---
base_model:
- bowen-upenn/Qwen3-4B-CreativeWriting-SFT
- Alibaba-Apsara/DASD-4B-Thinking
- WithinUsAI/Qwen3-Space.Agent.Claude.Uncensored-4B
license: apache-2.0
tags:
- merge
- slerp
- qwen3
- qwen3-4b
- creative-writing
- reasoning
- long-cot
- uncensored
- agentic
- slerp-merge
datasets:
- unalignment/toxic-dpo-v0.2
- NobodyExistsOnTheInternet/ToxicQAFinal
- Orion-zhen/dpo-toxic-zh
- Alibaba-Apsara/Superior-Reasoning-SFT-gpt-oss-120b-Logprob
- Alibaba-Apsara/Superior-Reasoning-SFT-gpt-oss-120b
---

# Qwen3-Space.Agent_DASD-Uncensored-4B

**A SLERP-merged Qwen3-4B combining creative writing, deep reasoning, and uncensored agentic capabilities.**

## Model Composition

Sequential **Spherical Linear Interpolation (SLERP)**:

1. **bowen-upenn/Qwen3-4B-CreativeWriting-SFT** + **Alibaba-Apsara/DASD-4B-Thinking** @ t=0.5
2. Result + **WithinUsAI/Qwen3-Space.Agent.Claude.Uncensored-4B** @ t=0.5

### Parent Models
- **Creative Writing SFT**: Storytelling, narrative, character development, stylistic prose.
- **DASD-4B-Thinking**: Long chain-of-thought reasoning, complex problem solving.
- **Claude-Uncensored Agent**: Agentic behavior, tool use, reduced refusals, Claude-like personality.

## Key Strengths
- Balanced creativity + deep reasoning
- Strong agentic / tool-use capabilities
- Reduced censorship compared to typical aligned models
- Good at long-context creative + reasoning tasks

## Usage

```python
from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

model_id = "GODsStrongestSoldier/Qwen3-Space.Agent_DASD-Uncensored-4B"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id, torch_dtype=torch.bfloat16, device_map="auto")

messages = [{"role": "user", "content": "Write a thoughtful sci-fi story with internal monologue."}]
inputs = tokenizer.apply_chat_template(messages, return_tensors="pt").to(model.device)
output = model.generate(inputs, max_new_tokens=1200, temperature=0.75, do_sample=True)
print(tokenizer.decode(output[0], skip_special_tokens=True))
```

## Recommended Settings
- Temperature: 0.7-0.9 for creative tasks
- Use system prompts for agent behavior
- Encourage step-by-step thinking for complex problems

## Technical Specs
- Base: Qwen3-4B (36 layers, GQA, 32k context)
- Merge: Sequential SLERP (no additional training)
- Size: ~4B parameters
- Precision: FP16 (safetensors)

## Acknowledgments
- bowen-upenn, Alibaba-Apsara, WithinUsAI, and the Qwen team.

*Merged on Kaggle via sequential SLERP — May 2026*