archive/README-v12.md

metadata

license: apache-2.0
tags:
  - jinja
  - chat-template
  - qwen
  - qwen3.5
  - qwen3.6
  - lm-studio
  - mlx
  - llama.cpp
  - tool-calling
  - thinking

Fixed jinja chat templates for Qwen 3.5 & 3.6 (v12)

2026-05-10 Update (v12): Fixed agent stalls, parameter data-loss, and hallucination bugs! Restored dynamic tool instructions and the <IMPORTANT> formatting reminder block to stop grammar parser crashes. Fixed a massive semantic data-loss bug in v10 by restoring parameter descriptions to the compacted tool schemas. Implemented a robust Pre-Parse Tag Normalization Pipeline to neutralize whitespace tag hallucinations (e.g., </ think>), preventing conversational text from being permanently trapped or deleted.

2026-05-10 Update (v11): Fixed agent looping and overthinking! Re-implemented preserve_thinking kwarg to properly strip reasoning blocks from history by default, and restored the reasoning bypass (<think>\n\n</think>\n\n) when enable_thinking is false to prevent the model from hallucinating reasoning tags. Also removed the redundant end-of-prompt system message.

2026-05-09 Update (v10): Massive stability and performance overhaul! Halved tool prompt tokens via compaction, eliminated C++ engine UndefinedValue crashes by removing brittle string slices, fixed prefix cache invalidation in llama.cpp by ensuring chronological state tracking, resolved vLLM crashes with <|think_off|>, and fixed tool parameter omission over long contexts via dynamic instruction reinforcement.

2026-05-08 Update (v9): Fixed 9th bug: Thinking-tool-call hallucination. Refactored system prompt parsing to enable dynamic tool instructions. The template now actively teaches the model how to safely combine <think> blocks and <tool_call> boundaries.

2026-05-07 Update (v8): Fixed 8th bug: Mid-conversation system messages no longer crash the template. Compatibility restored for agent frameworks (OpenCode, Docker Agent, oh-my-pi). Re-engineered Jinja string parsing for C++ engine stability.

These are drop-in Jinja templates that fix rendering errors, token waste, and missing features in the official Qwen chat templates.

They are tested to work across LM Studio, llama.cpp, vLLM, MLX, oMLX, and any engine that supports HuggingFace Jinja templates.

---

Why you need this

The official Qwen templates contain restrictions and Python-specific Jinja logic that break usage on many inference engines and agent frameworks.

Here are the 14 bugs this template fixes:

Problem	Impact	Fix
1. Tool calls fail on C++ engines	The |items filter doesn't exist in minijinja (LM Studio, llama.cpp, MLX). Tool calls instantly crash the template.	Rewritten for strict C++ engine compatibility.
2. Mid-conversation system crash	Frameworks injecting mid-conversation steering instructions trigger a hard crash.	Native, chronological rendering for system messages anywhere.
3. developer role rejected	Modern APIs send the developer role; the official template rejects it.	Added full support for "developer".
4. Empty thinking blocks spam	Every past turn gets wrapped in empty <think></think> tags, wasting context and breaking caching.	Strictly skips empty blocks unconditionally.
5. No way to toggle thinking	The user is restricted to the model defaults.	Intercepts <|think_off|> and <|think_on|> tags natively.
6. Whitespace tag hallucinations	Model hallucinates invalid boundaries (e.g., </ think>), swallowing conversational text.	Pre-Parse Tag Normalization Pipeline natively neutralizes variants.
7. No-user-query crash	raise_exception crashes agentic loops, system-only contexts, or /reset flows.	Removed backwards history scanning entirely.
8. Unclosed thinking before tool call	Model calls a tool without closing its reasoning, bleeding XML tags into tool parsers.	Auto-injects closing tags before tool boundaries.
9. Thinking tool_call hallucination	Model places <tool_call> inside <think> block because prompt forces <think>\n before a strict tool instruction.	Hoists system toggle to inject <think> natively into tool instructions.
10. Massive token waste on tools	The template dumps raw JSON schemas for tools, wasting ~50% of the prompt context and increasing TTFT.	Tools are now compacted into typed one-line signatures while preserving semantic parameter descriptions.
11. Cache invalidation on llama.cpp	Conditionally hiding past <think> blocks based on future messages breaks KV cache.	Refactored history rendering to strictly chronological forward state tracking.
12. Reasoning bypass hallucinations	When thinking is disabled, Qwen models inherently hallucinate reasoning tags anyway.	Injects an empty closed <think>\n\n</think>\n\n block to successfully force reasoning bypass.
13. Jinja C++ crashes (UndefinedValue)	Python negative indexing [-1] when closing tags fails on minijinja engines.	Replaced all string slicing with safe Jinja replace() operations.
14. Tool format drift & stalls	Model leaks conversational text between reasoning and tool calls, stalling strict parsers.	Re-engineered dynamic instructions and restored <IMPORTANT> formatting reminder to the initial system prompt.

---

Quick install

Choose your environment and update the template:

LM Studio

1. Open your Qwen model in the right-side panel.
2. Scroll down to Prompt Template.
3. Replace the template with the contents of qwen3.5/chat_template.jinja or qwen3.6/chat_template.jinja.
4. Click Save.

llama.cpp / koboldcpp

--jinja --chat-template-file qwen3.6/chat_template.jinja

vLLM / TextGen

Replace the "chat_template" string in your tokenizer_config.json with the raw file contents.

oMLX

Overwrite chat_template.jinja in your local model directory. Load with --jinja. Remove any chat_template_kwargs overrides because the template handles everything internally.

---

Which file do I use?

Template File	Supported Models
qwen3.5/chat_template.jinja	Qwen3.5-35B-A3B, Qwen3.5-32B, Qwen3.5-14B, and all Qwen 3.5 variants.
qwen3.6/chat_template.jinja	Qwen3.6-27B, Qwen3.6-35B-A3B, and all Qwen 3.6 variants.

Note: The 3.6 template is a superset. It additionally handles preserve_thinking, </thinking> hallucination recovery, and interrupted thought streams. If you are on 3.6, always use the 3.6 file.

---

The thinking toggle

You can control the model reasoning behavior. Insert <|think_on|> or <|think_off|> anywhere in your system or user prompt.

The template natively intercepts the tag, removes it from the final context so the model never sees it, and flips the reasoning mode instantly.

Fast answer, no reasoning:

System: You are a coding assistant. <|think_off|>
User: What's 2+2?

Deep reasoning:

System: You are a coding assistant. <|think_on|>
User: Implement a red-black tree in Rust.

(The tag syntax uses Qwen's control-token delimiters to guarantee it will never collide with legitimate text or file paths, unlike earlier community templates that used /think)

---

Preserving past thoughts

By default, Qwen models "forget" their previous <think> blocks in the chat history to prevent repetitive looping and save context tokens. If you are running an agentic workflow where the model needs to reference its past reasoning, you can enable the preserve_thinking flag in your engine's template kwargs:

{
  "preserve_thinking": true
}

(If your engine does not support passing kwargs, the template will default to standard Qwen behavior and strip past thoughts).

---

Pre-installed models

If you are using one of the following models, you already have an older version of this template installed.

• froggeric/Qwen3.6-27B-MLX-8bit
• froggeric/Qwen3.6-27B-MLX-4bit
• froggeric/Qwen3.5-35B-A3B-Uncensored-FernflowerAI-MLX-8bit
• froggeric/Qwen3.5-35B-A3B-Uncensored-FernflowerAI-MLX-4bit
• froggeric/Qwen3.6-27B-Uncensored-Heretic-v2-MLX-8bit
• froggeric/Qwen3.6-27B-Uncensored-Heretic-v2-MLX-6bit
• froggeric/Qwen3.6-27B-Uncensored-Heretic-v2-MLX-4bit
• froggeric/Qwen3.6-35B-A3B-Uncensored-Heretic-MLX-8bit
• froggeric/Qwen3.6-35B-A3B-Uncensored-Heretic-MLX-6bit
• froggeric/Qwen3.6-35B-A3B-Uncensored-Heretic-MLX-4bit

---

Technical Details of the 9 Fixes

1. Tool calls on C++ engines

The official template iterates tool call arguments with |items: {%- for key, value in tool_call.arguments|items %}

Python's Jinja supports |items. C++ runtimes (LM Studio, llama.cpp, MLX) do not, which produces a rendering error. This template uses direct dictionary key lookups instead. It also replaces is sequence with is iterable, removes Python-only |safe wrappers, and handles arguments returned as raw strings.

2. Mid-conversation system messages crash

The official template hard-crashes if a system or developer message appears anywhere except the first position. This breaks agentic frameworks (Codex CLI, Docker Agent, oh-my-pi, OpenCode) that inject steering instructions mid-conversation. The fix natively renders these messages chronologically to preserve LLM recency bias while enforcing strict image-blocking checks.

3. developer role

The OpenAI-compatible API spec sends message.role == "developer" for system-level instructions. The official Qwen template throws an exception. Both templates here accept "developer" and map it properly.

4. Empty thinking blocks

The official template wraps every past assistant turn in thinking tags, even when empty. When there is no reasoning content, those tags waste context tokens and break prefix caching. The 3.5 template checks reasoning_content before emitting. The 3.6 template checks reasoning_content|trim|length > 0 and ties history visibility to the <|think_off|> override.

5. </thinking> hallucination (Qwen 3.6 only)

The Qwen 3.6 model sometimes generates </thinking> instead of the expected </think>. The official parser splits on </think > only and fails. The 3.6 template detects which closing tag was actually used and splits dynamically. It also handles interrupted generation by rescuing incomplete streams.

6. Arguments serialization

The official template serializes argument values with |tojson unconditionally, failing when the value is already a string. The fixed templates check the type first. Strings pass through as-is, and everything else goes through |tojson.

7. Auto-close unclosed thinking before tool calls

The model sometimes starts a thinking block and immediately calls a tool without emitting the closing tag. The official template lets the unclosed thinking tag bleed into the tool call. The fixed templates detect this pattern and safely auto-inject the closing tag using standard Jinja split operations to guarantee 100% C++ compatibility.

8. No-user-query exception

The official template scans the message list in reverse. If all messages are tool results, or there are no user messages, it fires raise_exception('No user query found...') and hard-crashes. The fix replaces the exception with a graceful fallback {%- set ns.last_query_index = messages|length - 1 %}, enabling agentic tool-calling chains to function perfectly.

9. Thinking tool_call hallucination

The official template appends <think>\n to the end of the generation prompt to initiate reasoning. However, its system instructions rigidly demand the model to output only <tool_call> with no suffix. This contradictory state causes the model to improperly nest its tool call inside the thinking block. This template utilizes a global pre-scan to evaluate the final enable_thinking state across the entire conversation history, guaranteeing it can dynamically inject a proper <think>...</think> usage example into the tool instructions exactly when reasoning is enabled.

Comparison: Qwen 3.5 templates

Feature	Official	LuffyTheFox	mod-ellary	Pneuny	This (v12)
Tool arguments	Fails	Fixed	Missing	Fixed	Fixed
|safe removed	Fails	Fixed	Missing	Fixed	Fixed
developer role	Missing	Missing	Missing	Missing	Added
Thinking toggle	None	None	/think (system only)	None	<|think_off|> anywhere
Empty think in history	Broken	Broken	Tags omitted	Broken	Pruned dynamically
Mid-conversation system	Crashes	Crashes	Crashes	Crashes	Fixed
No-user-query crash	Crashes	Crashes	Crashes	Crashes	Graceful fallback
Auto-close thinking	Not handled	Not handled	Not handled	Not handled	Engine-safe auto-inject
Tool token optimization	None	None	None	None	~50% reduction
Long-context tool adherence	Fails	Fails	Fails	Fails	Dynamic reinforcement

Comparison: Qwen 3.6 template

Feature	Official	This (v12)
Tool arguments	Fails (|items)	Fixed
|safe removed	Fails	Fixed
developer role	Missing	Added
Thinking toggle	None	<|think_off|> anywhere
Empty think in history	Spams empty blocks	Pruned dynamically
KV prefix caching	Breaks on dynamic history	100% stable (Chronological)
Mid-conversation system	Crashes	Fixed
</thinking> hallucination	Fails	Detected and handled
Auto-close thinking before tool	Not handled	Engine-safe auto-inject
vLLM stop parsing	Crashes if thinking disabled	Fixed natively
Tool token optimization	None	~50% reduction
Long-context tool adherence	Fails	Dynamic reinforcement

---

Authorship

Role	Author
Original models	Alibaba Cloud (Qwen team)
Template fixes	froggeric

License

Apache-2.0, inherited from Qwen.