Analyzes internal computation of text-to-image diffusion transformers and shows structural template tokens act as implicit semantic registers that maintain object identity during denoising. Introduces a causal interpretability framework (attention decomposition + targeted interventions) and a training-free pruning rule that cuts ~20% attention FLOPs for a ~1.4-point GenEval drop.
AI Model2026
Efficient 4B native-resolution diffusion foundation model for text-to-image generation and instruction-based image editing. Uses a lightweight Mage‑VAE tokenizer and a 4B NR‑MMDiT backbone to produce 512–2048 outputs with low memory and fast inference; ships in Base, RL-aligned and few-step Turbo variants.
AI Model2026
Provides GGUF-format quantized shards of Laguna S 2.1 for local or self-hosted inference—packaged for llama.cpp/llama-server and usable with vLLM/Transformers runtimes; targeted at long-context, agentic coding workloads.
AI Model2026
1B-parameter text-generation model tuned for conversational and agentic workflows with a focus on security and vulnerability-detection; suited for low-cost or on-prem/edge deployments and terminal-agent integrations.
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
Compact 3B-scale agentic LLM for multi-step tool use and reasoning, using a Looped Transformer to increase capacity without adding parameters; built for local deployment with configurable "thinking" modes and benchmark gains vs larger open models.
AI Model2026
A 250B-parameter mixture-of-experts LLM that activates 15B parameters per token to lower inference cost for agentic tasks—tool calling, long-context reasoning, and coding. Uses a hybrid softmax+linear-attention stack with 1M-token context and supports English, Korean, and Japanese; requires H200/B200-class GPUs to run efficiently.
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.
AI Model2026
An open-weight LLM checkpoint post-trained for agentic deep search: Qwen-compatible reasoning and tool-call formats optimized for web browsing, multi-source evidence aggregation, long-horizon planning and recovery from failed interactions; typically paired with the AxisAgentic harness.
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
An open-weight, Qwen-derived thinking model optimized for agentic deep web search and long-horizon planning. Provides Qwen-compatible reasoning and tool-call formats for English/Chinese browsing, multi-source evidence aggregation, source verification, and recovery from failed environment interactions.
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.