BRIEF-DETAILS: Multilingual embedding model optimized for text representation across languages, available in various GGUF quantizations (q4-f16), recommended for q6_k/q8_0 usage
Brief-details: VRAM-80 is a model by unslothai focused on optimizing VRAM usage and efficiency, available on HuggingFace platform
Brief-details: A compact CLIP text encoder model developed by peft-internal-testing, designed for efficient text embedding and multimodal applications.
Brief-details: 8B parameter Llama 3.1 model optimized with FP8 quantization, achieving 99.52% accuracy recovery while reducing memory footprint by 50%
Brief Details: Cross-encoder model optimized for MS Marco passage ranking, offering excellent NDCG@10 (71.01) and MRR@10 (34.85) scores with 4100 docs/sec processing.
Brief Details: A model by unslothai hosted on Hugging Face, designed for collaborative AI development environments. Focus on accessibility and integration.
Brief Details: ResNet TDNN speaker verification model trained on VoxCeleb, achieving 1.05% EER. Handles speaker embedding extraction and verification using SpeechBrain.
Brief Details: Efficient cross-encoder model for MS Marco passage ranking, optimized for speed (9000 docs/sec) with NDCG@10 of 69.84 and MRR@10 of 32.56
Brief Details: GGML-optimized 13B parameter Vicuna language model (v1.1) - Obsolete version of the popular open-source chat assistant based on LLaMA architecture.
BRIEF-DETAILS: 176B parameter multilingual chat model built on BLOOM, instruction-tuned for conversation and QA across languages with advanced capabilities.
Brief-details: A Facebook-developed model focused on multi-token prediction capabilities, emphasizing privacy-compliant processing according to Meta's policies. Hosted on HuggingFace.
Brief-details: Oni_Mitsubishi_12B is a 12 billion parameter language model created by SicariusSicariiStuff, available on HuggingFace for open-source use and research purposes.
Brief-details: Experimental 12B parameter Gemma model with layerwise abliteration technique for uncensored outputs while maintaining coherent responses, featuring 90%+ acceptance rate.
Brief Details: T5XXL-Unchained is a modified T5-XXL encoder featuring an extended tokenizer for Flux uncensoring, with pending compatibility for ComfyUI and Kohya tools.
Brief-details: A 12B parameter fine-tuned version of Google's Gemma model optimized for creative freedom and reduced content restrictions, featuring 128K context window and enhanced character consistency
Brief Details: A state-of-the-art arbitrary-scale super-resolution model using neural heat fields, featuring anti-aliasing capabilities and RDN backbone architecture.
Brief-details: VisualPRM-8B is an 8B-parameter multimodal Process Reward Model that enhances reasoning capabilities of MLLMs through Best-of-N evaluation strategies
BRIEF-DETAILS: Audio Question Answering model achieving SOTA performance through reinforcement learning, based on Qwen2-Audio-7B-Instruct with GRPO optimization
Brief-details: Advanced chest X-ray interpretation model from Stanford AIMI, based on CLIP architecture with ViT-B/16 backbone, trained on LAION-2B dataset
BRIEF-DETAILS: 4-bit quantized version of Midnight-Miqu-70B-v1.5 using AWQ compression, optimized with lmdeploy v0.4.2 for efficient deployment
Brief Details: IndicTrans2 1B parameter model for high-quality Indian language to English translation, supporting all 22 scheduled Indian languages with advanced features like flash attention
