Potential numerical accuracy issue in the fused_moe implementation
Maintainers usually reply within 8 days
Nobody has claimed this yet.
Assessment
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Newbie friendliness
- 35/100
- Issue type
- Bug
- Clarity
- Mostly clear
- Activity status
- Stale
- Domain
- machine-learning, testing-qa
Research direction
Start with tests/ops/test_moe.py, reproduce the reported parameter set with torch.manual_seed(0), and compare bfloat16 against float16. Trace the fused_moe entry point used by that test and investigate the mismatched outputs for the larger dimensions. Done means the bfloat16 case matches the reference within the test tolerance without regressing the passing cases.
Written by the indexing model from the issue text.
Description
Test environment: CTK 13.1, torch = 2.9.1+cu130, cuda-tile = 1.0.0, single B200 GPU
When using a parameter set (num_tokens, hidden_size, moe_intermediate_size, n_experts, top_k) = (128, 4096, 2048, 16, 4) in tests/ops/test_moe.py and add torch.manual_seed(0) in line 101, will have mismatched results:
(Earlier output omitted)
> assert passed, f"\n{failed_msgs}"
^^^^^^
E AssertionError:
E *** OUTPUT 0 DID NOT MATCH THE REFERENCE (rtol=0.1, atol=0.1) ***
E allclose: False
E matched: 523466 / 524288 [99.84%]
E ref range: -1.1700e+02 : 1.2300e+02
E test range: -1.1650e+02 : 1.2300e+02
E |ref| range: 0.0000e+00 : 1.2300e+02
E |test| range: 0.0000e+00 : 1.2300e+02
E max absolute difference: 1.0000e+00
E max relative change: 4.3000e+01
E max max mean change: 2.0000e+00
E max arith mean change: 5.1400e+02
E shape: torch.Size([128, 4096]) stride: (4096, 1) dtype: torch.bfloat16
E mismatched indices:tensor([[ 0, 1842],
E [ 0, 2071],
E [ 1, 675],
E ...,
E [ 127, 1728],
E [ 127, 2207],
E [ 127, 2628]])
For the same parameter set, if dtype is changed to float16, the test can pass.
Using a smaller hidden_size or a smaller moe_intermediate_size can also pass the test.
- Dominant language
- Python
- Stars
- 830
- Forks
- 92
- Avg merge
- 8d 11h
- Merged PRs (30d)
- 8
Getting set up
- No Dockerfile or Docker Compose file
- Has a pull request template
- Read the contributing guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
More from NVIDIA/TileGym
-
Difficulty 5/5 Over a week Newbie friendliness 12/100
Maintainers usually reply within 8 days
-
Difficulty 5/5 Over a week Newbie friendliness 35/100
NVIDIA/TileGym#210 · 1 comment ·
Maintainers usually reply within 8 days
-
Difficulty 5/5 Over a week Newbie friendliness 35/100
NVIDIA/TileGym#195 · 2 comments ·
Maintainers usually reply within 8 days
-
Difficulty 5/5 Over a week Newbie friendliness 25/100
NVIDIA/TileGym#191 · 1 comment ·
Maintainers usually reply within 8 days
-
Difficulty 5/5 Over a week Newbie friendliness 35/100
NVIDIA/TileGym#190 · 2 comments ·
Maintainers usually reply within 8 days
Similar issues
-
ci-rerun: collect.py fails a whole pr-babysit cycle on one transient GitHub 5xx and never retriesOpenfollow-up
Difficulty 2/5 1-3 hours Newbie friendliness 62/100
Maintainers usually reply within 1 day
-
Difficulty 2/5 1-3 hours Newbie friendliness 70/100
-
Built-in QUBO fallback applies bitwise xor to floating binary statesPossibly taken @iloveeyjafjalla claimed this today. Open
Difficulty 2/5 1-3 hours Newbie friendliness 72/100
qboson/kaiwu-pytorch-plugin#1702 ·
Maintainers usually reply within 3 days
-
bug v2
Difficulty 2/5 1-3 hours Newbie friendliness 76/100
modelcontextprotocol/python-sdk#3811 ·
Maintainers usually reply within 1 day
-
cost:cheap practicality:seen-in-the-wild severity:medium
Difficulty 2/5 1-3 hours Newbie friendliness 72/100
fairagro/m4.2_advanced_middleware_api#620 ·
Maintainers usually reply within 1 day