tomhata/ul-wk2-gpt-neo-1.3B

Model Card for Model ID

Model Details

GPT-Neo 1.3B model, fine-tuned on math-based prompts using ORPO.

Model Description

Evaluated on mmlu elementary mathematics dataset using Eleuther evaluation harness

Tasks	Version	Filter	n-shot	Metric		Value		Stderr
elementary_mathematics	1	none	0	acc	↑	0.2778	±	0.023

This is a slight improvment over the base model, which has an accuracy of 0.2751 ± 0.023. However, this improvement in accuracy is within variance.

This is the model card of a 🤗 transformers model that has been pushed on the Hub. This model card has been automatically generated.

Uses

As a starting point for questions regarding math concepts. Actual calculations and values may not be reliable.

Out-of-Scope Use

Not to be used as a replacement to learn actual math.

Bias, Risks, and Limitations

This model is unlikely to produce accurate numerical results.

Training Details

Training Data

orpo-explorers/distilabel-math-preference

Training Procedure

1000 entries were randomly sampled. This subsample was further split 90/10 into train/test.

Preprocessing [optional]

Entries were split into prompt, chosen, rejected triplets. Maximum length for each of these entries was limited to 2048 characters.

Training Hyperparameters

• Training regime: [More Information Needed]