-
Notifications
You must be signed in to change notification settings - Fork 399
Pull requests: pytorch/torchtitan
Author
Label
Projects
Milestones
Reviews
Assignee
Sort
Pull requests list
[WIP] [float8] add float auto_filter_for_recipe
CLA Signed
This label is managed by the Meta Open Source bot.
#1319
opened Jun 18, 2025 by
danielvegamyhre
•
Draft
[not for land] testing out float8 128_1_128_128 blockwise scaling
CLA Signed
This label is managed by the Meta Open Source bot.
#1317
opened Jun 18, 2025 by
vkuzo
Loading…
Do not submit: Multinode training seems to be working
CLA Signed
This label is managed by the Meta Open Source bot.
#1314
opened Jun 17, 2025 by
ahmadsharif1
•
Draft
Do not submit: Multinode is working with multiple controllers
CLA Signed
This label is managed by the Meta Open Source bot.
#1313
opened Jun 17, 2025 by
ahmadsharif1
•
Draft
Add check for This label is managed by the Meta Open Source bot.
seq_len%tensor_parallel_degree==0
for parallelized Llama
CLA Signed
#1312
opened Jun 17, 2025 by
jc-audet
Loading…
[llama4][auxiliary-loss-free load balancing] update expert_bias without backward hooks
CLA Signed
This label is managed by the Meta Open Source bot.
#1304
opened Jun 16, 2025 by
hann-wang
Loading…
Finetune from pre-trained models
CLA Signed
This label is managed by the Meta Open Source bot.
#1300
opened Jun 15, 2025 by
vwxyzjn
Loading…
[not for land] Use new AC
CLA Signed
This label is managed by the Meta Open Source bot.
#1294
opened Jun 13, 2025 by
soulitzer
Loading…
WIP: Try to use monarch to run torchtitan.
CLA Signed
This label is managed by the Meta Open Source bot.
#1288
opened Jun 12, 2025 by
ahmadsharif1
•
Draft
[Do Not Merge] Test Titan changes to use DCP ZOC instead of titan default
CLA Signed
This label is managed by the Meta Open Source bot.
#1287
opened Jun 12, 2025 by
Saiteja64
Loading…
DO NOT SUBMIT: WIP: Try to use monarch to run torchtitan.
CLA Signed
This label is managed by the Meta Open Source bot.
#1286
opened Jun 12, 2025 by
ahmadsharif1
•
Draft
[deepseek][kernels][blackwell] Cutlass blackwell grouped gemm using cute dsl (forward)
CLA Signed
This label is managed by the Meta Open Source bot.
#1276
opened Jun 8, 2025 by
lessw2020
Loading…
[deepseek][blackwell] add Cutlass cute dsl blackwell dense based looping group gemm
CLA Signed
This label is managed by the Meta Open Source bot.
#1274
opened Jun 8, 2025 by
lessw2020
Loading…
[deepseek][blackwell] add manual looping group gemm to enable base working inference on Blackwell
CLA Signed
This label is managed by the Meta Open Source bot.
#1272
opened Jun 7, 2025 by
lessw2020
Loading…
[llama4] enable expert parallel on the same device mesh as tp (tp2ep)
CLA Signed
This label is managed by the Meta Open Source bot.
#1269
opened Jun 6, 2025 by
hann-wang
Loading…
Add support for creating ROCm docker image for torchtitan & enable ROCm CI support.
CLA Signed
This label is managed by the Meta Open Source bot.
module: rocm
#1260
opened Jun 4, 2025 by
akashveramd
•
Draft
[WIP][Blackwell Kernels] Blackwell group gemm and dense gemms with Python Cutlass
CLA Signed
This label is managed by the Meta Open Source bot.
#1256
opened Jun 3, 2025 by
lessw2020
Loading…
alternative implementation of create_indices_from_offsets_nosync compatible with torch.compile
CLA Signed
This label is managed by the Meta Open Source bot.
#1251
opened Jun 1, 2025 by
hann-wang
Loading…
[float8] add float8 rowwise MoE prototype
CLA Signed
This label is managed by the Meta Open Source bot.
#1245
opened May 30, 2025 by
danielvegamyhre
Loading…
[cp][flex_attention] integration test trial
CLA Signed
This label is managed by the Meta Open Source bot.
[Flux] Add batched inference
CLA Signed
This label is managed by the Meta Open Source bot.
#1227
opened May 27, 2025 by
CarlosGomes98
Loading…
[WIP] Implement the feature to save unsharded weights at the last step
CLA Signed
This label is managed by the Meta Open Source bot.
#1219
opened May 23, 2025 by
fegin
Loading…
[WIP][Experimental] Activation Offloading
CLA Signed
This label is managed by the Meta Open Source bot.
#1218
opened May 23, 2025 by
lessw2020
Loading…
Previous Next
ProTip!
Filter pull requests by the default branch with base:main.