LSW142857/OPSD-Qwen3.5-9B-Medium-545-Checkpoint7-Merged

VISIONPricing:Input $0.431 / Cached $0.0862 / Output $1.12Concurrent Unit Cost:1Model Size:9BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Sep 10, 2026Architecture:Transformer Featherless Exclusive Cold

LSW142857/OPSD-Qwen3.5-9B-Medium-545-Checkpoint7-Merged is a 9 billion parameter Qwen3.5-based language model, developed by LSW142857, specifically fine-tuned using the OPSD method on 545 student-error tasks. This model integrates merged LoRA updates and a complete trained MTP module, designed for robust performance in complex problem-solving scenarios. It is optimized for tasks requiring detailed reasoning and error correction, leveraging a 32768 token context length.

Loading preview...

Model Overview

LSW142857/OPSD-Qwen3.5-9B-Medium-545-Checkpoint7-Merged is a 9 billion parameter Qwen3.5 model, representing the seventh checkpoint from an ongoing validation run. It is initialized from an expert-SFT qwen35-9b-expert-sft-131k-lora64-block28 and further trained with the OPSD (Optimized Problem-Solving with Dynamic-PI) method on 545 student-error tasks. This model includes merged LoRA updates and a complete trained MTP (Multi-Task Pretraining) module, making it a self-contained inference artifact without requiring separate adapters.

Key Characteristics

  • Architecture: Qwen3.5-9B base with LoRA (rank 64, alpha 128) and MTP module.
  • Training: Fine-tuned on 545 student-error tasks using a stage-adaptive PI approach, with 8 optimizer updates completed.
  • Context Length: Supports a native context of 32768 tokens, with evaluation settings extending to 262144 tokens.
  • Integrated Components: Includes all model weight shards, tokenizer, chat template, processor configuration, and merged LoRA updates.

Use Cases

This model is particularly suited for applications requiring:

  • Complex Problem Solving: Its training on student-error tasks suggests proficiency in identifying and correcting errors.
  • Reasoning Tasks: The OPSD training method and MTP module aim to enhance reasoning capabilities.
  • Long Context Understanding: With a 32768 token context, it can process and understand extensive inputs.