Centralized repository for the Math-LM project, an open-source replication of the Minerva model.
This project contains the following directories
proof-pile-v2
: scripts for downloading and preprocessing data.task-finetunes
: scripts for fine-tuning models on task-specific datasets, such as MATH or GSM8k.
Model weights are stored on the Stability cluster and Huggingface.
In addition, this project makes contributions to EleutherAI/lm-evaluation-harness, such as PR no. 376.