Fixed RoPE scaling extension factors buffer size for partial_rotary_factor - #337
Open
harrywhoo wants to merge 6 commits into
Open
Fixed RoPE scaling extension factors buffer size for partial_rotary_factor#337harrywhoo wants to merge 6 commits into
harrywhoo wants to merge 6 commits into
Conversation
MLC local ci setup. Also CI for Windows and macOS building, which may take 90-100 mins. Co-authored-by: Siyuan Feng <Hzfengsy@sjtu.edu.cn>
- Revert "[CMake][MSVC] Disable permissive mode for MSVC builds (#16343)" - Skip MSC tests - Disable NNPack and TFLite - Tweak CMAKE_CUDA_ARCHITECTURES
This PR updates the NVSHMEM-based NDArray allocation, which was missed in the recent FFI refactor and thus fails to compile when building tvm with NVSHMEM.
…tor models - Change ext_factors buffer size from head_dim//2 to rotary_dim//2 - Fixes runtime error for models like Phi-4 with partial_rotary_factor < 1.0 - Resolves TVMError: Assert fail for fused_rope_longrope_scaling_ext_factors_handle
MasterJH5574
force-pushed
the
mlc
branch
3 times, most recently
from
August 9, 2025 20:50
7b965fe to
b5736b3
Compare
MasterJH5574
force-pushed
the
mlc
branch
4 times, most recently
from
August 27, 2025 14:42
7d8375b to
0a9fe8d
Compare
MasterJH5574
force-pushed
the
mlc
branch
5 times, most recently
from
August 31, 2025 14:06
d2a03b2 to
c3e156f
Compare
MasterJH5574
force-pushed
the
mlc
branch
7 times, most recently
from
September 14, 2025 03:49
424aaa4 to
10e31c7
Compare
MasterJH5574
force-pushed
the
mlc
branch
2 times, most recently
from
September 28, 2025 03:53
07529be to
e16f551
Compare
MasterJH5574
force-pushed
the
mlc
branch
2 times, most recently
from
December 9, 2025 01:27
851d153 to
da3714d
Compare
MasterJH5574
force-pushed
the
mlc
branch
3 times, most recently
from
February 9, 2026 17:27
90ab5d3 to
b3d4fe9
Compare
MasterJH5574
force-pushed
the
mlc
branch
4 times, most recently
from
March 3, 2026 15:45
2ef52fe to
bfd7787
Compare
MasterJH5574
force-pushed
the
mlc
branch
3 times, most recently
from
April 2, 2026 04:03
153ed4e to
2bf925c
Compare
MasterJH5574
force-pushed
the
mlc
branch
2 times, most recently
from
June 30, 2026 13:30
f5369ec to
06db68d
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Phi-4 was causing runtime errors in TVM due to its partial_rotary_factor of 0.75, which resulted in a mismatch between the expected ext_factor dimensions and the actual. Changed ext_factors buffer size from head_dim//2 to rotary_dim//2 to align with RoPE implementation. This resolves TVMError: Assert fail for fused_rope_longrope_scaling_ext_factors_handle. Tested other models to ensure this didn't break support.