OLMo-2-1124-7B-SFT
| Property | Value |
|---|---|
| Base Model | OLMo2-7B-1124 |
| License | Apache 2.0 |
| Language | English |
| Training Dataset | Tulu-3 SFT mixture |
| Paper | Tulu-3 Paper |
What is OLMo-2-1124-7B-SFT?
OLMo-2-1124-7B-SFT is a supervised fine-tuned variant of the OLMo 2 7B base model, specifically trained on an OLMo-adapted version of the Tulu-3 dataset. This model represents part of the Open Language Model (OLMo) series developed by Allen AI, designed to advance the science of language models while maintaining full transparency and openness.
Implementation Details
The model employs specific hyperparameters for its supervised fine-tuning, including a learning rate of 1E-5, an effective batch size of 64, and a maximum sequence length of 4096. It uses a linear learning rate schedule with a 0.03 warmup ratio over 2 epochs.
- Built on transformer architecture with 7B parameters
- Implements a specialized chat template for conversation
- Supports standard HuggingFace integration
- Trained with sum loss accumulation
Core Capabilities
- Strong performance on mathematical reasoning (71.2% on GSM8k)
- Excellent safety metrics (82.4% on safety benchmarks)
- Robust general knowledge (62.0% on MMLU)
- Enhanced performance on DROP (58.2%) and IFEval (68.0%)
Frequently Asked Questions
Q: What makes this model unique?
The model stands out for its fully open nature and strong performance across diverse tasks, particularly in safety and mathematical reasoning. It's part of a larger ecosystem of fully transparent AI models, with all training code and details publicly available.
Q: What are the recommended use cases?
The model is particularly well-suited for research and educational applications, excelling in tasks requiring mathematical reasoning, safety-conscious responses, and general knowledge applications. It's designed for both conversational and analytical tasks.
