Brief-details: Advanced 7B parameter GPTQ-quantized Mistral model optimized for helpful, compliant responses with multiple quantization options and ChatML format support
Brief Details: A powerful 52B parameter language model with strong Chinese language capabilities, featuring innovative fine-tuning and superior performance across multiple domains.
Brief-details: Fine-tuned 3B parameter FLAN-T5-XL model specialized in multi-purpose text summarization, supporting various summary types through prompt engineering
Brief-details: ELECTRA large discriminator model by Google - powerful transformer-based language model trained to detect real vs fake tokens, optimized for efficiency and performance.
Brief Details: Pix2struct-large: 1.34B parameter image-to-text model specialized in visual language understanding across multiple domains. Supports 5 languages.
BRIEF DETAILS: T5-Large fine-tuned model for deciphering ChatGPT's encrypted responses through token length analysis, part of USENIX Security '24 research.
Brief-details: BERT-based scientific document embedding model trained on citation graphs, optimized for research paper similarity with 110M parameters and SOTA performance on SciDocs benchmarks.
Brief-details: A specialized sentence embedding model based on DistilRoBERTa, designed for code search applications with 768-dimensional vectors, trained on code_search_net dataset.
Brief-details: Multilingual T5 model supporting 101 languages, pre-trained on mC4 dataset. Small variant ideal for text-to-text generation tasks requiring lower compute resources.
Brief Details: A specialized BERT-based embedding model fine-tuned on PubMed data, producing 768-dimensional vectors for medical text similarity tasks with SOTA performance.
Brief-details: A powerful multilingual embedding model (9.24B params) based on Gemma-2 that achieves SOTA results across multiple languages and benchmarks for text embedding tasks
Brief-details: Neural machine translation model for English to Italian conversion, developed by Helsinki-NLP with BLEU scores up to 48.2 on Tatoeba dataset.
Brief-details: A powerful 7B parameter language model trained on 1.5T tokens, featuring FlashAttention and multiquery architecture. Apache 2.0 licensed.
Brief-details: 4-bit quantized Mistral-7B Instruct v0.3 optimized for efficient inference with Unsloth, offering 2.2x faster performance and 62% less memory usage.
BRIEF DETAILS: T5-based headline generation model trained on 500k articles. Specialized in creating one-line headlines from article text. Popular with 122k+ downloads.
Brief Details: Quantized version of Meta's 405B parameter LLM, optimized for 8 languages. Uses 4-bit AWQ quantization, reducing model size while maintaining performance.
Brief-details: Spanish emotion analysis model for tweets, based on RoBERTuito architecture. Detects 6 Ekman emotions + neutral class. Strong performance with 0.560 F1 score.
Brief-details: BERT-based semantic search model with 768-dimensional embeddings, trained on 500K MS MARCO query-answer pairs. Optimized for dot-product similarity scoring.
Brief-details: A 7B parameter GPTQ-quantized chat model based on Meta's Llama 2, optimized for dialogue with 4-bit precision and multiple grouping size options.
Brief Details: Baichuan2-13B-Chat is a powerful bilingual LLM trained on 2.6T tokens, optimized for both Chinese and English tasks with state-of-the-art performance.
Brief Details: Phi-1.5: Microsoft's 1.3B parameter language model excelling in code, text & reasoning. MIT-licensed, trained on 150B tokens without RLHF.
