-
Notifications
You must be signed in to change notification settings - Fork 749
Pull requests: pytorch/torchtitan
Author
Label
Projects
Milestones
Reviews
Assignee
Sort
Pull requests list
Fix Pyrefly type errors for Tensor.item() return type
ciflow/8gpu
CLA Signed
This label is managed by the Meta Open Source bot.
#2626
opened Mar 18, 2026 by
fegin
Loading…
[Draft WIP] MoE with LocalMap
ciflow/8gpu
CLA Signed
This label is managed by the Meta Open Source bot.
#2625
opened Mar 18, 2026 by
acisseJZhong
•
Draft
[DONT LAND] Full Dtensor fully_shard + Local Map + FlexAttention
ciflow/8gpu
CLA Signed
This label is managed by the Meta Open Source bot.
#2621
opened Mar 18, 2026 by
fegin
Loading…
Only apply grouped GEMM padding for MXFP8 and FP8 non-HybridEP cases
ciflow/8gpu
CLA Signed
This label is managed by the Meta Open Source bot.
#2620
opened Mar 18, 2026 by
danielvegamyhre
Loading…
[graph_trainer] Log transformed graph to tlparse via trace_structured
ciflow/8gpu
CLA Signed
This label is managed by the Meta Open Source bot.
#2619
opened Mar 18, 2026 by
yiming0416
Loading…
FlattenThis label is managed by the Meta Open Source bot.
rl directory: remove vllm_compat, consolidate unified
ciflow/8gpu
CLA Signed
#2618
opened Mar 17, 2026 by
wwwjn
Loading…
[WIP][Full Dtensor] Work with full dtensor fully_shard
ciflow/8gpu
CLA Signed
This label is managed by the Meta Open Source bot.
Make MoE models non-strict tracing friendly
ciflow/8gpu
CLA Signed
This label is managed by the Meta Open Source bot.
#2612
opened Mar 16, 2026 by
ydwu4
Loading…
2 tasks done
Actually correct training w/ packed datasets defaults
ciflow/8gpu
CLA Signed
This label is managed by the Meta Open Source bot.
#2610
opened Mar 16, 2026 by
joecummings
•
Draft
[Module][2/2] Convert remaining nn.Module classes to Module protocol --
ciflow/8gpu
CLA Signed
This label is managed by the Meta Open Source bot.
#2608
opened Mar 16, 2026 by
fegin
Loading…
[Observability 7/7] Timing spans and step tags across all trainers
ciflow/8gpu
CLA Signed
This label is managed by the Meta Open Source bot.
#2607
opened Mar 16, 2026 by
felipemello1
Loading…
7 tasks done
[Observability 6/7] Wire observability into Trainer, Flux, FT, and Forge
ciflow/8gpu
CLA Signed
This label is managed by the Meta Open Source bot.
#2606
opened Mar 16, 2026 by
felipemello1
Loading…
5 tasks done
[Observability 5/7] RolloutLogger for RL rollout JSONL with filtering
ciflow/8gpu
CLA Signed
This label is managed by the Meta Open Source bot.
#2605
opened Mar 16, 2026 by
felipemello1
Loading…
3 tasks done
[Observability 4/7] MetricsProcessor — throughput, memory, validation, log scheduling
ciflow/8gpu
CLA Signed
This label is managed by the Meta Open Source bot.
#2604
opened Mar 16, 2026 by
felipemello1
Loading…
4 tasks done
[Observability 3/7] Experiment metrics with background aggregation subprocess
ciflow/8gpu
CLA Signed
This label is managed by the Meta Open Source bot.
#2603
opened Mar 16, 2026 by
felipemello1
Loading…
4 tasks done
[Observability 2/7] Structured system logging and gantt chart generation
ciflow/8gpu
CLA Signed
This label is managed by the Meta Open Source bot.
#2602
opened Mar 16, 2026 by
felipemello1
Loading…
4 tasks done
[Observability 1/7] Toy trainers for observability development
ciflow/8gpu
CLA Signed
This label is managed by the Meta Open Source bot.
#2601
opened Mar 16, 2026 by
felipemello1
Loading…
No Weight Decay Keywords
CLA Signed
This label is managed by the Meta Open Source bot.
#2600
opened Mar 16, 2026 by
francesco-bertolotti
Loading…
add async activation offloading to CPU with pinned memory pool
CLA Signed
This label is managed by the Meta Open Source bot.
#2581
opened Mar 15, 2026 by
dean-mccoppin
Loading…
Add weight tying support for Llama3
CLA Signed
This label is managed by the Meta Open Source bot.
#2580
opened Mar 15, 2026 by
dean-mccoppin
Loading…
Implement torch compile and mxfp8 for flux
CLA Signed
This label is managed by the Meta Open Source bot.
#2579
opened Mar 15, 2026 by
hlahkar
Loading…
Configurable Z Loss
CLA Signed
This label is managed by the Meta Open Source bot.
#2576
opened Mar 14, 2026 by
francesco-bertolotti
Loading…
[RL] Changes to enable compilation for trainer
ciflow/8gpu
CLA Signed
This label is managed by the Meta Open Source bot.
#2568
opened Mar 13, 2026 by
Lucaskabela
•
Draft
[mxfp8 training] add TP warning
ciflow/8gpu
CLA Signed
This label is managed by the Meta Open Source bot.
#2562
opened Mar 12, 2026 by
danielvegamyhre
Loading…
Previous Next
ProTip!
Updated in the last three days: updated:>2026-03-15.