feat: swap Tier 3 from GPT-OSS-20B to Qwen3.5-27B#573
Merged
Lightheartdevs merged 2 commits intomainfrom Mar 23, 2026
Merged
Conversation
GPT-OSS-20B uses special tokens (<|start|>, <|channel|>, <|constrain|>) for structured output that are incompatible with llama.cpp's JSON grammar mode. This causes Perplexica (which uses generateObject) to fail with HTTP 500 on every query. Pure chat inference worked fine but structured output / tool calling was broken. Qwen3.5-27B (16.7GB Q4_K_M) is the same model family as Tier 1-2 (Qwen 3.5), proven compatible with llama.cpp structured output, and fits in 20-39GB VRAM tier. Updated across all platforms, tests, agent templates, docs, and disk estimation. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Qwen3.5-27B-Q4_K_M.gguf not qwen3.5-27b-Q4_K_M.gguf — sed lowercased it during the bulk replace. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Lightheartdevs
added a commit
that referenced
this pull request
Mar 23, 2026
Follow-up to #573 — docs still referenced old Qwen3 8B/4B/14B models. Updated to match current tier map: - T1/T2/ARC: Qwen3.5 9B - T3: Qwen3.5 27B - ARC_LITE: Qwen3.5 4B Files: root README, FAQ, INTEL-ARC-GUIDE, MACOS-QUICKSTART, SUPPORT-MATRIX Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Lightheartdevs
added a commit
that referenced
this pull request
Mar 23, 2026
Follow-up to #573 — docs still referenced old Qwen3 8B/4B/14B models. Updated to match current tier map: - T1/T2/ARC: Qwen3.5 9B - T3: Qwen3.5 27B - ARC_LITE: Qwen3.5 4B Files: root README, FAQ, INTEL-ARC-GUIDE, MACOS-QUICKSTART, SUPPORT-MATRIX Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
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.
Summary
GPT-OSS-20B broke Perplexica and any service using structured output / JSON mode. The model uses special tokens (
<|start|>,<|channel|>,<|constrain|>) incompatible with llama.cpp's grammar-based JSON constraining. Pure chat worked great, structured output returned HTTP 500.Swapping to Qwen3.5-27B (16.7GB Q4_K_M) — same family as Tier 1-2, proven llama.cpp compatibility, fits 20-39GB VRAM.
Test plan
bash tests/test-tier-map.shpasses🤖 Generated with Claude Code