MMNeedle
Multimodal Needle in a Haystack
We introduce the MultiModal Needle-in-a-haystack (MMNeedle) benchmark, specifically designed to assess the long-context capabilities of MLLMs.
How to run
$ npm i -g lettokenburn-cli$ ltb bench info mmneedle
External harness. No Harbor adapter has been generated for this benchmark yet. `ltb bench info` prints the known links.
Runs happen on your machine with your own subscription; only the score summary is pushed. CLI docs
Scores
8 sourced entries| Harness | Source | Confidence | Link | |||
|---|---|---|---|---|---|---|
| GPT-4o | 83.0 | N/A | Independent | medium | Mar 15, 2023 | arxiv.org |
| GPT-4V | 54.7 | N/A | Independent | medium | Mar 15, 2023 | arxiv.org |
| Gemini Pro 1.5 | 39.9 | N/A | Independent | medium | Mar 8, 2024 | arxiv.org |
| Gemini Pro 1.0 | 24.8 | N/A | Independent | medium | Dec 19, 2023 | arxiv.org |
| LLaVA-Llama-3 | 17.5 | N/A | Independent | medium | Mar 18, 2024 | arxiv.org |
| Claude 3 Opus | 12.3 | N/A | Independent | medium | Mar 4, 2024 | anthropic.com |
| CogVLM2-Llama-3 | 0.9 | N/A | Independent | medium | Nov 6, 2023 | arxiv.org |
| CogVLM-17B | 0.1 | N/A | Independent | medium | Nov 6, 2023 | arxiv.org |
Every number links to where it was published. Low-confidence rows are muted and need per-cell verification.
Reproduced by the community
All runsAbout
We introduce the MultiModal Needle-in-a-haystack (MMNeedle) benchmark, specifically designed to assess the long-context capabilities of MLLMs. Besides multi-image input, we employ image stitching to further increase the input context length, and develop a protocol to automatically generate labels for sub-image level retrieval. Essentially, MMNeedle evaluates MLLMs by stress-testing their capability to locate a target sub-image (needle) within a set of images (haystack) based on textual instructions and descriptions of image contents. This setup necessitates an advanced understanding of extensive visual contexts and effective information retrieval within long-context image inputs.