BRIEF DETAILS: BLIP visual question-answering model with 385M parameters, combining vision-language understanding for image-based Q&A tasks. Built by Salesforce with PyTorch support.
Brief-details: Distil-large-v3 is a 756M parameter English speech recognition model, 6.3x faster than Whisper large-v3 with comparable accuracy and optimized for long-form transcription.
Brief Details: DPT-Hybrid (MiDaS 3.0) is Intel's state-of-the-art monocular depth estimation model using Vision Transformers, trained on 1.4M images with impressive zero-shot capabilities.
Brief-details: Creative writing control vectors for fine-tuned text generation control, supporting multiple LLMs with GGUF format compatibility and advanced axis manipulation
Brief-details: Qwen1.5-1.8B is a 1.84B parameter transformer-based language model, part of Qwen2's beta release, featuring 32K context length and improved multilingual capabilities.
Brief Details: CodeT5+ 110M embedding model for code understanding - generates 256-dim code embeddings with strong performance on code retrieval tasks
BRIEF DETAILS: Powerful multilingual speech encoder (580M params) supporting 96 languages, pre-trained on 4.5M hours of audio data. Ideal for feature extraction and ASR tasks.
BRIEF-DETAILS: Efficient vision-language model (1.87B params) for edge devices, capable of VQA tasks with strong benchmark performance and Apache 2.0 license
Brief Details: A powerful 6B parameter language model trained on The Pile dataset, offering strong performance in text generation and NLP tasks with public availability.
Brief-details: A specialized BERT model for chest X-ray radiology, achieving SOTA results in radiology NLI tasks with improved vocabulary and multi-modal capabilities.
Brief Details: A Chinese BERT model utilizing Whole Word Masking, developed by HFL team. Features improved masking strategy for Chinese text and 256K+ downloads. Apache 2.0 licensed.
Brief Details: A lightweight 160M parameter LLaMA-like model trained on Wikipedia and C4 datasets, designed for speculative inference acceleration research.
Brief Details: ATTACK-BERT is a specialized cybersecurity language model that uses sentence transformers to analyze and compare attack-related text, with strong focus on semantic similarity.
Brief Details: GLM-4V-9B is a powerful 13.9B parameter multimodal LLM with high-resolution image understanding capabilities and superior performance in Chinese/English tasks.
Brief Details: A traditional Chinese BERT model specialized in word segmentation tasks, part of CKIP Lab's NLP toolkit. Features GPL-3.0 license and extensive downloads (257k+).
Brief-details: GPTQ-quantized version of StarCoder2-15B optimized for code generation. Achieves 77.4% pass@1 on HumanEval-Python, using Alpaca instruction format.
**Brief Details:** A 2.7B parameter GPT-style language model trained on The Pile dataset, capable of sophisticated text generation with strong performance on various NLP tasks.
Brief Details: German BERT model for sentiment analysis with 109M params. Achieves 96.39% F1 score across datasets. Supports positive/negative/neutral classification.
Brief Details: BERT-large model fine-tuned for Named Entity Recognition, achieving 91.7% F1 score on CoNLL-2003. Identifies LOC, ORG, PER, MISC entities.
Brief Details: Turkish NER model using BERT, trained on MilliyetNER dataset. Achieves 96% F1-score for entity recognition. Supports person, location, and organization detection.
Brief-details: Multilingual text embedding model supporting 94+ languages, fine-tuned with instructions. 560M parameters, optimized for retrieval and classification tasks.
