Search Results for "checkpoint" - Page 3

Krea 2 Turbo is Krea AI’s flagship open-weight text-to-image diffusion model optimized for fast, high-quality image generation. Built on a 12-billion-parameter Diffusion Transformer (DiT), it is a distilled and post-trained version of the Krea 2 Raw checkpoint, enabling photorealistic and artistic image synthesis in as few as eight inference steps. Designed for creative professionals, developers, and researchers, it supports concept art, design exploration, marketing assets, illustrations, and commercial visual production. Unlike the Raw checkpoint, which is intended for fine-tuning and LoRA training, Turbo is optimized for direct inference, delivering native resolutions up to 2K with excellent prompt adherence and broad aesthetic diversity. ...

Krea 2 Raw is Krea AI’s base open-weight text-to-image diffusion checkpoint, designed primarily for fine-tuning, LoRA training, and post-training rather than direct inference. It is part of the Krea 2 model family and uses a 12-billion-parameter Diffusion Transformer architecture to generate images from natural-language prompts. Unlike Krea 2 Turbo, which is distilled and optimized for faster direct generation, Raw is the foundational checkpoint before additional post-training and fine-tuning. ...

mms-300m-1130-forced-aligner is a multilingual forced alignment model based on Meta’s MMS-300M wav2vec2 checkpoint, adapted for Hugging Face’s Transformers library. It supports forced alignment between audio and corresponding text across 158 languages, offering broad multilingual coverage. The model enables accurate word- or phoneme-level timestamping using Connectionist Temporal Classification (CTC) emissions. Unlike other tools, it provides significant memory efficiency compared to the TorchAudio forced alignment API. ...

...Developers can adapt it to new robotic tasks using the provided rank-32 LoRA training recipe. The checkpoint uses BF16 weights and includes its own observation history, camera configuration, preprocessing pipelines, normalization statistics, and action contract.

...The model learns to align product images and descriptive text using contrastive learning, enabling it to perform well across various fashion-related tasks without additional supervision. FashionCLIP 2.0, the latest version, uses the laion/CLIP-ViT-B-32-laion2B-s34B-b79K checkpoint for improved accuracy, achieving better F1 scores across multiple benchmarks compared to earlier versions. It supports multilingual fashion queries and works best with clean, product-style images against white backgrounds. The model can be used for product search, recommendation systems, or visual tagging in e-commerce platforms.

Mistral Large 3 675B Instruct 2512 NVFP4 is a frontier-scale multimodal Mixture-of-Experts model featuring 675B total parameters and 41B active parameters, trained from scratch on 3,000 H200 GPUs. This NVFP4 checkpoint is a post-training-activation quantized version of the original instruct model, created through a collaboration between Mistral AI, vLLM, and Red Hat using llm-compressor. It retains the same instruction-tuned behavior as the FP8 model, making it ideal for production assistants, agentic workflows, scientific tasks, and long-context enterprise systems. ...

...Training ran on TPU v6e using JAX and Pathways with privacy-preserving algorithms (DP-SGD, truncated Poisson subsampling) and DP scaling laws to balance compute and privacy budgets. Benchmarks on the 1B pre-trained checkpoint show expected utility trade-offs (e.g., HellaSwag 10-shot 39.09, BoolQ 0-shot 62.04, PIQA 0-shot 68.00), reflecting its privacy-first design.