Fixed/Improved Jinja Chat Template for Qwen 3.8: Reasoning & Tool Calls
Chromix_ · reddit · 2026-08-15
A deep dive into fixing and improving the Jinja chat template for Qwen 3.8. The author compared the original template with a recent modification, finding that while the original had issues with invalid reasoning effort and missing historical fields, the previous modification introduced deviations from the training format and new bugs. The new template maintains fidelity to the original training prompt format while fixing logic, such as aliasing high to xhigh, correctly handling JSON tool-call mode, and preventing Boolean argument regression. A detailed table compares the behavior of all three templates across various edge cases.
Related event: Developers Fix Qwen 3.8 Chat Template Bugs for Reasoning and Tool Calls(2 posts)→
More from coding & agent
- Terminal Night 3 talk slides released, missing demo video to follow — 4310sy · 2026-10-16
- Agent developers are sitting on unused API credits that can exceed actual usage — EnvironmentalDay8864 · 2026-10-16
- Notion's bet on agents: abstract away deployment so users and agents just write code — simonguozirui · 2026-10-16
- Four Gemini 4 Argon agents beat Minecraft's Ender Dragon autonomously in 27 hours — Maxence Faldor · 2026-10-16
- Tencent's Terminal-RSI builds recursively self-improving terminal agents, hitting 19.84% on Terminal-Bench 4.0 — tencent · 2026-10-16
- Tsinghua's AutoSFT uses multi-agent collaboration to automate SFT data research — TsinghuaNLP · 2026-10-16