Image created with gemini-3.1-flash-image-preview with claude-opus-4.7. Image prompt: A square Lutheran stained-glass panel of a bald athletic minister-saint in running gear, arms outstretched, with four small rose-window medallions around him depicting an eye, ear, mouth, and hand in jewel-tone glass, while overhead a haloed red-tailed hawk carries a cat in its talons rendered simultaneously as rippling sound waves, painted imagery, and inscribed glyphs across the leaded panes; cathedral cobalt, ruby, amber and emerald glass with radiant backlight and gothic tracery, a heavy blackletter title banner at the base reading MULTIMODALITY, medieval illuminated manuscript style with bold black came outlines.
LiteFrame: Efficient Vision Encoders Unlock Frame Scaling in Video LLMs
https://jjihwan.github.io/projects/LiteFrame/
interesting open model by cohere with lots of unusual architecture choices, here is a recap: > parallel transformer, so MoE and attention are computed in parallel. likely doing some kind of MLP/attention disaggregation here? > lots of query heads, query total dim is 4x hidden
https://x.com/eliebakouch/status/2057198733759008989
Searching through unstructured data, like scans of handwritten and typed declassified documents, can be challenging. But with Cohere Compass, it’s possible because it is built to process and retrieve across even the most challenging documents. This includes the Compass Visual
https://x.com/cohere/status/2055343638360752351
ByteDance open source Lance: Unified Multimodal Modeling by Multi-Task Synergy a lightweight native unified multimodal model for image and video understanding, generation, and editing. 3B for video part,3B for image part and 3B for decoder.
https://x.com/bdsqlsz/status/2056353648779907115
Great work at @baseten running vLLM-Omni in production — open-source, production-grade, cost-efficient omni-modal serving 🎙️ Multi-stage audio, streaming multi-modal, real-time TTS — workloads where closed-source APIs have been the default. →
https://x.com/vllm_project/status/2055136943550427242
OlmoEarth v1.1 just dropped (thx @allen_ai) 🌍 This family of Earth observation foundation models for satellite imagery tasks (e.g. mangrove change tracking, forest loss driver classification) just got 3X CHEAPER/FASTER to run. The trick is redesigning what a token represents.
https://x.com/cgeorgiaw/status/2057481909802774664
Leave a Reply