Brief Details: A lightweight ResNet model specialized for bean classification, developed by fxmarty. Ideal for testing and educational purposes in agricultural computer vision tasks.
Brief-details: Optimized 70B parameter LLM using FP8 quantization, achieving 99.8% accuracy recovery while reducing model size by 50% and improving inference speed up to 3x.
Brief-details: Test version of PatchTST time series forecasting model from IBM Research, referencing the official pretrained version available on Hugging Face.
Brief Details: Indic-Gemma-7B fine-tuned on 15 Indian languages + English, using 650K instruction samples. Built for multilingual capabilities using LoRA techniques.
Brief Details: DaViT Base - Vision Transformer with dual attention mechanism, 88M params, trained on ImageNet-1k. Achieves 84.63% top-1 accuracy.
Brief-details: Tiny-mbart is a lightweight version of mBART for multilingual translation tasks, created by sshleifer for testing and development purposes. Ideal for rapid prototyping and educational contexts.
Brief-details: 4-bit quantized version of DeepSeek-R1-Distill-Qwen-32B, a powerful reasoning model distilled from DeepSeek-R1, optimized for math and code tasks
Brief-details: State-of-the-art 8B parameter LLaMA-3-based model using iterative DPO training, outperforming larger models on key benchmarks including MT-Bench and Alpaca-Eval-V2
Brief Details: Fine-tuned speaker segmentation model achieving 18.28% DER, optimized for English CallHome dataset with enhanced detection capabilities
BRIEF-DETAILS: A 7B parameter bilingual Japanese-English instruction-tuned LLM based on Mistral architecture, achieving SOTA performance on Japanese benchmarks
BRIEF-DETAILS: A GGUF quantized version of Qwen2.5 14B model offering multiple quantization options (Q2-Q8) with file sizes ranging from 5.9GB to 15.8GB, optimized for different performance/quality tradeoffs.
BRIEF DETAILS: Compact testing model (<10MB) based on Falcon architecture, developed by fxmarty for alibi experimentation and lightweight applications.
Brief Details: Powerful 32B parameter RWKV-based language model converted from Qwen2.5, offering competitive performance with efficient linear attention and supporting 16K context length.
BRIEF-DETAILS: 8B parameter Llama-3 instruction-tuned model optimized with GGUF format, featuring 1048k context window and gradient-based improvements
Brief Details: TAIDE-LX-7B-Chat-4bit is a 4-bit quantized 7B parameter chat model from TAIDE, optimized for efficient deployment while maintaining performance.
Brief Details: OpenChat 3.5 0106 GGUF - High-performance 7B parameter model optimized for CPU/GPU inference, outperforms ChatGPT (March) in benchmarks
Brief-details: A SoftVC VITS singing voice conversion model trained on Genshin Impact character voices, enabling voice synthesis and conversion for game-like vocals.
Brief Details: Scientific sentence embedding model trained on 4.3M co-citation pairs, optimized for academic text similarity with strong performance in biomedical domains.
Brief-details: BioLinkBERT-base is a biomedical NLP model that leverages document citation links during pretraining, achieving SOTA performance on BLURB and MedQA benchmarks.
BRIEF-DETAILS: LinkBERT-base: BERT-like transformer model leveraging document links for enhanced language understanding, outperforming BERT on QA tasks.
BRIEF DETAILS: Advanced BERT variant leveraging document links and hyperlinks for enhanced NLP tasks. Achieves superior performance on QA and text classification tasks.
