Tag
Explore by tags
A vision-oriented foundation checkpoint for low-latency inference — DeepSeek V4 base in safetensors with FP8 optimizations. Designed for fast image generation and embedding use in inference pipelines; verify license and FP8/runtime compatibility before production use.
Base image-generation foundation model tuned for visual search and prompt-guided synthesis, intended as a compact starting point for local inference or fine-tuning. Emphasizes easy integration into image pipelines and suitability for downstream adaptation.
A 14B dense tri‑mode language model that supports autoregressive, diffusion‑based parallel decoding, and self‑speculation—designed to increase token throughput and acceptance length; best suited for researchers and engineers exploring decode‑efficiency tradeoffs on NVIDIA hardware under the Nemotron Open Model License.
A Qwen-3.6 27B model variant optimized for DFlash (speculative decoding) to reduce generation latency and increase throughput. Focuses on faster inference on serving stacks and is suitable for text-generation endpoints where lower latency and resource efficiency matter.
High-resolution vision transformers pretrained on one billion human images for human-centric tasks such as pose estimation, body-part segmentation, surface-normal and pointmap prediction. Provides multiple backbone sizes and task-specific checkpoints; released under the Sapiens2 license.
Performs task-aware generative video restoration and editing in latent video space — restoration, super-resolution, watermark and subtitle removal — adapting LTX‑2.3 with IC‑Edit/IC‑LoRA adapters to prioritize temporal consistency and occlusion-aware reconstruction.
A lightweight 'drafter' assistant for Gemma 4 31B that generates speculative token drafts to enable up-to-2× decoding speedups while preserving final output quality; compatible with Hugging Face Transformers and any-to-any pipelines.
Generates anime-style images from natural-language prompts with a full fine-tune family built on Z-Image Base — available as Base, 8-step and 4-step distillations, plus AIO and GGUF variants for 8GB/low-VRAM workflows (BF16/FP8 formats).
Open-source Mixture-of-Experts LLM designed for extremely long-context (up to 1M tokens) text generation and agentic workflows; uses a hybrid attention + MTP design to reduce KV-cache footprint while enabling 42B active parameters and FP8 mixed-precision training.
An instruct-focused LLM (104B total, 7.4B active) optimized for fast, token-efficient inference in agent workflows. Uses hybrid linear attention plus a sparse MoE to raise throughput and cut token use; suited for high-frequency production agents, with some trade-offs in very deep reasoning.
Curated multimodal training corpus for spatial intelligence: ~8.16M QA-style samples paired with ~2.72M unique images (≈1.1 TB). Provides JSONL annotations, a 1,000-sample preview, and 52 independent image archives — used to train SenseNova-SI models.
Provides the dataset and accompanying technical report for a DeepSeek project that interleaves spatial markers (points and boxes) into multimodal LLM reasoning. Includes a public subset of data and benchmarks under an MIT license; model weights are not included.