Brief-details: Apple's 3B parameter instruction-tuned language model, optimized with layer-wise scaling and available in multiple GGUF quantizations for efficient deployment.
Brief Details: ControlNet Tile SDXL - Advanced image manipulation model supporting image deblur, variation generation, and super-resolution with SDXL integration.
Brief Details: A 7.57B parameter medical vision-language model built on Mistral-7B, specialized for biomedical image analysis and Q&A, with Apache 2.0 license.
Brief Details: A 315M parameter speech recognition model fine-tuned on Esperanto, achieving 12.31% WER on Common Voice test set. Built on wav2vec2-large-xlsr-53.
Brief Details: Medical-focused 7B parameter LLM based on Mistral, trained on PubMed data. Supports 9 languages with strong performance in medical QA tasks.
Brief-details: A 6.74B parameter multimodal chatbot based on LLaMA, capable of understanding both text and images, trained on diverse datasets including LAION and GPT-generated instructions.
Brief Details: A 22.2B parameter chat model optimized for efficient deployment through various GGUF quantization options, featuring imatrix compression methods.
Brief-details: Qwen2.5's 32B parameter instruction-tuned model quantized to 8-bit precision. Features 128K context length, multilingual support, and enhanced capabilities in coding and mathematics.
Brief-details: Norwegian BERT-based sentence embedding model with 178M parameters, optimized for semantic similarity and cross-lingual tasks between Norwegian and English.
Brief-details: A lightweight MobileNetV4 variant with 3.8M parameters, trained on ImageNet-1k. Optimized for mobile devices with 74.6% top-1 accuracy at 256px resolution.
Brief Details: 405B parameter Llama 3.1 model optimized with FP8 quantization for improved efficiency while maintaining performance, featuring specialized KV cache optimization.
Brief-details: Thai speech recognition model based on wav2vec2, trained on CommonVoice V8 dataset. Achieves 12.58% WER with newmm tokenizer and language model.
Brief-details: SE-ResNeXt model with anti-aliasing and channel attention, 93.8M params, trained on ImageNet-12k and fine-tuned on ImageNet-1k, achieving 86.72% top-1 accuracy
Brief-Details: AMD's 70B parameter Llama 3.1 variant optimized with FP8 quantization for KV cache, offering efficient deployment while maintaining near-original perplexity scores
Brief-details: A sophisticated ControlNet v1.1 model specialized in soft edge detection for image generation, built on Stable Diffusion v1.5 with improved training and artifact reduction.
Brief-details: MiniLM is a compact 33M-parameter language model that achieves BERT-level performance while being 2.7x faster. Ideal for efficient NLP tasks.
Brief-details: Facebook's wav2vec2-large-robust is a large-scale speech recognition model trained on multiple domains, optimized for 16kHz audio processing with enhanced robustness.
Brief Details: A 6.45B parameter biological foundation model using StripedHyena architecture for long-context genomic sequence modeling up to 131k tokens.
Brief Details: SmolLM2-1.7B is a compact 1.7B parameter language model trained on 11T tokens, optimized for instruction-following and reasoning tasks with BF16 precision.
Brief Details: Advanced classifier model for detecting harmful AI behaviors, based on Llama-2 13B. Achieves 93.19% average agreement with human judgments.
Brief-details: A 10.7B parameter language model fine-tuned on truthy data, showing strong performance in reasoning tasks with 76.10% average on key benchmarks.
