Fara-7B is an efficient small language model specially designed by Microsoft for computer usage scenarios. It has only 7 billion parameters and performs excellently in advanced user tasks such as web operations, competing with larger agent systems.

MultimodalGguf

Fara-7B is a small language model developed by Microsoft Research, specifically designed for computer usage scenarios. It has only 7 billion parameters and achieves excellent performance among models of the same scale. It can perform computer interaction tasks such as web automation and multimodal understanding.

MultimodalTransformersEnglish

A three-stage training GRPO inference test model based on the microsoft/phi-4 (14B) base model, fine-tuned through the SFT→RLHF→SFT process, with excellent multi-step reasoning ability

Natural Language ProcessingTransformersMultiple Languages

DNA-R1 is a Korean-specialized reasoning model optimized from Microsoft's Phi-4, significantly enhancing Korean reasoning capabilities through reinforcement learning, excelling in mathematics, programming, and general reasoning tasks.

Natural Language ProcessingTransformersMultiple Languages

A Korean automatic speech recognition (ASR) and speech translation (AST) model fine-tuned based on microsoft/Phi-4-multimodal-instruct, demonstrating excellent performance on the zeroth-korean and fleurs datasets.

MultimodalTransformersKorean

Phi-4 is a text generation model developed by Microsoft. It is quantized and optimized through Intel auto-round technology and performs excellently in tasks such as natural language processing, mathematical reasoning, code generation, and dialogue communication.

Natural Language ProcessingTransformersMultiple Languages

Chocolatine-14B-Instruct-DPO-v1.3 is a large language model with 14 billion parameters based on the Microsoft Phi-4 model. It is fine-tuned through direct preference optimization using a French DPO dataset. It performs excellently on French and English tasks, supports a context length of 16k, and is the best Phi-4-based model on the OpenLLM leaderboard.

Natural Language ProcessingTransformersMultiple Languages

Phi-3.5-mini-instruct is an efficient small language model developed by Microsoft, optimized using advanced quantization techniques and designed specifically for NPU deployment. This model performs excellently in text generation tasks and supports natural language processing and code-related scenarios.

Natural Language ProcessingTransformersOther

Phi-1.5-Tele is a language model specifically designed for the telecommunications field. Based on Microsoft's phi-1.5 model, it has been continuously pre-trained on a large-scale telecommunications dataset and performs excellently in telecommunications benchmark tests. At the same time, it maintains comparable performance to the original model in benchmark tests such as common sense, language understanding, and logical reasoning.

Natural Language ProcessingSafetensorsMultiple Languages

A vision-language model based on Microsoft Florence-2-large, excelling in image understanding and text generation tasks

MultimodalTransformersMultiple Languages

WizardLM-2 7B is a highly efficient large language model developed by Microsoft AI team, based on the Mistral-7B architecture, excelling in multilingual, reasoning, and agent tasks.

Natural Language ProcessingTransformers

WizardLM-2 8x22B is the next-generation state-of-the-art large language model developed by Microsoft AI, featuring a Mixture of Experts (MoE) architecture, excelling in complex dialogue, multilingual capabilities, reasoning, and agent tasks.

Natural Language ProcessingTransformers

A document understanding model fine-tuned based on microsoft/layoutlmv3-base, excelling in document layout analysis tasks

MultimodalTransformers

A deepfake audio detection model fine-tuned based on Microsoft's WavLM-base, achieving excellent performance on the ASVspoof 2019 dataset with an accuracy of 99.79%

Audio ProcessingTransformers

EEVE-Korean-Instruct-2.8B-v1.0 is a large language model fine-tuned for Korean instructions based on the Microsoft Phi-2 architecture, with 2.8 billion parameters. This model is specifically optimized for Korean understanding and generation tasks and performs excellently on multiple evaluation benchmarks.

Natural Language ProcessingTransformersOther

EEVE-Korean-2.8B-v1.0 is a Korean large language model extended from the microsoft/phi-2 model by expanding Korean vocabulary and fine-tuning, and it performs excellently in Korean tasks. This model adopts a seven-stage training process and parameter freezing technology, effectively expanding its Korean vocabulary ability.

Natural Language ProcessingTransformersOther

Phi 2 - llamafile is the llamafile format version of the Microsoft Phi-2 model, offering multiple quantization methods and supporting operation on various operating systems. This model has 2.7 billion parameters and performs excellently in benchmark tests such as common sense, language understanding, and logical reasoning, approaching the current optimal level.

Natural Language ProcessingTransformersMultiple Languages

A visual question answering model fine-tuned on the textvqa dataset based on microsoft/git-base-textvqa, excelling at handling image-based question answering tasks involving text

Natural Language ProcessingTransformersOther

An image classification model fine-tuned based on microsoft/resnet-18, performing excellently on image folder datasets.

Computer VisionTransformers

This is a document understanding model fine-tuned on the CORD dataset based on the Microsoft LayoutLMv3-base model. It is specifically designed for extracting text and layout information from receipt documents and has achieved excellent performance on the evaluation set (F1 score of 0.9541).

MultimodalTransformers