---
license: apache-2.0
base_model:
- prithivMLmods/Q3.5-9B-DS-v4-Flash-v2.0
tags:
- text-generation-inference
- pytorch
- flash
- math
- reasoning
- sft
- deepseek-v4
- bf16
- abliterated
language:
- en
pipeline_tag: image-text-to-text
library_name: transformers
---

![1](https://cdn-uploads.huggingface.co/production/uploads/65bb837dbfb878f46c77de4c/SSnWXmMWnGJHLFnjcA4bG.png)

# **Qwen3.5-9B-DS-v4-Flash-v3.0**

> **Qwen3.5-9B-DS-v4-Flash-v3.0** is a reasoning-capable **9B-parameter** language model based on **prithivMLmods/Q3.5-9B-DS-v4-Flash-v2.0**, which is built on top of **Qwen/Qwen3.5-9B**. The model was trained through a multi-stage training pipeline using approximately **3.5K filtered samples** drawn from **DeepSeek V4 Flash reasoning traces**, along with additional high-quality reasoning datasets, to improve long-form reasoning, mathematical problem solving, scientific analysis, coding, and instruction-following capabilities.

> [!NOTE]
> This model is an experimental release and may generate unexpected behaviors or reasoning artifacts in certain scenarios.

## **Key Highlights**

* **Qwen 3.5 Foundation**: Built on top of **Qwen/Qwen3.5-9B** through the prior **prithivMLmods/Q3.5-9B-DS-v4-Flash-v2.0** release.
* **Multi-Stage Training**: Trained through multiple stages to progressively improve reasoning performance.
* **Long-Context Reasoning**: Incorporates approximately **3.5K filtered samples** from DeepSeek V4 Flash reasoning traces spanning mathematics, science, coding, and complex analytical tasks.
* **General Reasoning Enhancement**: Further trained on additional high-quality reasoning datasets to strengthen instruction following and multi-step reasoning.
* **Research-Focused Release**: Designed for reasoning research, experimentation, and evaluation.
* **Efficient 9B Deployment**: Suitable for local inference and research environments.

## **Quick Start with Transformers**

```bash
pip install transformers
pip install accelerate
```

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

model = AutoModelForCausalLM.from_pretrained(
    "prithivMLmods/Qwen3.5-9B-DS-v4-Flash-v3.0",
    torch_dtype="auto",
    device_map="auto"
)

tokenizer = AutoTokenizer.from_pretrained(
    "prithivMLmods/Qwen3.5-9B-DS-v4-Flash-v3.0"
)

messages = [
    {
        "role": "user",
        "content": "Explain how a transformer model processes text."
    }
]

inputs = tokenizer.apply_chat_template(
    messages,
    tokenize=True,
    add_generation_prompt=True,
    return_tensors="pt"
).to(model.device)

outputs = model.generate(
    inputs,
    max_new_tokens=512
)

print(
    tokenizer.decode(
        outputs[0][inputs.shape[-1]:],
        skip_special_tokens=True
    )
)
```

## **Training Details**

| Setting                            | Value                                                                                                                                                                                                                                                                |
| :--------------------------------- | :------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| **Base Model**                     | **prithivMLmods/Q3.5-9B-DS-v4-Flash-v2.0**                                                                                                                                                                                                                           |
| **Original Backbone**              | **Qwen/Qwen3.5-9B**                                                                                                                                                                                                                                                  |
| **Training Method**                | Multi-stage Supervised Fine-Tuning (SFT)                                                                                                                                                                                                                             |
| **Maximum Sequence Length**        | **32,768 tokens (Long Context)**                                                                                                                                                                                                                                     |
| **Training Precision**             | **BF16 (Full Precision)**                                                                                                                                                                                                                                            |
| **Training & Alignment Framework** | [TRL](https://github.com/huggingface/trl)                                                                                                                                                                                                                            |
| **Training Datasets**              | [Jackrong/DeepSeek-V4-Distill-8000x](https://huggingface.co/datasets/Jackrong/DeepSeek-V4-Distill-8000x), [sequelbox/Titanium4-DeepSeek-V4-Pro](https://huggingface.co/datasets/sequelbox/Titanium4-DeepSeek-V4-Pro), and additional high-quality reasoning datasets |

## **Intended Use**

* **Reasoning Research**: Studying long-context reasoning and multi-stage training techniques.
* **Mathematical Reasoning**: Solving complex mathematical problems with multi-step reasoning.
* **Scientific Reasoning**: Performing structured scientific analysis and problem solving.
* **Coding Assistance**: Improving code understanding and generation through long-context reasoning.
* **Instruction Following**: Evaluating and improving instruction-following capabilities.
* **Local Deployment**: Running efficient 9B reasoning models in research and experimentation environments.

## **Limitations**

* **Experimental Model**: Behavior may differ from the base model in certain scenarios.
* **Reasoning Artifacts**: Complex reasoning chains may occasionally produce incorrect intermediate steps or conclusions.
* **Training Biases**: Performance reflects the characteristics and coverage of the reasoning datasets used during training.

## **Acknowledgements**

* **[Qwen/Qwen3.5-9B](https://huggingface.co/Qwen/Qwen3.5-9B)**: Base model used for this project.

* **TRL - [Transformers Reinforcement Learning](https://huggingface.co/docs/trl/en/index)**: TRL is a full stack library providing tools to train transformer language models with methods including Supervised Fine-Tuning (SFT), Group Relative Policy Optimization (GRPO), Direct Preference Optimization (DPO), Reward Modeling, and more.

* **[Transformers](https://huggingface.co/docs/transformers/en/index)**: Transformers provides state-of-the-art machine learning models for text, computer vision, audio, video, and multimodal tasks, supporting both inference and training.