Add complete vision benchmark support - #2869
Open
ckuethe wants to merge 1 commit into
Open
Conversation
Contributor
Author
|
The vision benchmarks are in their own file to make review easier. GFX1152GFX1151GFX1100 RX7900 |
Contributor
Author
|
review/feedback can begin sooner than that. there's not much point in my rebasing to a pr that hasn't been merged or approved, especially if it changes parts my pr relies on. |
Add ability to benchmark vision (multimodal) models by embedding images into chat messages during text generation benchmarks. Vision scenarios specify an image_path and messages; the CLI loads and base64-encodes the image, then injects it into user message content arrays using the OpenAI multimodal format before sending to the server. Only allow vision-capable models to run vision scenarios Only load image files from the scenario file directory for security Source images for the tests taken from public domain sources: - https://en.wikipedia.org/wiki/United_States_Declaration_of_Independence - https://www.nro.gov/foia-home/ - https://ntrs.nasa.gov/ - https://www.cia.gov/readingroom/ Lots of challenging formatting, interesting fonts, and poor image quality. The scan of a fax of a photocopy of a microfiche of a mimeograph image path can really challenge the transcription models. Co-authored-by: opencode, North-Mini-Code (512k) Co-authored-by: opencode, Qwen3.6-35B-A3B (256k)
Contributor
Author
|
Rebased |
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
2 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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
feat(bench): add vision benchmark support for image-capable models
Add ability to benchmark vision (multimodal) models by embedding images
into chat messages during text generation benchmarks. Vision scenarios
specify an image_path and messages; the CLI loads and base64-encodes the
image, then injects it into user message content arrays using the OpenAI
multimodal format before sending to the server.
Only allow vision-capable models to run vision scenarios
Only load image files from the scenario file directory for security
Source images for the tests taken from public domain sources:
Lots of challenging formatting, interesting fonts, and poor image quality.
The scan of a fax of a photocopy of a microfiche of a mimeograph image path
can really challenge the transcription models.
Co-authored-by: opencode, North-Mini-Code (512k)
Co-authored-by: opencode, Qwen3.6-35B-A3B (256k)