qwen36_adapter.h File Reference

Qwen 3.6 chat adapter (v2.1.9, gh#45). More...

#include <entropic/inference/adapters/adapter_base.h>
#include <unordered_map>

Include dependency graph for qwen36_adapter.h:

This graph shows which files directly or indirectly include this file:

Go to the source code of this file.

Classes
class	entropic::Qwen36Adapter
	Qwen 3.6 MoE chat adapter. More...

Namespaces
namespace	entropic
	Activate model on GPU (WARM → ACTIVE).

Detailed Description

Qwen 3.6 chat adapter (v2.1.9, gh#45).

Tool-call format

Qwen 3.6 emits the same XML function-call format the vLLM qwen3_coder parser recognises:

<tool_call>
<function=filesystem.read_file>
<parameter=path>/src/main.cpp</parameter>
</function>
</tool_call>

That format overlaps with Qwen 3.5 today. The class remains distinct per the v2.1.9 "no version lumping" decision — template particulars across the 3.x line can diverge later; a QwenBaseAdapter extraction is a refactor for when overlap is proved, not a precondition.

Reasoning

Qwen 3.6 is a thinking model; reasoning is wrapped in <think>...</think> blocks. Stripping and extraction reuse the base class primitives.

Vision

Qwen 3.6 ships paired with an mmproj-F16.gguf (see registry key qwen3_6_a3b_mmproj) and follows the OpenAI content-array convention for image parts. The vision system-prompt extension mirrors Qwen 3.5.

Internal to inference .so.

Version

2.1.9

Definition in file qwen36_adapter.h.