# Multi-Dataset Training

TIPO is a text-to-image prompt optimization system based on text pre-sampling, which enhances the quality and usability of generative models by optimizing user input prompts through large language models.

ViTPose is a human pose estimation model based on Vision Transformer, achieving outstanding performance on benchmarks like MS COCO through simple architectural design.

A Hindi text-to-speech model trained from scratch based on the F5 architecture, developed by the SPRING Lab at the Indian Institute of Technology Madras.

TIPO is a 500-million-parameter model based on the LLaMA architecture, specifically designed for prompt optimization in text-to-image generation.

distilvit is an image-to-text model based on a VIT image encoder and a distilled GPT-2 text decoder, capable of generating textual descriptions of images.

A table structure recognition model developed based on Microsoft's Table Transformer for table detection and recognition tasks in documents

A high-performance cross-encoder model optimized for Japanese text reranking tasks, featuring a 24-layer architecture with 1024 hidden units

This is a Japanese Reranker (Cross-Encoder) model for text ranking tasks, featuring 24 layers and a hidden layer size of 1024.

This is a Japanese-trained Reranker (Cross-Encoder) model for text ranking tasks.

This is a Japanese-trained Reranker (Cross-Encoder) model for text ranking tasks.

PairRM is an efficient pairwise reward model for comparing and ranking output candidates from large language models, supporting various applications such as RLHF and Best-N sampling.

This model is trained using the Cross-Encoder class from SentenceTransformers to predict the semantic similarity score between two sentences.

This is a sentence embedding model based on the MiniLM architecture, capable of mapping text to a 384-dimensional vector space, suitable for semantic search and text similarity tasks.

A document image binarization model fine-tuned based on the SegFormer-B3 architecture, excelling in DIBCO evaluation metrics

This reward model is trained to predict which generated answer humans would prefer for a given question. Suitable for QA evaluation, RLHF reward scoring, and toxic answer detection.

An ONNX model based on sentence-transformers that maps text to a 384-dimensional vector space, suitable for semantic search and clustering tasks.

A BERT-base-cased based QA pair evaluation model for determining semantic relevance between questions and answers