# Multi-task fine-tuning

Tooka-SBERT-V2-Small is a trained sentence transformer model for semantic text similarity and embedding tasks. It can map sentences and paragraphs to a dense vector space where semantically similar texts are close to each other.

A Sanskrit-specific model fine-tuned based on Qwen2.5-7B, optimized for Sanskrit comprehension and translation

A Portuguese large language model fine-tuned from Qwen2.5-0.5B-Instruct, specializing in text generation tasks

A natural language inference model optimized through multi-task fine-tuning based on the ModernBERT-large model, excelling in zero-shot classification and NLI tasks.

ModernBERT is a model fine-tuned on multi-task natural language inference (NLI) tasks, excelling in zero-shot classification and long-context reasoning.

Moxin 7B is a powerful open-source large language model that offers various types such as base models and chat models, and has demonstrated good performance on multiple common datasets.

GreekBART is a Greek sequence-to-sequence pre-trained model based on BART, particularly suitable for generation tasks such as summarization.

Russian universal sentence encoder, based on the sentence-transformers framework, specifically designed to extract 1024-dimensional dense vectors for Russian text

Based on the DeBERTa-v3-base model, the context length is extended to 1280, and fine-tuned for 250,000 steps on the tasksource dataset, focusing on natural language inference and zero-shot classification tasks.

A pre-trained Amharic language model based on the bert-medium architecture, with 40.5 million parameters trained on 290 million tokens, achieving performance comparable to larger multilingual models.

Yi-1.5 is an upgraded version of the Yi model, demonstrating superior performance in programming, mathematics, reasoning, and instruction-following capabilities.

Yi-1.5 is an upgraded version of the Yi model, excelling in programming, mathematics, reasoning, and instruction-following capabilities while maintaining excellent language understanding, commonsense reasoning, and reading comprehension.

Yi-1.5 is an upgraded version of the Yi model, excelling in programming, mathematics, reasoning, and instruction-following capabilities while maintaining outstanding language understanding, commonsense reasoning, and reading comprehension.

AkaLlama is a Korean large language model fine-tuned from Meta-Llama-3-70b-Instruct, focusing on multi-task practical applications

OpenELM is a set of open-source and efficient language models. It adopts a hierarchical parameter allocation strategy to improve model accuracy and includes pre-trained and instruction-tuned versions with 270 million to 3 billion parameters.

A configurable large language model fine-tuned using Configurable Safety Tuning (CST) method, supporting behavior pattern configuration through system prompts.

A model fine-tuned on a specific Turkish dataset, capable of accurately answering information in Turkish and providing strong support for Turkish-related text generation tasks.

Canary-750M is a pre-trained Turkish GPT-J 750M model, part of the Turkish Data Depository initiative.

Russian-optimized model based on FLAN T5 3b, outperforming FRED T5XL

French sentence embedding model based on CamemBERT-large, providing powerful semantic search capabilities

This model is a lightweight natural language inference model based on DistilRoBERTa, supporting zero-shot classification tasks.

A DeBERTa-v3 model specifically designed for zero-shot classification tasks, excelling in various classification tasks

A Transformer-based language model released by OpenAI, pre-trained on large-scale corpora with powerful text generation capabilities

Mamba - GPT - 3B - V4 is an outstanding 3B parameter language model that performs excellently on the Open LLM leaderboard, surpassing dolly - v2 - 12b and providing high - quality language processing capabilities.

Camel-Platypus2-70B is a large language model merged from Platypus2-70B and qCammel-70-x, based on the LLaMA 2 architecture, focusing on STEM and logical reasoning tasks.

A Tigrinya abusive content detection model fine-tuned on TiRoBERTa, trained on 13,717 YouTube comments dataset

A Bengali pre-trained model based on the sequence-to-sequence Transformer architecture, optimized for natural language generation tasks

A Portuguese BERT model specialized for the legal domain based on the BERTimbau large model, supporting sentence similarity calculation and semantic search

A legal domain-specific Portuguese sentence transformer based on the BERTimbau large model, supporting semantic similarity calculation

XLMR-MaCoCu-is is a large-scale pre-trained language model based on Icelandic text, built by further training the XLM-RoBERTa-large model and belongs to the MaCoCu project.

A language model based on large-scale pre-training of Maltese text, further trained on the XLM-RoBERTa-large foundation

MaltBERTa is a large-scale pretrained language model based on Maltese text, using the RoBERTa architecture, developed by the MaCoCu project.

BanglaT5 is a Bengali sequence-to-sequence transformer model pre-trained with Span Corruption objectives, achieving state-of-the-art performance in multiple Bengali natural language generation tasks.

KoMiniLM is a lightweight Korean language model designed to address latency and capacity limitations of large language models in practical applications.

pko-t5 is a T5 model specifically optimized for Korean, trained exclusively on Korean data using BBPE tokenization to address Korean segmentation issues.

Latvian pre-trained language model based on BERT architecture, suitable for various natural language understanding tasks

RobBERT is the current state-of-the-art Dutch BERT model, optimized based on the RoBERTa architecture, suitable for various text classification and tagging tasks

IT5 is the first family of sequence-to-sequence Transformer models specifically pretrained at scale for Italian, following the T5 model approach.

T5 Version 1.1 is Google's improved text-to-text conversion model, using the GEGLU activation function, pretrained unsupervised only on the C4 dataset, and requires fine-tuning for use.

CZERT is a language representation model specifically trained for Czech, outperforming multilingual BERT models on various Czech NLP tasks