RL-Forgetting-Experiments-3/qwen2.5-3b-math-sft-shuffled-lr1e5-step1072
TEXT GENERATIONPricing:Input $0.32 / Cached $0.064 / Output $1.6Concurrent Unit Cost:1Model Size:3.1BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Sep 22, 2026Architecture:Transformer Featherless Exclusive Cold
The RL-Forgetting-Experiments-3/qwen2.5-3b-math-sft-shuffled-lr1e5-step1072 model is a 3.1 billion parameter Qwen2.5-based language model, specifically an SFT checkpoint from an RL-Forgetting experiment. This iteration is fine-tuned for mathematical tasks, originating from a shuffled dataset arm. It is designed for applications requiring mathematical reasoning and problem-solving capabilities.
Loading preview...
Model Overview
This model, qwen2.5-3b-math-sft-shuffled-lr1e5-step1072, is a 3.1 billion parameter SFT (Supervised Fine-Tuning) checkpoint based on the Qwen2.5 architecture. It is a specific iteration from an RL-Forgetting experiment, captured at optimizer step 1072.
Key Characteristics
- Architecture: Qwen2.5-3B base model.
- Fine-tuning: Supervised Fine-Tuning (SFT) specifically for mathematical tasks.
- Training Data: Derived from a "math shuffled" dataset arm, indicating a focus on mathematical content with a specific data ordering strategy.
- Origin: Part of a broader RL-Forgetting experiment, suggesting it might be evaluated for its ability to retain or forget specific information.
Intended Use Cases
- Mathematical Reasoning: Best suited for tasks involving mathematical problem-solving, calculations, and understanding mathematical concepts.
- Research in Forgetting/Retention: Potentially useful for researchers studying catastrophic forgetting or knowledge retention in large language models, given its experimental context.
- Specialized Math Applications: Can be integrated into applications requiring a compact, math-focused language model.