violetxi/qwen35-9b-equational-theory-sair-r3-mix10m

VISIONPricing:Input $0.431 / Cached $0.0862 / Output $1.12Concurrent Unit Cost:1Model Size:9BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Sep 24, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

The violetxi/qwen35-9b-equational-theory-sair-r3-mix10m is a 9 billion parameter Qwen3.5-based model, specifically fine-tuned for equational theory tasks. This model excels at mathematical reasoning, particularly in generating and evaluating proofs and counterexamples within equational logic. It features a 32,768-token context length and is optimized for tasks requiring structured logical deduction and formal verification.

Loading preview...

Overview

This model, violetxi/qwen35-9b-equational-theory-sair-r3-mix10m, is a 9 billion parameter Qwen3.5-based language model. It represents the final checkpoint of a full-model Supervised Fine-Tuning (SFT) run, specifically designed for tasks related to equational theory. The model is provided as a complete BF16 sharded safetensors model, including configuration, tokenizer, chat template, and processor files, requiring no adapter merge or custom code.

Key Capabilities

  • Equational Theory Reasoning: Specialized in generating and evaluating proofs and counterexamples in equational logic.
  • SAIR Evaluation: Achieves 63.66% Stage 1 accuracy and 20.42% Stage 2 judged validity on SAIR evaluations with a 24K output budget.
  • Extended Context: Supports a 32,768-token context length, crucial for complex logical sequences.
  • Native Qwen3.5 Layout: Fully compatible with Transformers and vLLM releases supporting Qwen3.5, ensuring straightforward integration.

Training Details

The model was trained on a nominal 10M nested mixture, comprising approximately 70% note tokens and 30% assistant-answer trajectories over two epochs. Training utilized eight GH200 GPUs with a learning rate of 5e-6 and a cosine schedule. It employs padding-free packing to isolate examples and masks prompts/assistant headers, focusing supervision on assistant answers and template markers.

Good for

  • Formal Verification: Tasks requiring the generation or validation of mathematical proofs.
  • Logical Deduction: Applications needing precise, structured reasoning in equational contexts.
  • Research in AI for Mathematics: Exploring advanced capabilities in automated theorem proving and mathematical reasoning.

Finetunes
1 models