Bee-8B is an advanced fully open-source multimodal large language model that focuses on data quality and aims to narrow the performance gap with proprietary models. By using the high-quality Honey-Data-15M corpus and the advanced data processing pipeline HoneyPipe, it demonstrates excellent performance in complex reasoning and other aspects.

MultimodalTransformersEnglish

This is the quantized version of the YanoljaNEXT-Rosetta-12B-2510 model from Yanolja. It is quantized using the llama.cpp tool and the imatrix dataset, offering multiple quantization types to adapt to different hardware environments, significantly reducing resource consumption while maintaining model quality.

Natural Language ProcessingGguf

Wan2.2 is a significant upgrade version of the basic video model. It introduces the Mixture of Experts (MoE) architecture, incorporates carefully curated aesthetic data, and is trained on larger datasets to enhance the ability to generate complex motions. This model supports generating 5-second videos with 480P and 720P resolutions, and has significant improvements in video generation quality and performance.

MultimodalDiffusers

This is a Swahili-English translation model fine-tuned based on the opus-mt-mul-en model of Helsinki-NLP. The model is specifically optimized and trained for the children's hotline service scenario, trained with synthetic hotline dialogue data, equipped with a mechanism to prevent hallucination generation, and the translation quality is guaranteed by monitoring the BLEU score through an early stopping strategy.

Natural Language ProcessingTensorboardMultiple Languages

Zonos-v0.1 is a leading open-source text-to-speech (TTS) model, trained on over 200,000 hours of multilingual speech data, comparable in expressiveness and quality to top TTS providers. It supports zero-shot voice cloning, multilingual synthesis, and fine-grained audio control.

Audio ProcessingOnnx

Discord-Hermes-3-8B is a large language model fine-tuned based on NousResearch/Hermes-3-Llama-3.1-8B, specifically optimized for conversation scenarios. Through training on a carefully curated Discord conversation dataset, it improves conversation ability and the quality of chat interaction.

Natural Language ProcessingTransformers

A speculative decoding model specifically designed for Qwen/Qwen3-8B. It uses the EAGLE-3 algorithm to improve the efficiency and quality of text generation and obtains excellent generalization ability through training on multiple high-quality datasets.

Natural Language ProcessingTransformers

Whisper is OpenAI's state-of-the-art automatic speech recognition (ASR) and speech translation model, trained on over 5 million hours of labeled data with strong zero-shot generalization capabilities. The Turbo version is a pruned and fine-tuned variant of the original, reducing decoder layers from 32 to 4, significantly improving speed with a slight quality trade-off.

Audio ProcessingTransformersMultiple Languages

A model for automatically assessing the quality of text data in both natural and programming languages, offering both unified and dual-model solutions.

Natural Language ProcessingTransformersMultiple Languages

Automatically assesses text data quality for both natural and programming languages, offering both unified and dual-model solutions.

Natural Language ProcessingTransformersMultiple Languages

A model for automatically evaluating the quality of text data in both natural and programming languages, offering both unified and independent model solutions.

Natural Language ProcessingTransformersMultiple Languages

A text-to-image model trained on a small-scale curated dataset through strategic data augmentation methods, significantly improving generation quality

MultimodalSafetensors

Zonos-v0.1 is a leading open-weight text-to-speech model trained on over 200,000 hours of multilingual speech data, delivering expressiveness and quality comparable to or even surpassing top-tier TTS service providers.

Audio ProcessingSafetensors

DeQA-Score-Mix3 is a no-reference image quality assessment model fine-tuned based on the MAGAer13/mplug-owl2-llama2-7b foundation model, demonstrating excellent performance across multiple datasets.

MultimodalTransformersEnglish

An image-to-video generation model fine-tuned on tens of millions of video data, achieving SBS-level quality comparable to CogVideoX-5B

Computer VisionDiffusersEnglish

Gemma-2-9b-it-TR-DPO-V1 is a Turkish language optimized model fine-tuned based on gemma-2-9b-it. It is trained with a synthetically generated preference dataset, which improves the format and content quality of Turkish language output.

Natural Language ProcessingSafetensorsOther

Magnum v4-22B is a large language model fine-tuned based on Mistral-Small-Instruct-2409, specifically dedicated to replicating the text quality of the Claude 3 series models (especially Sonnet and Opus). This model provides a GGUF quantized version, supports a 32K context length, and is trained on multiple high-quality datasets.

Natural Language ProcessingTransformersEnglish

Mini Magnum is a streamlined version of magnum-72b-v1, aiming to replicate the text quality of the Claude 3 series (especially Sonnet and Opus). It is fine-tuned based on Mistral-Nemo-Base-2407, and a new general instruction dataset is added to improve coherence and alignment.

Natural Language ProcessingTransformersMultiple Languages

Phi-3-mini-4k-instruct-graph is a fine-tuned version of Microsoft's Phi-3-mini-4k-instruct, specifically designed for entity relationship extraction from general text data, aiming to achieve comparable quality and accuracy to GPT-4 in generating entity relationship graphs.

Natural Language ProcessingTransformersEnglish

A language model fine-tuned based on Meta-Llama-3-8B-Instruct. It improves data quality through GPT-4o and focuses on enhancing Russian language capabilities. In the MT-Bench evaluation, its Russian score exceeds that of GPT-3.5-turbo.

Natural Language ProcessingTransformers