---
license: apache-2.0
language:
- en
- zh
- ko
- ja
- fr
- es
- de
- it
- ru
- ar
- multilingual
pipeline_tag: text-generation
tags:
- chat
- suzhou
- merged
- reasoning
- tool-use
- agent
library_name: transformers
base_model:
- tripplet-research/suzhou3.1
- Qwen/Qwen2.5-3B-Instruct
---

# Suzhou 3.2

A 12 billion parameter instruction-tuned language model by **Triplet Research**. Suzhou 3.2 is a weighted merge of Suzhou 3.1 and Qwen2.5-3B, designed to improve reasoning and math capabilities.

## Merge Details

- **Method**: Weighted blending (70% Suzhou 3.1 + 30% Qwen2.5-3B)
- **Model A**: Suzhou 3.1 - strong agent/tool-use, reasoning
- **Model B**: Qwen2.5-3B-Instruct - math reasoning, general knowledge
- **Target**: 12B parameters

## Key Features

- **12B parameters**
- **262K context window**
- Strong **reasoning** and **chain-of-thought** capabilities
- **Tool calling** and **agent** support
- **Multilingual** support (29+ languages)
- Mixed attention architecture (linear + full attention layers)

## Architecture

- Type: Causal Language Model
- Architecture: Qwen3.5 Text
- Layers: 32
- Parameters: 12B

## Quickstart

```python
from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("Triplet-Research/suzhou-3.2")
tokenizer = AutoTokenizer.from_pretrained("Triplet-Research/suzhou-3.2")
```