05:03:44 +00:00
b341b79652
Adding Conditioning Blocks to the Diffusion Transformer.
2025-01-15 22:29:47 +00:00
9f0fdb5055
Rename lora files ( #214 )
...
* rename files
* update
2025-01-14 01:17:42 +05:30
354a6b9e25
reposition private methods ( #213 )
2025-01-14 00:08:04 +05:30
e5df80cc36
Full Finetuning for LTX possibily extended to other models. ( #192 )
...
* Full Finetuning for LTX possibily extended to other models.
* Change name of the flag
* Used disable grad for component on lora fine tuning enabled
* Suggestions Addressed
Renamed to SFT
Added 2 other models.
Testing required.
* Switching to Full FineTuning
* Run linter.
* parse subfolder when needed.
* tackle saving and loading hooks.
* tackle validation.
* fix subfolder bug.
* remove __class__.
* refactor
* remove unnecessary changes
* handle saving of final model weights correctly
* remove unnecessary changes
* LTX uses a default frame rate of 24 FPS
We need to modify the output validation framerate to match that value.
Add Framerate args.
Add Update video output and inference frame rate
* There was a results_args mapping that needed to be modified.
* update
* update README
* Update README.md
* update docs
* add training configuration in cogvideox
---------
Co-authored-by: Sayak Paul <spsayakpaul@gmail.com >
Co-authored-by: Aryan <aryan@huggingface.co >
Co-authored-by: Aryan <contact.aryanvs@gmail.com >
2025-01-13 23:55:39 +05:30
a3898075d1
Helpful mesages ( #210 )
2025-01-11 06:35:28 +05:30
d3b797ee35
Improve argument handling ( #209 )
2025-01-11 06:18:13 +05:30
81bf590a91
Argument descriptions ( #208 )
...
* add argument descriptions
* update
* update
* update
2025-01-11 05:06:02 +05:30
b34091362e
Fix the checkpoint dir bug in get_intermediate_ckpt_path so that checkpointing_limit can actually work. ( #207 )
2025-01-10 13:18:32 +05:30
b9a6492ad1
Epoch loss ( #201 )
...
* update
* update
2025-01-09 14:22:56 +05:30
f311f16d98
[core] Fix loading of precomputed conditions and latents ( #199 )
...
* support precomputation.
* fixes
2025-01-09 12:43:03 +05:30
dbffc80e7c
Precomputation folder name based on model name ( #196 )
...
* update
* update
* fix
* address review comments
2025-01-08 08:45:44 +05:30
905c22edd8
Fix minor bug with function call that doesn't exist. ( #195 )
...
Co-authored-by: Aryan <aryan@huggingface.co >
2025-01-08 05:25:27 +05:30
38413aa167
Allow images; Remove LLM generated prefixes; Allow JSON/JSONL; Fix bugs ( #158 )
...
* update
* update
* fix
* Update finetrainers/dataset.py
* update
* argument for enabling remove of common llm prefixes
* update
* make new generator for validation
* update
* update
* update
* update
* update
2025-01-06 18:52:54 +05:30
b3d0ba8e12
support model cards ( #176 )
...
* introduce model cards and support passing FPS.
* updates
* fix
* fix
* updates
* revert changes
* revert args.py changes
* fix
* fix
* fix filename
2025-01-05 19:21:17 +05:30
a709a1cec1
fixes ( #178 )
2025-01-04 08:57:41 +05:30
071c140879
Fix scheduler bugs ( #177 )
...
* fix scheduler bugs.
* fix
2025-01-03 18:52:40 +05:30
b8352abf70
Support CogVideoX T2V ( #165 )
...
* support cog t2v.
* generator.
* updates
* style
* fixes
* fix padding frames.
Co-authored-by: zRzRzRzRzRzRzR <Yuxuan.Zhang2104@student.xjtlu.edu.cn >
* revert changes related to generator.
* refactor a lot of things.
* accept revision and cache_dir.
* remove unused var
* refactoring fixes.
* refactor
* update
* update
---------
Co-authored-by: zRzRzRzRzRzRzR <Yuxuan.Zhang2104@student.xjtlu.edu.cn >
Co-authored-by: Aryan <aryan@huggingface.co >
2025-01-03 10:53:16 +05:30
018953ed86
bulk update styling and formatting ( #170 )
2025-01-02 13:20:33 +05:30
191ea8b0cc
[optimization] support 8bit optims from bistandbytes ( #163 )
...
* support 8bit optims from bnb.
* fix exist_ok
* use_8bit_bnb.
* fix.
* note in readme.
2024-12-30 14:03:56 +05:30
84c17561eb
fixes
2024-12-27 15:02:21 +05:30
4461cdd98f
tracker logging
2024-12-27 12:25:11 +05:30
a19401e84f
update README; remove print; make style
2024-12-27 07:19:16 +01:00
87af1a0113
Merge branch 'main' into support-deepspeed
2024-12-26 08:17:48 +05:30
1d5e0566fb
private
2024-12-26 08:01:16 +05:30
0309f3bf67
Apply suggestions from code review
...
Co-authored-by: Aryan <aryan@huggingface.co >
2024-12-26 08:00:04 +05:30
0c0a0a5e4a
scheduler
2024-12-26 07:55:07 +05:30
28ecfe6b65
torch.is_tensor check
2024-12-26 07:49:51 +05:30
2a594ea2db
fix validation bug ( #149 )
2024-12-26 02:54:08 +05:30
5757b1b886
fix grad norm related logging
2024-12-25 19:25:22 +01:00
6b54867c3c
fixes
2024-12-25 21:34:56 +05:30
47e4aaac2f
track grad norm for deepspeed
2024-12-25 15:04:16 +01:00
e2178deea0
fix deadlock; make style
2024-12-25 14:16:31 +01:00
d7aa50917f
fixes
2024-12-25 12:07:50 +05:30
69b585a992
fixes
2024-12-25 11:35:09 +05:30
875cb72f63
address review feedback.
2024-12-24 21:20:37 +05:30
6329e9ed07
support DeepSpeed.
2024-12-24 10:59:21 +05:30
2858346b27
address review
2024-12-23 15:11:03 +01:00
78712f60a5
update comment
2024-12-23 14:01:39 +01:00
5261bed284
remove pipeline hooks
2024-12-23 13:58:37 +01:00
86e1d1888a
other condtions -> text conditions
2024-12-23 13:48:58 +01:00
363841ae3f
address some reviews; change hunyuan checkpoint repo-id
2024-12-23 12:54:52 +01:00
b045037850
make style
2024-12-21 15:19:03 +01:00
0db7bf28aa
update
2024-12-21 15:18:46 +01:00
2da8cb5798
update
2024-12-21 12:24:33 +01:00
66189c804b
update
2024-12-20 10:03:39 +01:00
223add1a59
Hunyuan Video LoRA ( #126 )
...
* add hunyuan-video lora support
* minor fixes; make style
* update readme
* update
* update
* Update README.md
* Update README.md
Co-authored-by: Sayak Paul <spsayakpaul@gmail.com >
* update
* update
* change move train scripts to internal directory
* update
---------
Co-authored-by: Sayak Paul <spsayakpaul@gmail.com >
2024-12-20 08:02:08 +05:30
9ef58e2f3a
LTX Video ( #123 )
...
* rename files
* ltx finetuning
* update
* update
* improvements
* make style
* gradient clipping
* update
* fix distributed inference
* update
* update
2024-12-19 04:27:36 +05:30
CrossProduct