issues
search
pytorch
/
torchtitan
A native PyTorch Library for large model training
BSD 3-Clause "New" or "Revised" License
1.26k
stars
115
forks
source link
issues
Newest
Newest
Most commented
Recently updated
Oldest
Least commented
Least recently updated
Update contributing.md
#385
H-Huang
closed
2 weeks ago
0
[torchtitan] Fix test runner fused optim tests
#384
wz337
closed
3 weeks ago
0
Make metrics logging work for pipeline parallelism
#383
wconstab
closed
3 weeks ago
1
[POC] Showed more memory efficient FSDP wrapping
#382
awgu
opened
3 weeks ago
0
Use general way to access and update submodules
#381
kwen2501
closed
3 weeks ago
0
Fix start/stop layer parsing
#380
wconstab
closed
3 weeks ago
0
Add PP tracer + DP test
#379
kwen2501
opened
4 weeks ago
1
Fix start/stop layer parsing
#378
wconstab
closed
4 weeks ago
1
Make seed checkpoint creation work on CPU
#377
wconstab
closed
3 weeks ago
0
IBM experimental dataloaders
#376
daviswer
opened
1 month ago
4
about reference of weight init according to layer depth or layer id
#375
SeunghyunSEO
closed
2 weeks ago
3
add compiled RMSNorm into the norm config
#374
tianyu-l
opened
1 month ago
0
merge depulicate integration tests into one
#373
tianyu-l
closed
1 month ago
0
replace old torch dependency in requirements.txt
#372
tianyu-l
closed
1 month ago
1
Use general way to access and update submodules
#371
kwen2501
closed
1 month ago
0
Lint
#370
kwen2501
closed
1 month ago
0
Add --test option to specify test to run
#369
kwen2501
closed
1 month ago
1
Add --test option to specify test to run
#368
kwen2501
closed
1 month ago
0
test changes
#367
wconstab
closed
1 month ago
0
keep only latest k checkpoints
#366
liangluofb
closed
1 month ago
0
expose optimizer params, log optimizer type and settings for the run
#365
lessw2020
opened
1 month ago
0
enable TritonFusedRMSNorm with local_map annotation
#364
XilunWu
closed
2 weeks ago
3
[do NOT land][experiment] use local_map to annotate TritonFusedRMSNorm
#363
XilunWu
closed
1 month ago
0
Fix 1D PP tracer test
#362
wconstab
closed
2 weeks ago
3
Remove PP+TP rmsnorm workaround
#361
wconstab
closed
2 weeks ago
1
[RFC] Allow ModelWrapper and OptimizerWrapper to accept multiple models
#360
fegin
closed
3 weeks ago
0
update .gitignore to screen out slew of new temp files
#359
lessw2020
closed
1 month ago
0
Support looped PP schedules in torchtitan
#358
wconstab
closed
1 week ago
0
Add test for PP tracer frontend
#357
wconstab
closed
1 month ago
0
Update pipelining import after change on pytorch
#356
wconstab
closed
1 month ago
0
[torchtitan][optim] Add fused as an option in train config
#355
wz337
closed
3 weeks ago
7
Fix bug in PP output layer shape
#354
wconstab
closed
1 month ago
0
fix periodic integration test and add helper message on torchdata import failure
#353
tianyu-l
closed
1 month ago
0
Make test_runner use separate logger with default INFO
#352
wconstab
closed
1 month ago
0
Add torchdata to requirements after release
#351
gokulavasan
opened
1 month ago
0
Fix llama_13b.toml -> llama2_13b.toml in multinode_trainer.slurm
#350
pbelevich
closed
1 month ago
0
Rmsnorm cuda
#349
lessw2020
closed
1 month ago
0
Expose mixed_precision dtype arguments
#348
wconstab
closed
1 month ago
0
Code change that changes the model semantics
#347
kwen2501
closed
1 month ago
3
use local_map to annotate fusedrmsnorm
#346
wanchaol
closed
2 weeks ago
0
Add a 3-stage PP config
#345
wconstab
closed
2 weeks ago
0
Add 3D support
#344
wconstab
closed
3 weeks ago
3
Make test_runner.py warn on non-empty output dir
#343
wconstab
closed
1 month ago
0
Make pip install torch quiet
#342
wconstab
closed
1 month ago
0
Modify FLOPs in MFU calculation for casual mask when using FlashAttention.
#341
Yuxin-CV
closed
1 month ago
1
Fixes for manual stage shape and pp_degree=3, WIP
#340
wconstab
closed
1 month ago
0
only produce tensorboard logs on rank 0 by default
#339
tianyu-l
closed
1 month ago
1
Add a workflow to build torchtitan-ubuntu-20.04-clang12 Docker image for CI
#338
huydhn
closed
1 month ago
4
Test 1f1b schedule
#337
wconstab
closed
1 month ago
0
checkpoint.model_weights_only Doesn't makes any difference
#336
TJ-Solergibert
closed
1 month ago
1
Previous
Next