# Synthetic data training

Phi-4-mini-reasoning is a lightweight open model built on synthetic data, focusing on high-quality, reasoning-rich data, and further fine-tuned for more advanced mathematical reasoning capabilities.

A zero-shot classification model fine-tuned based on roberta-base-zeroshot-v2.0-c, enhanced with SmartShot method and synthetic data

A vision-language model specialized in describing anime-style images, fine-tuned from SmolVLM-500M-Base, trained on 180K synthetic image/caption pairs generated by large language models.

GLiNER-Biomedical Edition is a specialized biomedical named entity recognition model developed based on the GLiNER framework, capable of identifying multiple biomedical entity types.

A Spanish grammar correction model based on the BART architecture, trained on the COWS-L2H dataset and 80,984 synthetic data entries, optimized for single-sentence correction

A large language model fine-tuned based on Qwen2.5-72B, specializing in text generation and instruction-following tasks

A fine-tuned version of Depth Anything V2 for outdoor metric depth estimation tasks, trained on the synthetic dataset Virtual KITTI

An efficient zero-shot classifier trained on synthetic data, suitable for topic classification, sentiment analysis, and reranking tasks in RAG workflows.

GLiClass is an efficient zero-shot classifier inspired by GLiNER, suitable for text classification, sentiment analysis, and reranking tasks in RAG workflows.

GLiClass is an efficient zero-shot classifier trained on synthetic data, suitable for text classification, sentiment analysis, and reranking tasks in RAG workflows.

A large language model developed in the third iteration using the Self-Play Preference Optimization method based on the Meta-Llama-3-8B-Instruct architecture.

Gemma is an advanced open-source model trained on high-quality datasets, supporting different context length requirements.

GLiClass is an efficient zero-shot classifier trained on synthetic data, suitable for topic classification, sentiment analysis, and reranking tasks in RAG workflows.

This model is based on the T5 architecture, specifically designed to correct spelling and punctuation errors in French text.

BERT-based Named Entity Recognition (NER) context reranking model for evaluating the helpfulness of contextual sentences for NER predictions

An agricultural disease detection model fine-tuned based on Qwen-VL-chat, specializing in disease identification and treatment recommendations for three major crops: rice, corn, and wheat.

A Russian and English spelling correction model based on the mT5-large architecture, normalizing words to correct spelling and typographical errors.

Nous Hermes 2 is a next-generation flagship 7B Hermes model based on Mistral 7B DPO, optimized with DPO and demonstrating excellent performance across multiple benchmarks.

The OpenMath model is specifically designed for solving mathematical problems by integrating textual reasoning with Python interpreter-executed code blocks. Trained on the OpenMathInstruct-1 dataset containing 1.8 million math problem-solution pairs.

A 7B-parameter causal language model compatible with Meta LLaMA 2 architecture, outperforming similar models under 33B in multiple evaluations

Phi-1.5 model fine-tuned on the Hermes dataset, primarily used for text generation tasks

A small sequence classifier trained based on the ESM-2 protein language model, capable of classifying protein sequences into three categories: enzymes, receptor proteins, and structural proteins.

This is a 13-billion-parameter language model based on the LlaMa architecture, fine-tuned using synthetic data, primarily for research purposes.

An OCR system based on Transformer architecture, specifically designed for recognizing Central Kurdish text, trained using synthetic data.

A fine-tuned printed text recognition model based on microsoft/trocr-base-printed, optimized for synthetic OCR datasets

An automatic paraphrase model based on BART-large architecture, trained on the ParaBank 2 dataset, capable of generating high-quality English sentence paraphrases

A T5-based model for correcting sentence errors in English, German, and French texts