# Long Context Processing

InternVL3-78B is an advanced multimodal large language model developed by OpenGVLab, demonstrating exceptional comprehensive performance. Compared to its predecessor InternVL 2.5, it possesses stronger multimodal perception and reasoning capabilities, extending its abilities to new domains such as tool usage, GUI agents, industrial image analysis, and 3D visual perception.

InternVL3-2B-Instruct is a supervised fine-tuned version based on InternVL3-2B, undergoing native multimodal pretraining and SFT processing, equipped with powerful multimodal perception and reasoning capabilities.

A code-reasoning large language model fine-tuned based on DeepSeek-R1-Distilled-Qwen-1.5B, utilizing distributed reinforcement learning technology to extend long-context processing capabilities

A next-generation language model based on the Qwen 2.5 14B architecture, specifically optimized for mathematical reasoning, programming, and general logical tasks.

A French language model pre-trained on 1 trillion high-quality French texts, the French version of ModernBERT

A compact language model based on the Llama architecture, supporting English and Portuguese, with 96 million parameters and a context length of 4096 tokens.

The current V3-0324 model is the best-performing quantized version in its size category, significantly reducing volume while maintaining performance close to Q8_0

Granite-3.2-2B-Instruct is a 2-billion-parameter long-context AI model, fine-tuned for cognitive reasoning capabilities, supporting 12 languages and multitasking.

Granite-3.2-8B-Instruct is an 8-billion-parameter long-context AI model, specifically fine-tuned for cognitive reasoning capabilities, supporting multiple languages and tasks.

mmMamba-linear is the first pure decoder multimodal state space model to achieve quadratic-to-linear distillation with moderate academic computing resources, featuring efficient multimodal processing capabilities.

A multilingual BERT model developed by the Algomatic team, supporting mask-filling tasks with an 8192 context length and a vocabulary of 151,680.

A modern Russian version of a unidirectional and bidirectional encoder Transformer model, pre-trained on approximately 2 trillion tokens of Russian, English, and code data, with a context length of up to 8,192 tokens.

Phi-4-Model-Stock-v2 is a large language model merged from multiple Phi-4 variant models using the model_stock merging method, demonstrating strong performance across multiple benchmarks.

Qwen2-VL-2B-Instruct is a multimodal vision-language model that supports interaction between images and text, suitable for image understanding and generation tasks.

An HTML pruning model designed for RAG systems where HTML is more suitable than plain text for modeling retrieval results

The Jais series is a comprehensive English-Arabic bilingual large language model, optimized for Arabic while maintaining strong English capabilities. This model has been fine-tuned for instruction following, making it suitable for conversational scenarios.

The Jais series is a large English-Arabic bilingual language model specifically optimized for Arabic, with strong English capabilities and 670 million parameters

Jais is a bilingual large language model family specifically optimized for Arabic, with strong English capabilities, ranging from 590 million to 70 billion parameters

The Jais series is a bilingual large language model based on the Llama-2 architecture, specifically optimized for Arabic while maintaining strong English capabilities. This model is a 70-billion-parameter Arabic-adapted version, supporting a context length of 4,096 tokens.

Phi-3-Vision-128K-Instruct is a lightweight, cutting-edge open multimodal model supporting a 128K token context length, focusing on high-quality reasoning in text and visual domains.

Phi-3 Mini 128K Instruct is a 3.8B parameter lightweight open-source model focused on reasoning capabilities, supporting 128K context length.

A 7B-parameter language model merged from Mistral 7B v0.1, combining multiple fine-tuned models via task arithmetic, supporting 32K context length, balancing creativity and reasoning

xLAM-v0.1 is a major upgrade in the Large Action Model series, fine-tuned across a wide range of agent tasks and scenarios while maintaining the original model's capabilities with the same parameter count.

A 120B hybrid large language model generated by interleaving fusion of miqu-1-70b-sf layers using the mergekit tool based on miqu-1-70b

Pruned version based on alnrg2arg/blockchainlabs_7B_merged_test2_4, integrating 7B-parameter large language models from mlabonne/NeuralBeagle14-7B and udkai/Turdus

Deepmoney is a large language model focused on the financial investment domain, trained on high-quality research reports and financial knowledge, aiming to provide professional investment analysis and decision-making support.

A 7-billion-parameter large language model fine-tuned on Intel Gaudi 2 processors based on Mistral-7B, aligned using DPO method, suitable for various language tasks

A large language model fine-tuned based on Mistral-7B-v0.1, incorporating Self-RAG technology and supporting a 32k context length

LongLoRA is an efficient fine-tuning technique for large language models with long context processing capabilities, achieving this through shifted short attention mechanisms, supporting context lengths from 8k to 100k.

Open LLM Search is a specialized adaptation of Together AI's llama-2-7b-32k model, specifically built for extracting information from web pages.

XLNet is a model pre-trained on English language using generalized permutation language modeling objectives and Transformer-XL architecture, achieving SOTA results on multiple language tasks.

XLNet is an unsupervised language representation learning method based on a generalized permutation language modeling objective, using Transformer-XL as the backbone model, excelling in long-context tasks.