GLM-4.5-Air-REAP-82B-A12B is an efficiently compressed version of GLM-4.5-Air. Through the REAP pruning technology, the parameter scale is compressed from 106B to 82B, reducing the memory requirement by 25% while maintaining nearly lossless performance.

Natural Language ProcessingTransformersEnglish

MobileLLM-Pro is a 1-billion-parameter efficient device-side language model launched by Meta, optimized for mobile devices, supporting a context length of 128k and providing high-quality inference capabilities. This model is trained using knowledge distillation technology, outperforming models of the same scale in multiple benchmark tests, and supporting nearly lossless 4-bit quantization.

Natural Language ProcessingTransformersEnglish

Sugoi 14B Ultra is a quantized version optimized based on the previous Sugoi 14B model. It performs excellently in Japanese-to-English translation tasks, with the BLEU score nearly doubling compared to the previous quantized version. It is particularly good at handling bracketed text in RPG Maker projects.

Natural Language ProcessingGgufMultiple Languages

An efficient inference model fine-tuned based on Qwen 30B-A3B (MOE), achieving nearly double the speed improvement by reducing the number of experts, supporting multiple quantization formats and a 40K context length.

Natural Language ProcessingTransformers

This model is distilled from all-MiniLM-L12-v2, achieving nearly 2x faster inference speed while maintaining high accuracy on both CPU and GPU.

Natural Language ProcessingSafetensorsMultiple Languages

The Linglong Linear Preview is a hybrid linear sparse large language model open-sourced by InclusionAI, with a total of 17.1B parameters and 3.0B activated parameters. This model implements long-text reasoning based on a hybrid linear attention mechanism, achieving near-linear computational complexity and near-constant space complexity during inference.

Natural Language ProcessingSafetensorsMultiple Languages

IndicF5 is a near-human level multilingual text-to-speech (TTS) model supporting 11 Indian languages.

Audio ProcessingSafetensorsOther

Gemma 3 is a lightweight open-source multimodal model series by Google, supporting text and image inputs with text generation capabilities. This version is a 27B parameter instruction-tuned model using quantization-aware training, offering lower memory requirements while maintaining near-original quality.

MultimodalGguf

IndicF5 is a near-human-level multilingual text-to-speech (TTS) model supporting 11 Indian languages, trained on 1417 hours of high-quality speech data.

Audio ProcessingSafetensorsOther

QwQ-32B-FP8 is the FP8 quantized version of the QwQ-32B model, maintaining nearly the same accuracy as the BF16 version while supporting faster inference speed.

Natural Language ProcessingTransformers

Light-R1-32B-DS is a near-SOTA level 32B mathematical model, fine-tuned based on DeepSeek-R1-Distill-Qwen-32B, achieving high performance with only 3K SFT data.

Natural Language ProcessingTransformers

IndicF5 is a near-human multilingual text-to-speech (TTS) model trained on 1,417 hours of high-quality speech data, supporting 11 Indian languages.

Audio ProcessingSafetensorsOther

A speech recognition model optimized for chess coordinate recognition, fine-tuned using nearly 2,500 English chess coordinate audio files

Audio ProcessingTransformersMultiple Languages

ModernBertMultilingual is a multilingual model trained from scratch, supporting Chinese, English, Japanese, and Korean, excelling in mixed East Asian language text tasks.

Natural Language ProcessingSafetensorsMultiple Languages

ModernBertMultilingual is a multilingual model trained from scratch, supporting Chinese, English, Japanese, and Korean, with excellent performance in mixed East Asian language text tasks.

Natural Language ProcessingSafetensorsMultiple Languages

SenecaLLM is a model focused on the field of cybersecurity. After nearly a month (about 100 hours) of training and fine-tuning, it can think like a cybersecurity expert and answer relevant questions.

Natural Language ProcessingGgufEnglish

Chronos-Bolt is a series of pre-trained time series forecasting models that support zero-shot prediction, based on the T5 encoder-decoder architecture, trained on nearly 100 billion time series observations.

Scientific ComputingSafetensors

Chronos-Bolt is a series of pretrained time series forecasting models supporting zero-shot prediction, based on T5 encoder-decoder architecture, trained on nearly 100 billion time series observations.

Scientific ComputingSafetensors

Chronos-Bolt is a series of pretrained time series forecasting models supporting zero-shot forecasting, based on T5 encoder-decoder architecture, trained on nearly 100 billion time series data points.

Scientific ComputingSafetensors

Chronos-Bolt is a series of pretrained time series forecasting models supporting zero-shot prediction, based on T5 encoder-decoder architecture, trained on nearly 100 billion time series observations.

Scientific ComputingSafetensors