Prime the opening <think> in the generation prompt and remove enable_thinking

README.md

@@ -106,11 +106,12 @@ Okay, the user is asking who Molière was, and they want a concise answer. Let m
 Jean-Baptiste Poquelin, known as Molière, was a 17th-century French playwright, actor, and poet. Born in Paris in 1622, … <|im_end|>
 ``` 
 
-Thinking can be turned off by using ```tokenizer.apply_chat_template(…, enable_thinking=False)```. This will append empty thinking tags (```<think>\n\n</think>\n\n```) in the prompt, immediately after "`assistant\n`".
-However, for testing a non-thinking model, it is better to conduct experiments on [Luciole-8B-Instruct-1.1](https://huggingface.co/OpenLLM-France/Luciole-8B-Instruct-1.1).
+For a non-thinking model, use [Luciole-8B-Instruct-1.1](https://huggingface.co/OpenLLM-France/Luciole-8B-Instruct-1.1) instead.
 
 By default, the `<think>…</think>` reasoning of all previous assistant turns is kept when formatting a multi-turn conversation. Setting ```tokenizer.apply_chat_template(…, preserve_thinking=False)``` drops it from every assistant turn preceding the last user message (only the most recent turn keeps its reasoning), which shortens the prompt and reduces the KV-cache footprint. With vLLM, set it per request via ```extra_body={"chat_template_kwargs": {"preserve_thinking": false}}```. Ollama does not use this parameter, as it formats prompts with its own built-in renderer.
 
+The chat template prefills the opening `<think>` tag in the generation prompt, so the model's output starts directly with the reasoning and contains only the closing `</think>` (reasoning, then `</think>`, then the answer). When serving with vLLM, add `--reasoning-parser qwen3` so the reasoning is returned separately in `reasoning_content` and the final answer in `content` (SGLang uses the same `--reasoning-parser qwen3`). This requires a recent vLLM (≥ 0.24; prefer the latest).
+
 ### Training procedure
 
 Luciole-8B-KD-Math-Thinking-1.1 was trained in two phases. The model was first finetuned on thinking data, followed by DPO on preference data with thinking traces. 
@@ -186,7 +187,8 @@ docker run --runtime nvidia --gpus=all \
     -p 8000:8000 \
     --ipc=host \
     vllm/vllm-openai:latest \
-    --model OpenLLM-France/Luciole-8B-KD-Math-Thinking-1.1
+    --model OpenLLM-France/Luciole-8B-KD-Math-Thinking-1.1 \
+    --reasoning-parser qwen3
 ```
 
 #### 2. Test using OpenAI Client in Python

chat_template.jinja

@@ -1,6 +1,9 @@
 {%- if preserve_thinking is not defined %}
     {%- set preserve_thinking = true %}
 {%- endif %}
+{%- if enable_thinking is defined and enable_thinking is false %}
+    {{- raise_exception('Disabling thinking (enable_thinking=False) is not supported by this model. For a non-thinking model, use an OpenLLM-France Luciole Instruct model.') }}
+{%- endif %}
 {%- if messages[0].role == 'system' and messages[0].content.strip() != "" %}
     {%- set system_message = messages[0].content %}
 {%- else %}
@@ -84,8 +87,5 @@
     {%- endif %}
 {%- endfor %}
 {%- if add_generation_prompt %}
-    {{- '<|im_start|>assistant\n' }}
-    {%- if enable_thinking is defined and enable_thinking is false %}
-        {{- '<think>\n\n</think>\n\n' }}
-    {%- endif %}
+    {{- '<|im_start|>assistant\n<think>\n' }}
 {%- endif %}