Tag

PaddleOCR-VL-1.6

Performs image-to-text document parsing and OCR for complex elements (tables, formulas, charts, seals), with multilingual support (en/zh). It uses region-aware data optimization and progressive post-training to improve weak-region supervision and is plug-and-play compatible with PaddleOCR-VL-1.5.

nvidia/Qwen3.6-35B-A3B-NVFP4

Quantized NVFP4 build of the Qwen3.6-35B MoE language model, optimized with NVIDIA Model Optimizer to cut model size and GPU memory by ~3.06× for inference. Designed for vLLM and NVIDIA GPU deployments (Hopper/Blackwell).

Cosmos3-Super-Text2Image

Generates high-fidelity images from text prompts using NVIDIA's 64B Cosmos3-Super multimodal foundation model. Integrates with Hugging Face Diffusers and vLLM‑Omni, is released under OpenMDW1.1 for commercial use, and is optimized for Physical AI workflows (robotics, AV, simulation).

LFM2.5-8B-A1B

Hybrid LFM2.5 text-generation model optimized for on-device assistants and agentic workflows — 8.3B total / 1.5B active parameters with 131,072-token context. Prioritizes low-latency, high-throughput inference and multilingual instruction-following; not optimized for pure heavy programming or knowledge-heavy QA without retrieval.

EarlyTom: Early Token Compression Completes Fast Video Understanding

Performs training-free early-stage visual token compression inside the vision encoder to cut time-to-first-token (TTFT) and FLOPs for Video-LLMs. Introduces a decoupled spatial token selection strategy and reports up to 2.65× TTFT reduction and 61% FLOPs savings on LLaVA-OneVision-7B (NVIDIA A100) while preserving full-token accuracy — aimed at latency-sensitive video understanding.

Step-3.7-Flash (GGUF quantizations)

GGUF quantizations of Step-3.7-Flash: a sparse multimodal Mixture-of-Experts LLM with native image understanding, selectable reasoning levels, and a 256K context window. Ships multiple calibrated Q3/Q4/IQ quant files plus an mmproj vision projector for local llama.cpp inference on high-memory hosts.

SANA-Streaming: Real-time Streaming Video Editing with Hybrid Diffusion Transformer

Enables real-time streaming video-to-video editing (1280×704 @24 FPS) on a single RTX 5090 GPU. Uses a Hybrid Diffusion Transformer for balanced local/global modeling, Cycle‑Reverse Regularization for temporal consistency, and system-level mixed-precision and fused kernels to maximize throughput.

Draft-OPD: On-Policy Distillation for Speculative Draft Models

Introduces Draft-OPD, an on-policy distillation method for training lightweight draft models used in speculative decoding — it focuses learning on draft-induced errors via target-assisted rollouts and replay, improving acceptance length and enabling >5× lossless LLM inference acceleration.

Colored Noise Diffusion Sampling

Reallocates injected noise energy across frequency bands to match a diffusion model's spectral bias, improving sampling fidelity without retraining. Uses a timestep- and frequency-dependent colored-noise schedule as a plug-and-play inference-time SDE solver; shows sizable FID drops on ImageNet-256.

unsloth/gemma-4-12b-it-GGUF

A GGUF-quantized, locally runnable build of Gemma 4 12B Unified (image-text-to-text) packaged by unsloth; preserves multimodal (image/audio) input support under an Apache-2.0 license and is compatible with common GGUF runtimes and Unsloth Studio.

Cosmos3-Super

Generates and reasons about multimodal physical-world content—text, images, video, audio, and robot/action trajectories—conditioned on combinations of text, image, video and action inputs. The 64B “Super” variant targets Physical AI use cases and supports vLLM‑Omni, Diffusers, and action prediction.

ByteDance/Bernini-R

Provides the renderer weights and inference code for Bernini’s video renderer, enabling text→video, image→video and video editing inference. Offers a ready diffusers-format bundle or safetensors checkpoints under Apache‑2.0; intended for multi‑GPU/Hopper inference and reproducible research.