AI Model2026
A 9B-parameter Qwen3.5-based multimodal model tuned to preserve chain-of-thought reasoning while eliminating repetition loops; restores native multi-token prediction, supports 1,048,576-token context, and targets research/red-team use.
Explores unsupervised visual pretraining on visually rich documents to improve language-model intelligence; shows visual-pretrained models outperform text-only counterparts on the same corpora. Key aspects: direct use of images/layouts (no OCR-only pipeline), scalable across backbones and benchmarks.
Proposes Riemannian Isometric Policy Optimization (RIPO) to fix exploration collapse in PPO-style RL for LLMs by aligning policy updates with the policy manifold's Riemannian geometry, improving exploration–exploitation balance and optimization stability across competition benchmarks.
AI Model2026
A GGUF-format Qwen3.6 35B base model image-text-to-text release repaired via tensor-level SVD/scale correction and packaged with Hermes agent tweaks; multimodal (vision + text), MoE architecture, ready for GGUF runtimes like llama.cpp.
AI Model2026
A GGUF-distributed Qwen3.6 35B MoE model variant repaired with a
AI Model2026
GGUF conversions of Laguna S 2.1 for llama.cpp, including quantized builds (Q4_K_M, Q8_0, F16) and a small DFlash drafter for speculative decoding; configured for a 256K default context window and intended for local inference and serving with Poolside's llama.cpp fork.
AI Model2026
A GGUF build of Qwen3.6 (35B) post-processed with the Genesis numerical repair to reduce training noise and restore weight distributions; provides a more stable, uncensored multimodal (image+text) MoE model with long-context support for local use.
AI Model2026
27B multimodal reasoning model built on Qwen3.5-27B that preserves the base model's native multi-token-prediction head, full vision tower, and a 1,048,576-token YaRN context window. Designed for agentic tool use, long-context reasoning, and research deployments; released under Apache-2.0.
AI Model2026
End-to-end 0.8B multimodal OCR and page-level document parser that converts page images into structured Markdown (text, LaTeX formulas, HTML tables, and image crops). Post-trained from Qwen3.5-0.8B using mixed real/synthetic data and SFT+RL+OPD; achieves 96.58 on OmniDocBench v1.6.
Uses pretrained multimodal LLMs as zero-shot, training-free reward models for text-to-image RL by scoring how well the original text prompt can be recovered from a generated image via image-conditioned prompt log-likelihood; includes a Self-SpectraReward closed-loop variant.
AI Agent Papers2026
Acquires repository knowledge via a targeted QA loop before generating patches, decoupling knowledge acquisition from repair. A Questioner and Answerer produce evidence-grounded QA pairs that a Resolver uses to generate fixes; improves Pass@1 on SWE-bench Verified with modest overhead.
AI Model2026
A GGUF-local variant of Qwen3.6-35B that applies a non-training 'Genesis' tensor-repair process and Hermes-agent fine-tuning to enable uncensored, multimodal (text+image) local inference. Highlights: MoE 35B spec, large native context, Hermes function-calling dataset transfer, and recommended quantization/runtime settings.