docs: add Papers with Code badges for LLaVA-OneVision

[lorenzozanee]

Summary

• Add Papers with Code live leaderboard badges for LLaVA-OneVision (NLVR2 and CLIP-CC-Bench) to README as suggested in Verify evals on Papers with Code #1483.
• Fixes Verify evals on Papers with Code #1483.

In scope

• Add two PwC badges (eval=7144 NLVR2, eval=28575 CLIP-CC-Bench) alongside existing shields in README.md.

Out of scope

• No model, task or evaluation code changes.
• No verification or correction of individual scores (maintainers can edit via PwC page as noted in Verify evals on Papers with Code #1483).

Validation

• git diff --check | sample size: N=1 file | key metrics: no whitespace errors | result: pass
• git diff --stat | sample size: N=1 file | key metrics: README.md | 2 ++ | result: pass
• Manual README preview: badges render as live PwC SVGs and link to paperswithcode.co/paper/2408.03326

Risk / Compatibility

• Documentation-only change, no runtime or API impact.
• External badge SVGs are live from Papers with Code; no local asset change.

Type of Change

• Bug fix (non-breaking change)
• New feature
• New benchmark/task
• New model integration
• Breaking change
• Documentation update
• Refactoring (no functional changes)

[pufanyi]

Thank you for taking the time to put this together. These badges look helpful for showcasing the LLaVA-OneVision results.

Since the rankings are specific to LLaVA-OneVision rather than lmms-eval itself, we feel they would be a better fit for the official LLaVA-OneVision repository. We’d also like to keep the top-level lmms-eval README focused on the evaluation framework as a whole, since it supports many different model families.

With that in mind, we’ll pass on adding these badges here, but we really appreciate the contribution and the suggestion.