Falcon-H1 is a new family of large language models developed by the Technology Innovation Institute of the United Arab Emirates. It adopts a hybrid architecture that combines the Transformer attention mechanism and the state space model (SSMs), with excellent long-context memory ability and computational efficiency. This series of models offers various configurations with parameters ranging from 0.5B to 34B, and performs excellently in tasks such as inference, mathematics, and multilingual tasks.

Natural Language ProcessingTransformersMultiple Languages

Ovis2-1B is the latest member of the Ovis series of multimodal large language models (MLLM), focusing on structural alignment of vision and text embeddings, featuring high performance for small models, enhanced reasoning capabilities, video and multi-image processing, and multilingual OCR enhancement.

MultimodalTransformersMultiple Languages

OlympicCoder-32B is a code generation model based on Qwen2.5-Coder-32B-Instruct, employing IQ-DynamicGate ultra-low-bit quantization technology for efficient inference in memory-constrained environments.

Natural Language ProcessingGgufEnglish

Qwen2.5-7B-Instruct-1M is an instruction-tuned version based on Qwen2.5-7B, employing IQ-DynamicGate ultra-low-bit quantization (1-2 bits), suitable for efficient inference in memory-constrained environments.

Natural Language ProcessingGgufEnglish

Mistral-7B-Instruct-v0.1 is a fine-tuned model based on Mistral-7B-v0.1, supporting text generation tasks. It employs IQ-DynamicGate ultra-low bit quantization technology, making it suitable for memory-constrained deployment environments.

Natural Language ProcessingGguf

Hermes 3 3B is a small yet powerful new member of the Hermes series LLMs by Nous Research, focusing on aligning LLMs with users and empowering end-users with strong guidance and control capabilities.

Natural Language ProcessingTransformersEnglish

MobiLlama-1B-Chat is an instruction-following model fine-tuned from MobiLlama-1B, specifically designed for resource-constrained devices, emphasizing efficiency, low memory footprint, and fast response.

Natural Language ProcessingTransformersEnglish

FalconLite is a quantized version of the Falcon 40B SFT OASST-TOP1 model, capable of processing long input sequences up to 11K tokens while reducing GPU memory consumption to one-fourth of the original. Through 4-bit GPTQ quantization and adaptive dynamic NTK rotation embedding technology, it strikes a balance between latency, accuracy, and memory efficiency.

Natural Language ProcessingTransformers

This is a voice conversion model based on RVC (Retrieval-Based Voice Conversion) technology, specifically trained for 1000 epochs to emulate the voice of BTS member Jungkook.

Audio ProcessingTransformers

ALBERT XLarge v2 is an English pretrained model based on the Transformer architecture, employing parameter-sharing mechanisms to reduce memory usage, trained with masked language modeling and sentence order prediction objectives.

Natural Language ProcessingTransformersEnglish