# DPO Optimization

A creative writing enhancement model fine-tuned and optimized based on DeepSeek-R1-Distill-Qwen-14B, showing significant improvements in creative writing

This is a speech language model based on discrete Hubert tokens, focusing on efficient training and capable of generating speech segment continuations.

SummLlama3.1-8B is a text summarization model initialized from Llama3.1-8B-Instruct, optimized through large-scale summarization feedback via Direct Preference Optimization (DPO), excelling in fidelity, completeness, and conciseness.

UNA-ThePitbull-21.4B-v2 is a large language model based on 21.4B parameters, with performance close to 70B models, integrating emotional intelligence and IQ, excelling in dialogue and text generation.

OpenBioLLM-70B is an advanced open-source language model specifically designed for the biomedical field, fine-tuned based on Meta-Llama-3-70B-Instruct, demonstrating outstanding performance in biomedical tasks.

A human-preference-aligned chat model supporting Hungarian and English, adapted from Llama-2-7b for Hungarian

LLaVA-v1.5-13B-DPO is a vision-language model based on the LLaVA framework, trained with Direct Preference Optimization (DPO) and converted to GGUF quantized format to improve inference efficiency.

A Chinese text error correction model trained with DPO, capable of automatically detecting and correcting spelling and grammar errors in Chinese text.

TheBeagle is a 7-billion-parameter model trained on The Bagel dataset, optimized with DPO (Direct Preference Optimization) and UNA (Unified Neural Architecture) techniques, demonstrating excellent performance in multi-task scenarios.

Rocket-3B is a 3-billion-parameter large language model trained on public datasets through Direct Preference Optimization (DPO), outperforming many larger-scale models.