Repository navigation
llama-quant : fix 4D tensor quantization using ggml_nrows() and name-based expert detection - #30820
Merged
Merged
Conversation
CISC
approved these changes
Oct 19, 2026
anthony-wss
force-pushed
the
quantize-fix-4d-tensors
branch
from
October 21, 2026 08:54
e8efc1d to
1eebadc
Compare
nrows_total was ne[1]*ne[2], so for a 4D tensor only the first ne[2] slices were quantized and the rest of the rows were dropped. Use ggml_nrows(tensor) to count all rows, including ne[3]. Fixes ggml-org#29916
…4D tensors The MXFP4_MOE path used ne[2] > 1 to detect MoE expert tensors. This is wrong for non-expert tensors with 3 or more dims, e.g. MLA projections (BailingMoE3) and 4D conv kernels. Use the tensor name instead: all tensors used by GGML_OP_MUL_MAT_ID are named *_exps or *_chexps. This also removes the BailingMoE3 special case. The imatrix holds one vector of ne[0] values per 2D slice, so the expected size is ne[0]*ne[2]*ne[3], not ne[0]*ne[2]. This matches how the imatrix is indexed per row for 4D tensors in llama_tensor_quantize_impl. Assisted-by: Claude Opus 5.5
anthony-wss
force-pushed
the
quantize-fix-4d-tensors
branch
from
October 25, 2026 01:27
1eebadc to
e9e884e
Compare
CISC
added
the
merge ready
3 of 6 tasks
auRose94
pushed a commit
to auRose94/llama.cpp
that referenced
this pull request
Oct 28, 2026
…based expert detection (ggml-org#30820) * llama-quant : fix quantizing 4D tensors nrows_total was ne[1]*ne[2], so for a 4D tensor only the first ne[2] slices were quantized and the rest of the rows were dropped. Use ggml_nrows(tensor) to count all rows, including ne[3]. Fixes ggml-org#29916 * llama-quant : detect expert tensors by name and fix imatrix size for 4D tensors The MXFP4_MOE path used ne[2] > 1 to detect MoE expert tensors. This is wrong for non-expert tensors with 3 or more dims, e.g. MLA projections (BailingMoE3) and 4D conv kernels. Use the tensor name instead: all tensors used by GGML_OP_MUL_MAT_ID are named *_exps or *_chexps. This also removes the BailingMoE3 special case. The imatrix holds one vector of ne[0] values per 2D slice, so the expected size is ne[0]*ne[2]*ne[3], not ne[0]*ne[2]. This matches how the imatrix is indexed per row for 4D tensors in llama_tensor_quantize_impl. Assisted-by: Claude Opus 5.5
edwardyoon
pushed a commit
to edwardyoon/focus-llama
that referenced
this pull request
Oct 30, 2026
…based expert detection (ggml-org#30820) * llama-quant : fix quantizing 4D tensors nrows_total was ne[1]*ne[2], so for a 4D tensor only the first ne[2] slices were quantized and the rest of the rows were dropped. Use ggml_nrows(tensor) to count all rows, including ne[3]. Fixes ggml-org#29916 * llama-quant : detect expert tensors by name and fix imatrix size for 4D tensors The MXFP4_MOE path used ne[2] > 1 to detect MoE expert tensors. This is wrong for non-expert tensors with 3 or more dims, e.g. MLA projections (BailingMoE3) and 4D conv kernels. Use the tensor name instead: all tensors used by GGML_OP_MUL_MAT_ID are named *_exps or *_chexps. This also removes the BailingMoE3 special case. The imatrix holds one vector of ne[0] values per 2D slice, so the expected size is ne[0]*ne[2]*ne[3], not ne[0]*ne[2]. This matches how the imatrix is indexed per row for 4D tensors in llama_tensor_quantize_impl. Assisted-by: Claude Opus 5.5
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
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.
Overview
Fix quantization of 4D tensors by using
ggml_nrows()to count rows, and detect expert tensors by name instead of by shape.Additional information
Fixes #29916 in 64d1bbb. The row count was computed from
ne[1]andne[2]only, ignoringne[3]. For tensors likea.conv1d.0.weight [3, 3, 1, 128]this produces the wrong size and trips thegguf_get_tensor_size(...) == new_sizeassert.ggml_nrows()is now used instead. Other places insrc/llama-quant.cppthat ignoredne[3]are fixed as well.Before fix (reproduced.log)
After fix (after-fix.log)
On top of that, we previously used
tensor->ne[2] > 1to detect expert tensors, but that did not cover all cases and required some hard-coded exclusions for the bailingmoe3 model. Expert tensors are now identified by_exps.or_chexps.in their name. This new implementation does not break existing models, and it even caught an edge case where the attention weights for deepseek-v3.1 were wrongly treated as expert tensors when they should be quantized asQ8_0instead. The test schema has also been updated.Requirements
Log files