VibeThinker-1.5B is a 1.5-billion-parameter dense language model launched by Weibo AI. It is fine-tuned based on Qwen2.5-Math-1.5B and is specifically designed for mathematical and algorithmic coding problems. Trained using the 'Spectrum to Signal Principle' framework, it outperforms larger models in multiple math competition tests. The training cost is approximately $7,800, and it supports an output of up to about 40k tokens.

Natural Language ProcessingTransformersMultiple Languages

This is a Transformer model published on the Hugging Face model hub. The model card documentation is automatically generated, and currently lacks specific detailed information about the model.

Natural Language ProcessingTransformers

LFM2-8B-A1B is an 8-bit quantized MLX build version optimized for Apple Silicon chips, adopting the Mixture of Experts (MoE) architecture. It has approximately 8 billion total parameters, with about 1 billion parameters activated per token, and supports fast inference on the device side.

Natural Language ProcessingMlxEnglish

KaniTTS is a high-speed, high-fidelity text-to-speech model optimized for real-time conversational artificial intelligence applications. The model uses a two-stage processing flow, combining a large language model and an efficient audio codec. On an Nvidia RTX 5080, the latency for generating 15 seconds of audio is only about 1 second, and the MOS naturalness score reaches 4.3/5. It supports multiple languages such as English, Chinese, and Japanese.

Audio ProcessingTransformersMultiple Languages

MobileLLM-R1 is a series of efficient inference models launched by Facebook, focusing on solving mathematical, programming, and scientific problems. This model has achieved excellent performance in multiple benchmark tests with only about 2T high-quality labels for pre-training.

Natural Language ProcessingTransformersEnglish

The Japanese Receipt Vision-Language Model lfm2-450M is a vision-language model specifically designed for understanding and processing Japanese receipts. It is built on LiquidAI's LFM2-VL-450M base model, capable of analyzing receipt images, extracting structured information, answering questions about the receipt content, and providing detailed descriptions in Japanese and English.

MultimodalTensorboardMultiple Languages

The W4A16 quantized version of Qwen3-30B-A3B-Thinking-2507, using LLM-Compressor for 4-bit weight and 16-bit activation quantization. Compared with the FP16 version, the memory usage is reduced by about 75%, and faster inference speed is achieved on supported hardware. This model has excellent reasoning ability, tool invocation ability, and long context support.

Natural Language ProcessingTransformers

This is a version of Qwen3-30B-A3B-Instruct-2507 quantized with W8A8 (8-bit weights and activations) based on LLM-Compressor. Compared with the FP16 version, the memory usage is reduced by about 50%, and faster inference speed is achieved on supported hardware, which is particularly suitable for Ampere and older GPU architectures.

Natural Language ProcessingTransformers

This is a Transformers model that has been published to the Hugging Face model hub. Specific information needs to be further supplemented. The model card is automatically generated and contains basic information about the model, usage instructions, training details, etc.

Natural Language ProcessingTransformers

This is a model specifically designed for gene/protein entity recognition, capable of accurately identifying and extracting information about gene and protein mentions in biomedical texts. The model is trained on the BC2GM dataset, featuring high precision and domain-specific optimization, providing reliable support for clinical and research applications.

Natural Language ProcessingTransformersEnglish

This is a Transformers model uploaded to the Hugging Face Hub. Specific information is to be supplemented by the developer. The model card is automatically generated and contains basic framework information about the model.

Natural Language ProcessingTransformers

An experimental diffusion model based on FLUX.1-Kontext-dev, focusing on image-to-image conversion tasks. Quantization processing has been completed, significantly improving inference performance. The video memory usage is about 10GB, and only 6 - 8 steps are required for inference.

Computer VisionDiffusers

LoRA trained on the Wan2.1 14B I2V 480p model for generating smooth overhead view videos rising above the subject

Computer VisionDiffusersEnglish

This is a transformers model hosted on Hugging Face Hub, with no detailed information provided about its specific functions and uses.

Natural Language ProcessingTransformers

A Q&A model fine-tuned based on Warren Buffett's shareholder letters from 1998-2024, specifically answering questions about his investment philosophy and decision logic

Natural Language ProcessingTransformersEnglish

AuriStream-1B is a biologically inspired GPT-style autoregressive Transformer model specifically designed to predict cochlear markers in long speech contexts. The model uses a long context window of approximately 20 seconds (4096 tokens) and was trained on the LibriLight dataset (approximately 60,000 hours) for about 500,000 steps. It can learn rich time-aligned representations and generate speech continuations.

Audio ProcessingTransformersEnglish

This is a 12B-parameter language model merged using the DARE TIES method, combining multiple pre-trained models to provide powerful text generation capabilities.

Natural Language ProcessingTransformers

BiRefNet is a bilateral reference framework model for high-resolution binary image segmentation, focusing on tasks such as background removal and mask generation, supporting image processing at resolutions of 2048x2048 and above.

Computer VisionSafetensors

A generative model fine-tuned on an unknown dataset. Specific information about the training data and usage is missing.

Natural Language ProcessingTransformers

FlowerVLA is a vision-language-action flow model pre-trained on the CALVIN D dataset, employing an efficient flow-matching architecture that achieves general-purpose robot operation strategies with only about 1 billion parameters.

MultimodalSafetensorsEnglish