model: LFM2-VL fixes#17577
Merged
Merged
Conversation
tdakhran
commented
Nov 28, 2025
ngxson
reviewed
Nov 28, 2025
ngxson
reviewed
Nov 28, 2025
ngxson
left a comment
Collaborator
There was a problem hiding this comment.
I think we probably need to update ggml_backend_*_supports_op across backend, to avoid some backend fallback to the non-antialias kernel, which will result in wrong results.
For backend that does not support this mode, it will be fallback to CPU
Contributor
|
On my end (CPU) the outputs of fp32 and bf16 450M looked good, tested on a variety of small images (< 16/32px one side). Also checked a few personal tuned 1.6B Q3s (which should be more sensitive) and the outputs were great - it didn't go into a repetitive "breaking" state like before! It couldn't have been easy to figure this issue out... Thank you guys for looking into this! |
50ba22e to
2385ecf
Compare
Contributor
Author
|
@SmartestWashingMachine, great to see it work for you! @ngxson, I rolled back changes related to default marker placement, for now Added a fallback to CPU for backends as well. |
ngxson
reviewed
Nov 30, 2025
ngxson
left a comment
Collaborator
There was a problem hiding this comment.
The mtmd changes look good to me. I'll also need a review on backend code (cpu/cuda) before merging this. CC @CISC @JohannesGaessler if you have a bit of time.
|
|
||
| clip_image_u8 resized_img; | ||
| img_tool::resize(*img, resized_img, target_size, img_tool::RESIZE_ALGO_BILINEAR, true, pad_color); | ||
| const bool pad = (ctx->proj_type() != PROJECTOR_TYPE_LFM2); |
Collaborator
There was a problem hiding this comment.
self-note: probably kimi-vl also need padding, to be verified later
JohannesGaessler
approved these changes
Nov 30, 2025
JohannesGaessler
left a comment
Contributor
There was a problem hiding this comment.
The CUDA changes look good to me.
CISC
approved these changes
Nov 30, 2025
Anico2
added a commit
to Anico2/llama.cpp
that referenced
this pull request
Jan 15, 2026
* Adjust to pytorch * Add antialiasing upscale * Increase number of patches to 1024 * Handle default marker insertion for LFM2 * Switch to flag * Reformat * Cuda implementation of antialias kernel * Change placement in ops.cpp * consistent float literals * Pad only for LFM2 * Address PR feedback * Rollback default marker placement changes * Fallback to CPU implementation for antialias implementation of upscale
blime4
referenced
this pull request
in blime4/llama.cpp
Feb 5, 2026
* Adjust to pytorch * Add antialiasing upscale * Increase number of patches to 1024 * Handle default marker insertion for LFM2 * Switch to flag * Reformat * Cuda implementation of antialias kernel * Change placement in ops.cpp * consistent float literals * Pad only for LFM2 * Address PR feedback * Rollback default marker placement changes * Fallback to CPU implementation for antialias implementation of upscale
Seunghhon
pushed a commit
to Seunghhon/llama.cpp
that referenced
this pull request
Apr 26, 2026
* Adjust to pytorch * Add antialiasing upscale * Increase number of patches to 1024 * Handle default marker insertion for LFM2 * Switch to flag * Reformat * Cuda implementation of antialias kernel * Change placement in ops.cpp * consistent float literals * Pad only for LFM2 * Address PR feedback * Rollback default marker placement changes * Fallback to CPU implementation for antialias implementation of upscale
ljubomirj
pushed a commit
to ljubomirj/llama.cpp
that referenced
this pull request
May 6, 2026
* Adjust to pytorch * Add antialiasing upscale * Increase number of patches to 1024 * Handle default marker insertion for LFM2 * Switch to flag * Reformat * Cuda implementation of antialias kernel * Change placement in ops.cpp * consistent float literals * Pad only for LFM2 * Address PR feedback * Rollback default marker placement changes * Fallback to CPU implementation for antialias implementation of upscale
my-other-github-account
pushed a commit
to my-other-github-account/llama.cpp
that referenced
this pull request
May 15, 2026
* Adjust to pytorch * Add antialiasing upscale * Increase number of patches to 1024 * Handle default marker insertion for LFM2 * Switch to flag * Reformat * Cuda implementation of antialias kernel * Change placement in ops.cpp * consistent float literals * Pad only for LFM2 * Address PR feedback * Rollback default marker placement changes * Fallback to CPU implementation for antialias implementation of upscale
my-other-github-account
pushed a commit
to my-other-github-account/llama.cpp
that referenced
this pull request
May 15, 2026
* Adjust to pytorch * Add antialiasing upscale * Increase number of patches to 1024 * Handle default marker insertion for LFM2 * Switch to flag * Reformat * Cuda implementation of antialias kernel * Change placement in ops.cpp * consistent float literals * Pad only for LFM2 * Address PR feedback * Rollback default marker placement changes * Fallback to CPU implementation for antialias implementation of upscale
phibya
pushed a commit
to ziee-ai/llama.cpp
that referenced
this pull request
May 29, 2026
* Adjust to pytorch * Add antialiasing upscale * Increase number of patches to 1024 * Handle default marker insertion for LFM2 * Switch to flag * Reformat * Cuda implementation of antialias kernel * Change placement in ops.cpp * consistent float literals * Pad only for LFM2 * Address PR feedback * Rollback default marker placement changes * Fallback to CPU implementation for antialias implementation of upscale
fewtarius
pushed a commit
to fewtarius/llama.cpp
that referenced
this pull request
May 30, 2026
* Adjust to pytorch * Add antialiasing upscale * Increase number of patches to 1024 * Handle default marker insertion for LFM2 * Switch to flag * Reformat * Cuda implementation of antialias kernel * Change placement in ops.cpp * consistent float literals * Pad only for LFM2 * Address PR feedback * Rollback default marker placement changes * Fallback to CPU implementation for antialias implementation of upscale
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
5 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.
Debugging of #17290 revealed multiple issues with LFM2-VL.
This PR fixes the following issues and makes the output of
llama.cppequivalent to PyTorchround_by_factorto calculate targetwidthandheightin "smart resize".The central issue was the resizing of positional embeddings. Siglip2 implementation in PyTorch uses
F.interpolate(..., mode="bilinear", align_corners=False, antialias=True). Antialiasing only contributes during downscaling. When the image width or height is less than 256, the scaling of positional embeddings inllama.cppproduced numerically different results from PyTorch.A new flag, ' GGML_SCALE_FLAG_ANTIALIAS', has been added for the upscale function, with implementations for CPU and CUDA.
Now outputs match:
PyTorch (fp32)
this PR (
bin/llama-mtmd-cli -m $CKPT/LFM2-VL-1.6B-F32.gguf --mmproj $CKPT/mmproj-LFM2-VL-1.6B-F32.gguf -n 64 -t 4 --image /data/playground/issue_17290/siglip_1024.png -p "OCR." --temp 0.0 --top-k 1)