AI Image2026
Performs instruction-based image editing from reference images using a 4B native-resolution diffusion transformer; the Turbo variant uses 4-step distillation for interactive latency (≈1.02 s per 1024² edit on A100) while supporting semantic, appearance, structure-aware and restoration edits.
AI Model2026
4-bit NVFP4 (W4A4) quantized pack of Upstage Solar Open2 250B for vLLM serving on NVIDIA Blackwell GPUs, preserving MoE routing and near-BF16 quality while cutting model size from 500.6 GB to 153.3 GB.
AI Model2026
Provides a 2‑bit quantized build of Qwen3.6‑35B‑A3B for local serving via an OpenAI‑compatible HTTP API. Key features: 12.3 GB on disk, eschamoe mixed 2/3‑bit expert quantization with int8 dense layers, runs on a single 16–24 GB NVIDIA GPU and ships with Escha SGLang and ZML runtimes.
Performs full-parameter post-training of trillion-parameter MoE DeepSeek-V4 models on an Ascend NPU SuperPOD, using a hierarchical optimization of model parallelism, communication orchestration, and kernel execution to increase Model FLOPs Utilization. Also builds CPT/SFT pipelines with solver-verified synthetic data for Operations Research, reporting strong zero-shot Pass@1 results.
AI Model2026
A 124B hybrid-linear Mixture-of-Experts language model optimized for instruction following, long-context reasoning and agentic workflows, activating ~5.1B parameters per token. Key features include a 256K native context (extendable to 1M), alternating KDA/MLA attention layers, and vLLM/SGLang inference support.
AI Model2026
A sparsely activated Mixture-of-Experts (MoE) causal language model with 16B total parameters and 2.8B active parameters per token, released with end-to-end checkpoints and training recipe; trained on AMD Instinct GPUs and licensed for research use.
AI Audio2026
Multilingual, real-time ASR for edge CPUs that uses heterogeneous quantization to reduce model size (4.62→1.58 GB) and lower inference latency. Trades some accuracy for 1.6–2.3× faster inference vs. Whisper.cpp and real-time capability on a few CPU threads, making it suitable for memory- and compute-constrained on-device transcription.
AI Model2026
Generates text from text, image, or audio inputs using a native multimodal, Mixture-of-Experts autoregressive transformer (276B total / 12B active) with up to 1M-token context; targeted at conversational, agentic, coding and multimodal applications.
AI Model2026
Open-weight multimodal Mixture-of-Experts LLM with native vision and a 1,048,576-token context window. 2.8T parameters (104B activated), MXFP4 quantization, released for agentic long-horizon coding, knowledge work, and vision-in-the-loop workflows.
AI Model2026
GGUF-quantized build of Moonshot AI's Kimi K3 for local inference: MXFP4-aware quantization, image-text-to-text pipeline support, native vision and a 1,048,576-token context window. Intended for local GGUF runtimes (vLLM, SGLang, TokenSpeed) with Kimi K3 license constraints.
AI Video Papers2026
Real-time streaming multimodal foundation model that uses a codec-native tokenizer (Mage-ViT) to encode motion- and residual-rich regions from video I/P frames, reducing visual token usage by over 75% and enabling up to ~3.5× wall-clock inference speedup after training on ~560M images and 100M video frames.
AI Model2026
A 350M-parameter multilingual bidirectional masked-language encoder with an 8,192-token context window, intended for fine-tuning on classification, token-level tasks, retrieval/reranking and semantic-similarity; optimized for long-context CPU inference and on-device use.