Repository navigation
Conversation
HDCharles
requested review from
AndreasKaratzas,
mgoin,
pavanimajety,
robertgshaw2-redhat,
tlrmchlsmth,
yewentao256 and
zyongye
as code owners
August 18, 2026 16:30
![claude[bot]](https://avatars.githubusercontent.com/in/1236702?s=60&v=4){kind=small}
Contributor
|
This pull request has merge conflicts that must be resolved before it can be |
CompressedTensorsWNA16MoEMethod previously chose weight buffer layout (N-first vs K-first) based on which backend the oracle selected, leaking backend-specific concerns into the integration layer. This required an ever-growing exclusion list and caused unnecessary transpose round-trips for some backends (e.g. Triton: load K-first → transpose back to N-first). Now CT MoE always loads weights in the checkpoint's native N-first format [E, N, K_packed]. Backend-specific transposition (to K-first for Marlin, CPU, XPU, emulation) is handled in convert_to_wna16_moe_kernel_format where it naturally belongs. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
HDCharles
force-pushed
the
yiliu/ct-humming-w2a16-v2
branch
from
September 16, 2026 18:57
2798577 to
a6bd447
Compare
Contributor
|
This pull request has merge conflicts that must be resolved before it can be |
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
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Summary
CompressedTensorsWNA16MoEMethodto support arbitrary integer bit widths (2–8) for Humming backend, usingmath.ceilpacked dimension calculation instead of32 // num_bits_humming_wna16_weight_schemaforQuantizationArgs), quant config (get_humming_moe_quant_config), and kernel selection_is_supported_wna16_weight_keyto allow Humming experts to accept WNA16QuantKeytypes withScalarTypedtypes_dtype_abbrhelper to fixQuantKey.__str__/ScaleDesc.__str__crash onScalarTypedtypesDepends on #52798 — must be merged first (this branch is stacked on it).
Test plan
test_humming_supports_compressed_tensors_wna16_quant_key— verifies Humming accepts WNA16 keys for bit widths 2–8test_quant_key_str_supports_scalar_type_dtypes— verifiesQuantKey.__str__works withScalarTypetest_compressed_tensors_wna16_moe_create_weights_uses_ceil_packed_shapes— verifies ceil-based packed shapes for 3/5/6/7-bit (requires humming + CUDA)test_compressed_tensors_wna16_moe_converts_and_sets_up_humming_kernel— end-to-end weight conversion and kernel setup (requires humming + CUDA)🤖 Generated with Claude Code