---
base_model: Qwen/Qwen3-4B
license: apache-2.0
library_name: transformers
tags:
- thinking-preservation
- qwen3
- terminal-agent
---

# Qwen3-4B-Thinking-Preservation

Derived from Qwen/Qwen3-4B (hybrid thinking model). The chat template no longer strips `<think>` from prior assistant turns and the nonthinking branch is removed, so the generation prompt always opens `<think>` (like Qwen3-4B-Thinking-2507).

**Thinking is always preserved across multi-turn history (append-only).** Every
assistant turn keeps its `<think>...</think>` reasoning, not just the latest one, and
the generation prompt always opens `<think>` (passing `enable_thinking=False` has no
effect). This makes multi-turn agent **training match evaluation** — the model always
sees its own prior reasoning. Model weights are identical to [Qwen/Qwen3-4B](https://huggingface.co/Qwen/Qwen3-4B);
only the chat template differs.