pytorch torchtune issues

pytorch / torchtune

PyTorch native finetuning library

https://pytorch.org/torchtune/main/

BSD 3-Clause "New" or "Revised" License

4.35k stars 440 forks source link

issues

Newest

Newest Most commented Recently updated Oldest Least commented Least recently updated

Clean up PyTorch profiler usage in recipes + configs

#2053 joecummings opened 1 hour ago
0
improve _optim_ckpt_wrapper so it is a drop in replacement of optimizer

#2052 felipemello1 opened 2 hours ago
1
Do not require lr_scheduler for all other recipes

#2051 thomasjpfan opened 2 hours ago
2
Recipe documentation

#2050 joecummings opened 3 hours ago
1
[export] Add exportable attention and kv cache

#2049 larryliu0820 opened 17 hours ago
1
`FullModelHFCheckpointer` saved checkpoint isn't compatible with Huggingface `transformers` model loading

#2048 vancoykendall opened 20 hours ago
4
Remove deprecated `TiedEmbeddingTransformerDecoder`

#2047 EmilyIsCoding closed 18 hours ago
1
Use hf transfer as default

#2046 felipemello1 closed 5 hours ago
3
Change llama 3 and 2 configs to use hf ckpt instead of meta ckpt

#2045 felipemello1 opened 1 day ago
0
PPO slow start

#2044 felipemello1 opened 1 day ago
0
Fix issue in loading mixed precision vocab pruned models during torchtune generation for evaluation

#2043 ifed-ucsd opened 1 day ago
4
remove default to ignore safetensors

#2042 felipemello1 closed 19 hours ago
3
Remove deprecated `TiedEmbeddingTransformerDecoder`

#2041 SalmanMohammadi closed 18 hours ago
0
Vector Quantized Embeddings

#2040 RdoubleA opened 1 day ago
1
[BUG] Linter is not capturing issues, like unused arguments

#2039 felipemello1 opened 1 day ago
0
[BUG] decoder_lora and fusion_lora is a no-op in llama 3.2 vision

#2038 felipemello1 opened 1 day ago
1
Apply gradient accumulation fix to DPO/PPO recipes

#2037 SalmanMohammadi opened 2 days ago
0
update tests and readme cu121 --> cu124

#2036 felipemello1 opened 2 days ago
1
Add RPO, DPOP losses, add lambda_dpop to basic DPO loss

#2035 krammnic opened 2 days ago
3
Added new datasets

#2034 shahin-trunk closed 2 days ago
3
adding inception datasets

#2033 shahin-trunk closed 2 days ago
2
Add NLL and DPOP weighting to DPO losses

#2032 RdoubleA opened 2 days ago
0
Error: Are you sure that 'metric_logging' is importable from module 'torchtune.utils'?

#2031 zhoufy20 opened 2 days ago
3
log rank zero everywhere

#2030 RdoubleA closed 2 days ago
1
Add config for distilling Llama3.1 70B into 1B

#2029 joecummings opened 2 days ago
0
Fix Qlora/lora for 3.2 vision

#2028 felipemello1 closed 1 day ago
2
[WIP] compile graph breaks

#2027 felipemello1 opened 3 days ago
1
[Test] Save local model path in PEFT adapter config

#2026 ebsmothers opened 3 days ago
1
Integrate Lora fine-tuned model with HF

#2025 GLorenzo679 opened 3 days ago
5
Hown to config generation length on Android device

#2024 HaomingSong opened 3 days ago
1
Support for unsharded parameters in state_dict APIs

#2023 RdoubleA closed 2 days ago
1
[WIP] Reducing eval vision tests runtime

#2022 SalmanMohammadi closed 2 days ago
2
Allow recipes to work without a LR scheduler

#2021 felipemello1 opened 4 days ago
0
Some KD recipe cleanup

#2020 ebsmothers closed 4 days ago
1
Dataset packing does not work in knowledge distillation recipe

#2019 joecummings closed 4 days ago
0
Does torchtune support multi-node training?

#2018 tginart opened 4 days ago
3
Add LR Scheduler to full finetune distributed

#2017 parthsarthi03 closed 2 days ago
5
Remove unused FSDP components

#2016 ebsmothers closed 5 days ago
5
torchrun defaults for concurrent distributed training jobs

#2015 ebsmothers closed 6 days ago
4
Add comments explaining why 128011 is skipped for Llama3 Tokenizer

#2014 RdoubleA opened 6 days ago
0
gemma2 had wrong path to scheduler

#2013 felipemello1 closed 6 days ago
2
Add custom dataset section in Basics in live docs

#2012 RdoubleA opened 6 days ago
0
Create _export directory in torchtune

#2011 Jack-Khuu closed 6 days ago
9
Warmup starts from lr=0

#2010 cfhammill opened 1 week ago
9
Bump version to 0.5.0

#2009 joecummings closed 1 week ago
1
v0.5.0 tracker

#2008 joecummings opened 1 week ago
0
Allow input labels to be logits for knowledge distilation

#2007 felipemello1 opened 1 week ago
0
[DCP][RFC] Faster intermediate checkpoints with DCP async save in TorchTune

#2006 saumishr opened 1 week ago
9
Can we get XPU Nightlies?

#2005 Jack-Khuu opened 1 week ago
2
Fine tuning on custom dataset with custom prompt template

#2004 GLorenzo679 opened 1 week ago
9