microsoft/Tutel

View on GitHub

Tutel MoE: Optimized Mixture-of-Experts Library, Support GptOss/DeepSeek/Kimi-K2/Qwen3 using FP8/NVFP4/MXFP4

Stars: 1k
Forks: 111
Open beginner issues: 0
Indexed issues: 55
Avg merge: 26m
Merged PRs (30d): 5

Dominant language: C
License: MIT
Last GitHub push: Sep 15, 2026
Latest indexed: Sep 19, 2026
Contributing guide: No contributing guide
Code of conduct: Code of conduct
Beginner labels: No beginner labels indexed

Beginner friendly Everything

Open All

55 open issues indexed

OutOfMemoryError Open

Difficulty 4/5 3-5 days Newbie friendliness 35/100

microsoft/Tutel#329 · 10 comments · Opened Jul 2, 2026
BUG: system.init_data_model_parallel() Prevents Nsight Systems (nsys) from Tracing GPU Hardware in Distributed Mode Open

Difficulty 4/5 3-5 days Newbie friendliness 35/100

microsoft/Tutel#315 · 1 comment · Opened Nov 16, 2025
Tutel as an MoE backend in Nanotron for Qwen3-MoE 15B (128 experts, top-k=8) Open

Difficulty 5/5 Over a week Newbie friendliness 20/100

microsoft/Tutel#310 · 3 comments · Opened Sep 17, 2025
How to install tutel on Huawei Ascend Open

Difficulty 5/5 Over a week Newbie friendliness 20/100

microsoft/Tutel#306 · 1 comment · Opened Sep 2, 2025
How to set dp MoE Open

Difficulty 4/5 3-5 days Newbie friendliness 25/100

microsoft/Tutel#304 · 1 comment · Opened Aug 24, 2025
Login Open Web UI Open

Difficulty 5/5 Over a week Newbie friendliness 20/100

microsoft/Tutel#301 · 1 comment · Opened Aug 8, 2025
Support for Blackwell? Open

Difficulty 4/5 3-5 days Newbie friendliness 35/100

microsoft/Tutel#298 · 20 comments · Opened Jul 28, 2025
does it support Qwen3-235B-A22B-Instruct-2507-FP4 Open

Difficulty 4/5 3-5 days Newbie friendliness 25/100

microsoft/Tutel#297 · 8 comments · Opened Jul 24, 2025
[Question] How can I set the max tokens with tutelgroup/deepseek-671b Open

Difficulty 3/5 1-2 days Newbie friendliness 20/100

microsoft/Tutel#294 · 4 comments · Opened Jul 15, 2025
About adapting Tutel in fairseq Open

Difficulty 4/5 3-5 days Newbie friendliness 25/100

microsoft/Tutel#293 · 11 comments · Opened Jul 15, 2025
where can I find the source code for specific mod file? Open

Difficulty 2/5 1-3 hours Newbie friendliness 25/100

microsoft/Tutel#290 · 3 comments · Opened Jul 5, 2025
Examples integrated with Megatron-LM Open

Difficulty 5/5 Over a week Newbie friendliness 25/100

microsoft/Tutel#288 · 1 comment · Opened Jun 20, 2025
[Question] How does the gate coordinate across ranks in expert parallelism? Open

Difficulty 4/5 3-5 days Newbie friendliness 25/100

microsoft/Tutel#278 · 3 comments · Opened Apr 27, 2025
RuntimeError: (true) == (fp != nullptr) INTERNAL ASSERT FAILED at "/home/xxx/tutel/tutel/custom/custom_kernel.cpp":33, please report a bug to PyTorch. CHECK_EQ fails. Open

Difficulty 4/5 3-5 days Newbie friendliness 25/100

microsoft/Tutel#271 · 4 comments · 1 reaction · Opened Mar 27, 2025
custom_kernel.o: No such file or directory Open

Difficulty 3/5 1-2 days Newbie friendliness 25/100

microsoft/Tutel#263 · 1 comment · Opened Mar 4, 2025
How to define a custom gate function in moe_layer Open

Difficulty 3/5 1-2 days Newbie friendliness 20/100

microsoft/Tutel#256 · 1 comment · Opened Jan 17, 2025
Inference code for 2 models evaluated in the paper Open

Difficulty 5/5 Over a week Newbie friendliness 25/100

microsoft/Tutel#255 · 3 comments · Opened Dec 3, 2024
How expert parameters are distributed in the cluster when using the Tutel framework? Open

Difficulty 4/5 3-5 days Newbie friendliness 20/100

microsoft/Tutel#251 · 1 comment · Opened Oct 30, 2024
How to load 32-experts Swin-transformer-moe on a 2-GPU machine. Open

Difficulty 4/5 3-5 days Newbie friendliness 30/100

microsoft/Tutel#248 · 8 comments · Opened Oct 27, 2024
How to convert checkpoint files that adapt to different distributed world sizes Open

Difficulty 4/5 3-5 days Newbie friendliness 25/100

microsoft/Tutel#246 · 1 comment · 1 reaction · Opened Aug 27, 2024
Question regarding the load importance loss calculation Open

Difficulty 4/5 3-5 days Newbie friendliness 25/100

microsoft/Tutel#240 · 1 comment · Opened Jun 15, 2024
How about the cost of TUTEL features? Open

Difficulty 5/5 Over a week Newbie friendliness 20/100

microsoft/Tutel#239 · 1 comment · Opened Jun 6, 2024
How to use Megablocks in MoE training Open

Difficulty 4/5 3-5 days Newbie friendliness 25/100

microsoft/Tutel#236 · 1 comment · Opened Jun 5, 2024
[Question] Comparison to FasterMoE Open

Difficulty 5/5 Over a week Newbie friendliness 15/100

microsoft/Tutel#232 · 4 comments · Opened Apr 17, 2024
Qs Open

Difficulty 4/5 3-5 days Newbie friendliness 25/100

microsoft/Tutel#230 · 3 comments · Opened Apr 16, 2024
How to create a custom expert with tutel? Open

Difficulty 3/5 1-2 days Newbie friendliness 35/100

microsoft/Tutel#226 · 19 comments · Opened Mar 22, 2024
[Question] Why use datatype ncclInt8 in nccl_all_to_all_scatter_async. Open

Difficulty 4/5 3-5 days Newbie friendliness 20/100

microsoft/Tutel#220 · 1 comment · Opened Dec 18, 2023
How to implement Fairseq-MoE training checkpoint like Swin-MoE? Open

Difficulty 5/5 Over a week Newbie friendliness 20/100

microsoft/Tutel#219 · 1 comment · Opened Nov 10, 2023
Non-surface function utilities only work for contiguous input data Open

Difficulty 4/5 3-5 days Newbie friendliness 35/100

microsoft/Tutel#218 · 12 comments · Opened Nov 6, 2023
tutel is slower than the naive p2p using 2DH for small scale Open

Difficulty 4/5 3-5 days Newbie friendliness 25/100

microsoft/Tutel#214 · 3 comments · Opened Aug 27, 2023
how to use tutel on Megatron Deepspeed Open

Difficulty 4/5 3-5 days Newbie friendliness 25/100

microsoft/Tutel#207 · 4 comments · Opened Jul 15, 2023
Can this package support the one-gpu machine Open

Difficulty 4/5 3-5 days Newbie friendliness 25/100

microsoft/Tutel#206 · 5 comments · Opened May 31, 2023
Training with Data and Expert Parallelism Open

Difficulty 4/5 3-5 days Newbie friendliness 25/100

microsoft/Tutel#204 · 11 comments · Opened May 15, 2023
INTERNAL ASSERT FAILED Open

Difficulty 4/5 3-5 days Newbie friendliness 18/100

microsoft/Tutel#203 · 5 comments · Opened May 2, 2023
about compute_location and locations Open

Difficulty 2/5 1-3 hours Newbie friendliness 30/100

microsoft/Tutel#201 · 1 comment · Opened Apr 10, 2023
ImportError: cannot import name 'tutel_custom_kernel' from 'tutel.impls.jit_compiler' Open

environmental issue

Difficulty 4/5 3-5 days Newbie friendliness 20/100

microsoft/Tutel#198 · 16 comments · Opened Mar 30, 2023
tutel/jit_kernels/sparse.py torch.float16 There is a bug in the calculation: the cuda calculation result is inconsistent with the CPU calculation result and the array is out of bounds Open

invalid

Difficulty 4/5 3-5 days Newbie friendliness 35/100

microsoft/Tutel#196 · 1 comment · Opened Mar 8, 2023
All2All precision always in fp32 Open

Difficulty 4/5 3-5 days Newbie friendliness 35/100

microsoft/Tutel#195 · 1 comment · 1 reaction · Opened Feb 21, 2023
How the experts' gradients are handled under data parallelism? Open

Difficulty 4/5 3-5 days Newbie friendliness 30/100

microsoft/Tutel#192 · 1 comment · Opened Dec 26, 2022
RuntimeError: No such operator tutel_ops::cumsum Open

Difficulty 4/5 3-5 days Newbie friendliness 25/100

microsoft/Tutel#190 · 10 comments · 1 reaction · Opened Nov 14, 2022
[installation errors] fatal error: nccl.h: No such file or directory Open

Difficulty 3/5 1-2 days Newbie friendliness 35/100

microsoft/Tutel#189 · 6 comments · Opened Oct 19, 2022
Multi-nodes training is much more slower than single node Open

Difficulty 4/5 3-5 days Newbie friendliness 25/100

microsoft/Tutel#187 · 1 comment · Opened Sep 29, 2022
NCCL Asynchronous update timeout crash with Tutel MoE Open

microsoft/Tutel#185 · 5 comments · 1 assignee · Opened Sep 6, 2022
My code seems to hang when skip_remainder_batch=False. Open

application patch

Difficulty 5/5 Over a week Newbie friendliness 20/100

microsoft/Tutel#182 · 7 comments · Opened Aug 9, 2022
Cannot import JIT optimized kernels. Did you forget to install Custom Kernel Extension? Open

environmental issue

Difficulty 4/5 3-5 days Newbie friendliness 25/100

microsoft/Tutel#180 · 1 comment · Opened Aug 7, 2022
Pretrained MoE model Open

question

Difficulty 5/5 Over a week Newbie friendliness 25/100

microsoft/Tutel#179 · 2 comments · Opened Aug 7, 2022
Example on saving experts to one model when using distributed training Open

duplicate

Difficulty 4/5 3-5 days Newbie friendliness 35/100

microsoft/Tutel#178 · 2 comments · Opened Aug 7, 2022
Error when doing deepcopy of the model Open

enhancement

Difficulty 4/5 3-5 days Newbie friendliness 25/100

microsoft/Tutel#177 · 5 comments · Opened Aug 3, 2022
how can I install this pack on conda environment?? Open

setup

microsoft/Tutel#169 · 11 comments · 1 assignee · Opened Jul 11, 2022
Error in load_importance_loss Open

enhancement

Difficulty 4/5 3-5 days Newbie friendliness 20/100

microsoft/Tutel#167 · 7 comments · Opened Jul 6, 2022

microsoft/Tutel

Get new issues in your inbox