Daily AI

Learns fine-grained preferences over sub-trajectories to identify and penalize redundant steps in long chain-of-thoughts, letting models "fold" reasoning chains into concise paths; reports ~56% token reduction on DeepSeek-R1-Distill-Qwen-7B while keeping accuracy.

Evaluates multimodal LLMs on streaming egocentric video for spatial intelligence using 1,680 human-annotated questions across 348 videos; organizes tasks into four hierarchical levels (perception → tracking → simulation → allocentric mapping) and highlights allocentric mapping as the main bottleneck.

Explores how training recipe — data composition, teacher guidance, and task mixture — shapes few-step distillation for text-to-image generation and instruction-guided image editing; introduces Qwen-Image-Flash and empirical findings that training pipeline organization matters as much as distillation objectives.

Localizes harmful span-level errors inside long research-agent trajectories to show which trajectory segments make final answers unreliable. Provides a 1,000-instance TELBench of annotated spans and DRIFT, a claim-centric auditing method that improves span-level localization and first-error accuracy by up to 30 percentage points.

Omnimodal world model that jointly processes and generates text, images, video, audio, and action trajectories for physical AI. Uses a mixture-of-transformers to combine autoregressive reasoning and diffusion-based multimodal generation; released open-source with checkpoints, datasets and benchmarks for robotics and simulation.

Provides ~1M synthetic Salvadoran‑Spanish personas (148k records, ~300M tokens) grounded in 2024 census distributions for demographics, occupations and locations; intended for training/evaluating localized LLMs and synthetic-data workflows. CC BY 4.0, adults only.

NF4-quantized text-to-image diffusion model released as safetensors and compatible with the Diffusers Ideogram4Pipeline — optimized for lower-memory local inference and faster deployments while preserving the original model's text-to-image capabilities.

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.

Text-to-image model packaged for Diffusers that uses fp8 quantization to lower memory and speed up inference. Delivered as a safetensors checkpoint on Hugging Face with an Ideogram pipeline; created May 30, 2026 — license unspecified.

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.

A 12B unified, encoder-free multimodal model that directly ingests text, images and audio and returns text; supports very long contexts (up to 256K tokens), native function-calling/thinking modes, and small-model deployment for local or on-device use.

Instruction-tuned, unified Gemma 4 12B multimodal model that accepts text, image and audio inputs and generates text outputs locally. Encoder-free design reduces multimodal latency and fits on consumer devices while offering long-context support and native thinking/system-prompt features.

Web-based resume editor with real-time preview, custom themes, dark mode, auto-save and PDF export, plus built-in AI-assisted writing and a custom model for polishing content. Open-source under Apache-2.0 but requires a commercial license for paid/enterprise use.

Renders Mermaid source synchronously to themeable SVG or ASCII/Unicode art for UIs and terminals. Includes 15 built-in themes, Shiki theme extraction, mono mode and zero-DOM dependencies so diagrams render instantly in React, CLIs, or chat/agent UIs.

Aggregates and deduplicates stories from Hacker News, Reddit, RSS, Telegram, GitHub and more, then uses LLMs to score, enrich, and produce bilingual (EN/CN) daily briefings. Supports customizable sources, comment summarization, multi-provider scoring, and delivery via GitHub Pages, email, or webhooks — designed for self-hosted, configurable news digests.

Parses PDF resumes into structured JSON using LLMs, enriches profiles with GitHub signals, and outputs explainable category scores, evidence, bonuses and deductions. Runs fully local with Ollama or via Google Gemini; designed for reproducible, fairness-constrained resume scoring in hiring workflows.

Runs 70B-class LLM inference on a single 4GB GPU without quantization and supports Llama3.1 405B on 8GB VRAM. Uses layer-splitting and block-wise model compression (4/8-bit) to reduce disk load and can speed up inference loading by up to ~3x; integrates with Hugging Face models.