Part A: qwen3-no-think.jinja.j2 Ansible template
- New template: templates/qwen3-no-think.jinja.j2
Standard Qwen3 chat template with enable_thinking unconditionally false
(hardcoded empty <think></think> prefix at add_generation_prompt step).
Deployed to /opt/models/templates/qwen3-no-think.jinja on astro-orbiter.
- tasks/models.yml: deploy templates dir + qwen3-no-think.jinja via
ansible.builtin.template task (tags: models, chat_templates).
Part B: INI preset template — two Qwen3-8B sections
- templates/llama-server-router-preset.ini.j2:
[Qwen3-8B-Q4_K_M] — thinking variant, same GGUF, baked-in template.
[Qwen3-8B-Q4_K_M-no_think] — no-think variant, same GGUF,
chat-template-file = /opt/models/templates/qwen3-no-think.jinja.
Both sections: n-gpu-layers=99, ctx-size=32768, flash-attn=true,
q4_0 KV cache, sleep-idle-seconds=60.
Part C: defaults/main.yml — llama-swap models + matrix
- llm_swapmode_models: added Qwen3-8B-Q4_K_M (port 8106, GPU) and
Qwen3-8B-Q4_K_M-no_think (port 8107, GPU, chat_template_file set).
- llm_swapmode_matrix_rows: row5 (Qwen3-8B thinking + embed),
row6 (Qwen3-8B no_think + embed). Neither co-resident with Qwen3.8-27B.
- templates/llama-swap-config.yaml.j2: added chat_template_file support
(--chat-template-file flag conditional on model.chat_template_file).
- tasks/swapmode.yml: GATE 2 assert updated 5 -> 7 models.
Part D: war-machine config.yaml (not tracked in git)
- custom_providers.astro-orbiter.models: added Qwen3-8B-Q4_K_M-no_think
(context_length: 32768).
- Reassigned 5 latency-sensitive aux tasks from Meta-Llama-3.1-8B to
Qwen3-8B-Q4_K_M-no_think: skills_hub, approval, mcp, title_generation,
profile_describer. Rationale: GPU-resident, lower latency, json_schema OK.
- web_extract and compression remain on Phi-3.5-mini (long scrapes).
VRAM: both Qwen3-8B variants co-reside with nomic-embed only (~10.4GB +
84MB). Cannot co-reside with Qwen3.8-27B (17.8GB); LRU eviction applies.
104 lines
5.0 KiB
Django/Jinja
104 lines
5.0 KiB
Django/Jinja
{#
|
|
FILE: roles/llm-inference-multimodel/templates/qwen3-no-think.jinja.j2
|
|
DESCRIPTION: Qwen3 chat template with enable_thinking unconditionally false.
|
|
|
|
This is a chat-template FILE deployed to {{ llm_models_dir }}/templates/qwen3-no-think.jinja
|
|
on astro-orbiter and referenced via chat-template-file in the INI preset for
|
|
[Qwen3-8B-Q4_K_M-no_think]. The [Qwen3-8B-Q4_K_M] section uses the model's
|
|
baked-in default template (thinking ON by default).
|
|
|
|
Mechanism: at the add_generation_prompt step, instead of checking
|
|
"enable_thinking is defined and enable_thinking is false"
|
|
we UNCONDITIONALLY emit the empty <think></think> prefix that suppresses CoT.
|
|
This is the hard-switch documented in the Qwen3 template spec and confirmed in
|
|
https://huggingface.co/blog/qwen-3-chat-template-deep-dive (section 1).
|
|
|
|
All other logic is identical to /opt/llama.cpp/models/templates/Qwen-Qwen3-0.6B.jinja
|
|
(the shipped template for Qwen3). Only the final add_generation_prompt block differs.
|
|
|
|
Added: t_664289a0 (2026-08-19, War Machine) — dual thinking deployment.
|
|
#}
|
|
{%- if tools %}
|
|
{{- '<|im_start|>system\n' }}
|
|
{%- if messages[0].role == 'system' %}
|
|
{{- messages[0].content + '\n\n' }}
|
|
{%- endif %}
|
|
{{- "# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>" }}
|
|
{%- for tool in tools %}
|
|
{{- "\n" }}
|
|
{{- tool | tojson }}
|
|
{%- endfor %}
|
|
{{- "\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call><|im_end|>\n" }}
|
|
{%- else %}
|
|
{%- if messages[0].role == 'system' %}
|
|
{{- '<|im_start|>system\n' + messages[0].content + '<|im_end|>\n' }}
|
|
{%- endif %}
|
|
{%- endif %}
|
|
{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
|
|
{%- for message in messages[::-1] %}
|
|
{%- set index = (messages|length - 1) - loop.index0 %}
|
|
{%- if ns.multi_step_tool and message.role == "user" and not(message.content.startswith('<tool_response>') and message.content.endswith('</tool_response>')) %}
|
|
{%- set ns.multi_step_tool = false %}
|
|
{%- set ns.last_query_index = index %}
|
|
{%- endif %}
|
|
{%- endfor %}
|
|
{%- for message in messages %}
|
|
{%- if (message.role == "user") or (message.role == "system" and not loop.first) %}
|
|
{{- '<|im_start|>' + message.role + '\n' + message.content + '<|im_end|>' + '\n' }}
|
|
{%- elif message.role == "assistant" %}
|
|
{%- set content = message.content %}
|
|
{%- set reasoning_content = '' %}
|
|
{%- if message.reasoning_content is defined and message.reasoning_content is not none %}
|
|
{%- set reasoning_content = message.reasoning_content %}
|
|
{%- else %}
|
|
{%- if '</think>' in message.content %}
|
|
{%- set content = message.content.split('</think>')[-1].lstrip('\n') %}
|
|
{%- set reasoning_content = message.content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
|
|
{%- endif %}
|
|
{%- endif %}
|
|
{%- if loop.index0 > ns.last_query_index %}
|
|
{%- if loop.last or (not loop.last and reasoning_content) %}
|
|
{{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content.strip('\n') + '\n</think>\n\n' + content.lstrip('\n') }}
|
|
{%- else %}
|
|
{{- '<|im_start|>' + message.role + '\n' + content }}
|
|
{%- endif %}
|
|
{%- else %}
|
|
{{- '<|im_start|>' + message.role + '\n' + content }}
|
|
{%- endif %}
|
|
{%- if message.tool_calls %}
|
|
{%- for tool_call in message.tool_calls %}
|
|
{%- if (loop.first and content) or (not loop.first) %}
|
|
{{- '\n' }}
|
|
{%- endif %}
|
|
{%- if tool_call.function %}
|
|
{%- set tool_call = tool_call.function %}
|
|
{%- endif %}
|
|
{{- '<tool_call>\n{"name": "' }}
|
|
{{- tool_call.name }}
|
|
{{- '", "arguments": ' }}
|
|
{%- if tool_call.arguments is string %}
|
|
{{- tool_call.arguments }}
|
|
{%- else %}
|
|
{{- tool_call.arguments | tojson }}
|
|
{%- endif %}
|
|
{{- '}\n</tool_call>' }}
|
|
{%- endfor %}
|
|
{%- endif %}
|
|
{{- '<|im_end|>\n' }}
|
|
{%- elif message.role == "tool" %}
|
|
{%- if loop.first or (messages[loop.index0 - 1].role != "tool") %}
|
|
{{- '<|im_start|>user' }}
|
|
{%- endif %}
|
|
{{- '\n<tool_response>\n' }}
|
|
{{- message.content }}
|
|
{{- '\n</tool_response>' }}
|
|
{%- if loop.last or (messages[loop.index0 + 1].role != "tool") %}
|
|
{{- '<|im_end|>\n' }}
|
|
{%- endif %}
|
|
{%- endif %}
|
|
{%- endfor %}
|
|
{%- if add_generation_prompt %}
|
|
{{- '<|im_start|>assistant\n' }}
|
|
{{- '<think>\n\n</think>\n\n' }}
|
|
{%- endif %}
|