kyujinpy/Sakura-SOLRCA-Math-Instruct-DPO-v1

Hugging Face
TEXT GENERATIONConcurrency Cost:1Model Size:10.7BQuant:FP8Ctx Length:4kPublished:Dec 25, 2023License:cc-by-nc-sa-4.0Architecture:Transformer Open Weights Warm

The Sakura-SOLRCA-Math-Instruct-DPO-v1 is a 10.7 billion parameter instruction-tuned causal language model developed by Kyujin Han (kyujinpy). It was fine-tuned using the DPO method on a combination of Intel/orca_dpo_pairs and argilla/distilabel-math-preference-dpo datasets. This model is specifically optimized for mathematical reasoning and general instruction following, achieving a 63.84 score on GSM8K and a 74.13 average on the Open LLM Leaderboard.

Loading preview...

Overview

The kyujinpy/Sakura-SOLRCA-Math-Instruct-DPO-v1 is a 10.7 billion parameter instruction-tuned language model developed by Kyujin Han (kyujinpy). This model leverages the Direct Preference Optimization (DPO) method, fine-tuned on a combination of the Intel/orca_dpo_pairs and argilla/distilabel-math-preference-dpo datasets. A merged version of these datasets, kyujinpy/orca_math_dpo, was also utilized in its development.

Key Capabilities

  • Mathematical Reasoning: Demonstrates strong performance in mathematical problem-solving, as indicated by its 63.84 score on the GSM8K benchmark.
  • Instruction Following: Designed to accurately follow instructions across a range of tasks, achieving an average score of 74.13 on the Open LLM Leaderboard.
  • General Language Understanding: Exhibits solid performance on various benchmarks including ARC (71.25), HellaSwag (88.48), MMLU (66.21), TruthfulQA (72.12), and Winogrande (82.87).

Good For

  • Applications requiring robust mathematical problem-solving capabilities.
  • General-purpose instruction-following tasks where accuracy is critical.
  • Developers looking for a DPO-tuned model with competitive benchmark performance in its size class.

Popular Sampler Settings

Top 3 parameter combinations used by Featherless users for this model. Click a tab to see each config.

temperature
top_p
top_k
frequency_penalty
presence_penalty
repetition_penalty
min_p