Bump Qwen-Fixed-Chat-Templates to v19 and adopt unified chat_template.jinja

Upstream restructured the repository: the qwen3.5/ and qwen3.6/ subdirectories are gone, replaced by a single chat_template.jinja at the root that handles both variants. v19 also flips preserve_thinking to default true, so the explicit chat-template-kwargs override is no longer needed and is dropped from all presets.

Author: countzero

CHANGELOG.md

@@ -7,6 +7,11 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+### Changed
+- [Vendor] Bump Qwen-Fixed-Chat-Templates submodule to v19
+- [Presets] Repoint chat-template-file to unified chat_template.jinja
+- [Presets] Drop chat-template-kwargs preserve_thinking override
+
 ### Removed
 - [Build] Drop webui-download.cmake npm-resolver patch (fixed upstream in #23064)
 

CLAUDE.md

@@ -44,20 +44,25 @@ See `presets/README.md` for the user-facing quick-start; notes below are for edi
   compute buffer OOMs but the server keeps running — only image requests error at generation
   time. Set `false` on tiers where LLM + KV already saturate VRAM.
 
-- **All Qwen 3.6 entries pin `chat-template-file = vendor\Qwen-Fixed-Chat-Templates\qwen3.6\chat_template-v13.jinja`.**
+- **All Qwen 3.6 entries pin `chat-template-file = vendor\Qwen-Fixed-Chat-Templates\chat_template.jinja`.**
   Required, *not* redundant with `jinja = true` — `chat-template-file` *replaces*
   the GGUF-embedded template entirely (`vendor/llama.cpp/common/arg.cpp:3142`,
   `params.chat_template = read_file(value)`). The upstream embedded template has
-  fourteen documented bugs affecting tool calls, role handling, `<think>` block
-  rendering, agentic loops, and llama.cpp KV-prefix cache stability; the vendored
-  template fixes all of them (full list in `vendor/Qwen-Fixed-Chat-Templates/README-v13.md`).
-  The template also adds a `<|think_on|>` / `<|think_off|>` toggle that coexists
-  with `chat-template-kwargs = {"preserve_thinking":true}` (default strips past
-  `<think>` blocks from history; set the kwarg to keep them for agentic reasoning).
-  Path is repo-relative, so `llama-server` must be launched from the repo root —
-  `read_file()` resolves against the process CWD, not the INI file's directory.
-  Gemma and `Qwen3-Coder-Next` entries deliberately keep their GGUF-embedded
-  templates; upstream's README only claims compatibility for Qwen 3.5 / 3.6 variants.
+  documented issues with tool calls, role handling, `<think>` block rendering,
+  agentic loops, and llama.cpp KV-prefix cache stability; the vendored template
+  fixes all of them (full list in `vendor/Qwen-Fixed-Chat-Templates/README.md`).
+  Since v19 the template is a single unified file covering both Qwen 3.5 and 3.6
+  variants (the old `qwen3.5/` and `qwen3.6/` subdirectories now live under
+  `archive/`). The template adds a `<|think_on|>` / `<|think_off|>` toggle, and
+  v19 defaults `preserve_thinking` to `true` (past `<think>` blocks are kept
+  chronologically for 100% KV prefix cache stability and agentic reasoning
+  continuity). To strip past `<think>` blocks instead, set
+  `chat-template-kwargs = {"preserve_thinking":false}` — at the cost of a lower
+  KV cache hit rate. Path is repo-relative, so `llama-server` must be launched
+  from the repo root — `read_file()` resolves against the process CWD, not the
+  INI file's directory. Gemma and `Qwen3-Coder-Next` entries deliberately keep
+  their GGUF-embedded templates; upstream's README only claims compatibility for
+  Qwen 3.5 / 3.6 variants.
 
 **ngram-mod speculative decoding** (`--spec-type ngram-mod`): model-agnostic, works on any model.
 - All models: `spec-ngram-mod-n-match = 24`, `spec-ngram-mod-n-min = 48`, `spec-ngram-mod-n-max = 64`

presets/README.md

@@ -39,8 +39,9 @@ The section header (e.g. `[gemma-4-31B-it.IQ4_XS.gguf]`) is the model name clien
 > See `llama-server --help` for all flags.
 
 > [!IMPORTANT]
-> All `Qwen3.6-*` entries set `chat-template-file = vendor\Qwen-Fixed-Chat-Templates\qwen3.6\chat_template-v13.jinja`,
-> overriding the buggy template embedded in the GGUF. The path is repo-relative,
+> All `Qwen3.6-*` entries set `chat-template-file = vendor\Qwen-Fixed-Chat-Templates\chat_template.jinja`,
+> overriding the buggy template embedded in the GGUF. The vendored template is a
+> single unified file that handles both Qwen 3.5 and 3.6 variants. The path is repo-relative,
 > so launch `llama-server` from the repository root (as the examples above do).
 > If you cloned without `--recurse-submodules`, run `git submodule update --init`
 > first — otherwise startup fails with a missing-file error. Gemma and

presets/models_16GB_VRAM.ini

@@ -91,7 +91,7 @@ cache-type-v = q8_0
 flash-attn = true
 reasoning = on
 jinja = true
-chat-template-file = vendor\Qwen-Fixed-Chat-Templates\qwen3.6\chat_template-v13.jinja
+chat-template-file = vendor\Qwen-Fixed-Chat-Templates\chat_template.jinja
 kv-unified = true
 ctx-checkpoints = 32
 cache-ram = 51200
@@ -120,8 +120,7 @@ cache-type-v = q8_0
 flash-attn = true
 reasoning = on
 jinja = true
-chat-template-file = vendor\Qwen-Fixed-Chat-Templates\qwen3.6\chat_template-v13.jinja
-chat-template-kwargs = {"preserve_thinking":true}
+chat-template-file = vendor\Qwen-Fixed-Chat-Templates\chat_template.jinja
 kv-unified = true
 ctx-checkpoints = 32
 cache-ram = 51200
@@ -150,8 +149,7 @@ cache-type-v = q8_0
 flash-attn = true
 reasoning = on
 jinja = true
-chat-template-file = vendor\Qwen-Fixed-Chat-Templates\qwen3.6\chat_template-v13.jinja
-chat-template-kwargs = {"preserve_thinking":true}
+chat-template-file = vendor\Qwen-Fixed-Chat-Templates\chat_template.jinja
 kv-unified = true
 ctx-checkpoints = 32
 cache-ram = 51200

presets/models_24GB_VRAM.ini

@@ -64,8 +64,7 @@ cache-type-v = q8_0
 flash-attn = true
 reasoning = on
 jinja = true
-chat-template-file = vendor\Qwen-Fixed-Chat-Templates\qwen3.6\chat_template-v13.jinja
-chat-template-kwargs = {"preserve_thinking":true}
+chat-template-file = vendor\Qwen-Fixed-Chat-Templates\chat_template.jinja
 kv-unified = true
 ctx-checkpoints = 32
 cache-ram = 51200
@@ -95,8 +94,7 @@ cache-type-v = q8_0
 flash-attn = true
 reasoning = on
 jinja = true
-chat-template-file = vendor\Qwen-Fixed-Chat-Templates\qwen3.6\chat_template-v13.jinja
-chat-template-kwargs = {"preserve_thinking":true}
+chat-template-file = vendor\Qwen-Fixed-Chat-Templates\chat_template.jinja
 kv-unified = true
 ctx-checkpoints = 32
 cache-ram = 51200