eval(qwen): add Qwen 3.8 quality benchmarks and native 14px pilot - #244
Merged
Conversation
info.imageCount already tracked the static slab, and the slabs are inserted into req.messages before applyChatHistoryCollapse runs, so adding both double-counted them. With 14 slab images the budget went 28 > 24 and remainingMax clamped to 0, disabling history collapse for Qwen-sized contexts. Walk the request messages only, matching the Responses path.
GPT-5 keeps detail:'original'; every other upstream (Workers AI, GPT-4o, ...) validates against auto|low|high and 400s on 'original'. The bridge already has req.model, so derive the detail from it instead of hard-coding.
The measured 14px/84-col/512px geometry is specific to Qwen 3.8 27B. Other Qwen variants now fall to the family-id guard (refused) instead of being gated with an unmeasured profile.
The local gateway is private tooling; the repo now references it only as 'a local gateway' with neutral example ports, and eval clients require OPENAI_BASE_URL instead of assuming a default.
The profile's 24 stood in for the total, so client-attached images starved the history budget (14 attached + 14 slab = 28 > 24 left zero headroom and history collapse never fired). Add providerImageCap to the profile and budget history as min(history cap, provider cap - images already in the request), so the final request can never overshoot the provider's hard limit while still using whatever headroom is left.
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
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Problem
Evaluated Qwen 3.8 27B (
workers-ai/@cf/qwen/qwen3.8-27b) across pxpipe's quality benchmarks via ocproxy Chat Completions.Findings
max_tokens>= 8,192 for multi-image prompts.Changes
eval/qwen-profile/suite and receipts (novel-arithmetic,gist-recall,verbatim-hex,verbatim-hex-14px).eval/qwen-profile/QUALITY_RESULTS.mdand updatedREADME.mdquality matrix.