prithivMLmods/Qwen3.5-9B-DS-v4-Flash-v3.0
prithivMLmods/Qwen3.5-9B-DS-v4-Flash-v3.0 is a 9-billion parameter language model built upon Qwen/Qwen3.5-9B, further developed from prithivMLmods/Q3.5-9B-DS-v4-Flash-v2.0. This model is specifically fine-tuned through a multi-stage process using DeepSeek V4 Flash reasoning traces and other high-quality reasoning datasets. It excels in long-form reasoning, mathematical problem-solving, scientific analysis, coding, and instruction-following, making it suitable for research and experimentation in complex analytical tasks.
Loading preview...
Model Overview
prithivMLmods/Qwen3.5-9B-DS-v4-Flash-v3.0 is a 9-billion parameter language model designed for enhanced reasoning capabilities. It is built on the Qwen/Qwen3.5-9B foundation, evolving from the prior prithivMLmods/Q3.5-9B-DS-v4-Flash-v2.0 release. The model underwent a multi-stage supervised fine-tuning (SFT) process, utilizing approximately 3.5K filtered samples from DeepSeek V4 Flash reasoning traces, alongside additional high-quality reasoning datasets.
Key Capabilities
- Enhanced Reasoning: Significantly improved in long-form reasoning, mathematical problem-solving, scientific analysis, and coding tasks.
- Instruction Following: Strengthened ability to follow complex instructions through targeted training.
- Long Context: Supports a maximum sequence length of 32,768 tokens, enabling processing of extensive inputs.
- Research-Focused: An experimental release primarily intended for reasoning research, evaluation, and studying multi-stage training techniques.
- Efficient Deployment: A 9B-parameter model suitable for local inference and research environments.
Intended Use Cases
- Reasoning Research: Ideal for investigating long-context reasoning and multi-stage training methodologies.
- Complex Problem Solving: Effective for mathematical and scientific problems requiring multi-step reasoning.
- Coding Assistance: Useful for improving code understanding and generation, especially with long contexts.
- Instruction Following Evaluation: A tool for assessing and enhancing instruction-following performance.
Limitations
As an experimental model, it may exhibit unexpected behaviors or reasoning artifacts. Performance is influenced by the characteristics and coverage of its training datasets, and complex reasoning chains might occasionally produce incorrect intermediate steps or conclusions.