OLMo-2-1124-7B-Instruct SFT mathfisher
fine-tuned language model for math tasks
1 min readupdated
This model is a fine-tuned version of the base allenai/OLMo-2-1124-7B-Instruct model with 7.3 billion parameters. Fine-tuning was performed using the TRL library (a library for reinforcement learning and supervised fine-tuning of language models). The new model focuses on solving tasks that require reasoning and math while retaining the capabilities of the base architecture from AllenAI.
Key features
- Based on allenai/OLMo-2-1124-7B-Instruct
- Uses the TRL (Transformers Reinforcement Learning) framework
- SFT (supervised fine-tuning) to improve reasoning skills
- 7.3B parameter size to run on consumer GPUs
What you can do
- Integrate into Telegram bots or local assistants to help with calculations
- Use for text generation and answering technical questions
- Run experiments with math prompts and benchmarks
Example
from transformers import pipeline
question = "If you had a time machine, but could only go to the past or the future once and never return, which would you choose and why?"
generator = pipeline("text-generation", model="Neelectric/OLMo-2-1124-7B-Instruct_SFT_mathfisher_v00.01_s43", device="cuda")
output = generator([{"role": "user", "content": question}], max_new_tokens=128, return_full_text=False)[0]
print(output["generated_text"])Read more
DetailsBase model: allenai/OLMo-2-1124-7B-Instruct
Training method: SFT
Library versions: TRL 1.1.0.dev0, Transformers 4.57.6, Pytorch 2.9.0










Comments