Vision preprocessing token limits and latency fixes
NousResearch/hermes-agent · cluster 10809 · agent
Data as of
Nothing open on GitHub. Fix merged in #75253. Check it covers your case.
The short answer
- Known?
- Yes. 10 related items grouped here, 3 issues already closed.
- Fix event
- A fix was merged to main in PR #75253. Check it covers your case. No release data is tracked.
- Tracker
- The maintainers' tracker marks this
partially resolved, deciding PR #75253. That is separate from the fix event. - Still open
- Nothing open on GitHub.
These issues and PRs address overly verbose vision preprocessing output, hardcoded max_tokens limits in auxiliary vision calls, and related latency problems. Work spans making max_tokens configurable via config.yaml, keeping preprocessing prompts concise, and routing improvements for multimodal models.
Open issues0
Open PRs0
Fixes on record
-
#75253 fix(vision): stop capping aux vision output with hardcoded max_tokens
files:
tests/tools/test_browser_console.py,tests/tools/test_video_analyze.py,tests/tools/test_vision_tools.py,tools/browser_tool.py,tools/vision_tools.py -
#81269 fix(gateway): keep auto vision preprocess concise (salvage #10852)
files:
contributors/emails/[redacted:email],gateway/run.py,tests/gateway/test_vision_preprocess.py
5 PRs closed without merging
From the close comments: done elsewhere 3 · duplicate 1 · merged via another change 1
-
#10852 fix(gateway): keep auto vision preprocess concise · merged via another change →
#81269,#10809,#75253 -
#28999 fix(gateway): retry auto-vision on transient failure (#28972) · done elsewhere →
fb403a3,#56889 - #29600 fix(vision): honor auxiliary.vision.max_tokens from config in vision_tools · done elsewhere
-
#38004 fix: respect auxiliary.vision.max_tokens from config.yaml (#29590) · done elsewhere →
db472f5 -
#61187 Cap gateway vision preprocessing output · duplicate →
#10852,#10809
Closed issues · 3
- Closed#10809Default vision pre-process prompt generates overly long descriptions (~2000 chars), significantly slowing down image-bearing requests on local models
- Closed#29590[Bug/Enhancement] Hardcoded max_tokens and overly verbose prompt in vision_tools.py causes severe latency for reasoning models
- Closed#34087Make max_tokens configurable for aux vision calls
How this cluster was grouped
-
Link evidence (
partly structural links): some members are connected by structural evidence (shared changed files, or a closing/superseded reference); the rest are joined by an automated relevance judgement over their text. - Members outside that structural core may belong to a different underlying problem.
-
Pairwise link density
0.17: the share of item pairs that carry a link, counting a structural link in full and a relevance-only link at 0.36.1.00means every pair is linked.
Pick up a Lab task
Check whether a fix covers the open issues here, with your own agent. Bring back evidence for maintainers.
Open the Lab