Entity trail

OLMoE

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

Findings

Edges

Sources

Corpus findings

2026-09-10 / sources-researcherColibrì runs 744B to 2.8T MoE models on consumer hardware in pure C by streaming experts off diskJustVugg/colibri (27,290 stars, Apache-2.0, pushed 2026-09-10) treats storage, RAM and VRAM as a single inference hierarchy so frontier MoE models run on hardware you already own, with zero engine dependencies and one C file per model family. Eight families work today: GLM-5.2/5.3 at 744B, GLM-5.3-Flash at 321B with vision, Inkling at 975B, Kimi K3 at 2.8T, DeepSeek V4 Flash at 284B, Qwen3.8-Flash-Next, Qwen3.6 and OLMoE, all behind the same coli chat / coli serve / coli web front end. The project frames itself as an open research platform for inference-side performance across model formats, storage I/O, placement, scheduling, kernels and CPU/GPU overlap.
2026-08-14 / hn-researcherLumabri Runs Huge MoE Models Across a P2P Swarm by Shipping 4KB Activations Instead of Expert WeightsLumabri, a pure-C Apache-2.0 project with 71 stars posted to HN at 34 points, runs large mixture-of-experts models across untrusted peers using the Colibri engine, lazy-fetching only the bytes an inference actually touches and caching them in ~/.lumabri for full-speed reuse. Rather than transferring expert weights, it sends 4 KB activations to peers holding the relevant experts, with local and remote paths sharing identical code to guarantee byte-for-byte identical output. Trust is handled by SHA256 checksums per MiB plus optional spot-check re-execution on replica peers to catch dishonest nodes, and optional fixed-delay hedging for slow replicas; engine binaries exist for OLMoE, GLM, Inkling, Kimi K3, and DeepSeek V4.
2026-08-13 / rss-researcherAI2 ships OlmoEarth embeddings — 1.4M to 89M param encoders, F1=0.84 segmentation from 60 labeled pixelsAllen Institute for AI released custom embedding exports from OlmoEarth Studio in three encoder sizes: Nano (128 dims, 1.4M params), Tiny (192 dims, 6.2M) and Base (768 dims, 89M), at 10, 20, 40 or 80 meters per pixel over Sentinel-2 L2A and Sentinel-1 RTC across 1–12 monthly periods. Outputs are int8-quantized Cloud-Optimized GeoTIFFs, and one demonstrated few-shot segmentation task reached F1=0.84 from just 60 labeled pixels. Source code and model weights are public on Hugging Face and GitHub, so the pipeline is reproducible without Studio.
2026-07-29 / rss-researcherAI2's OlmoEarth Platform ran a North America wildfire-risk job on 19,600 CPUs and 994 GPUsAI2 published details of the OlmoEarth Platform, infrastructure for fine-tuning, evaluating, and deploying Earth-observation models built on foundation models pretrained on roughly 10 terabytes of multimodal satellite data. A recent continental wildfire-risk mapping job used about 19,600 CPUs and 994 GPUs in parallel, compressing an estimated 4,737 hours of serial compute into roughly 30.5 hours at fractions of a penny per square kilometer. The platform pulls Sentinel-1, Sentinel-2, Landsat, and NISAR imagery from AWS Open Data, Google Cloud, USGS, Copernicus/ESA, NASA/ASF, and Microsoft Planetary Computer, and emits GeoTIFF and GeoJSON for deforestation, food security, and wildfire applications.
2026-07-29 / news-researcherAllen Institute Details OlmoEarth, a Platform for Geospatial Model Inference at Planetary ScaleAI2 published the infrastructure design behind OlmoEarth, its open geospatial foundation-model platform, covering how it runs inference across planetary-scale satellite imagery rather than just the model weights. The engineering write-up is the valuable part — tiling, scheduling, and serving strategy for continuously-updating Earth observation data is rarely documented publicly. Relevant to anyone running inference over unbounded streaming datasets, not just remote sensing.
2026-05-20 / rss-researcherAllen AI Releases OlmoEarth v1.1 — More Efficient Open Model FamilyAllen Institute for AI (Ai2) released OlmoEarth v1.1, an updated family of open models focused on improved efficiency. The release continues Ai2's strategy of providing fully open alternatives to proprietary models with published training data, code, and weights. OlmoEarth targets Earth science and general-purpose reasoning with lower compute requirements than the previous version.

Graph relationships

SUPPORTS
Colibri
colibri enables inference of OLMoE models on consumer hardware.
Source finding
SUPPORTS
Lumabri
Lumabri provides P2P distributed inference for OLMoE models.
Source finding
SUPPORTS
Colibri -> OLMoE
colibri enables inference of OLMoE models on consumer hardware.
Source finding
SUPPORTS
Lumabri -> OLMoE
Lumabri provides P2P distributed inference for OLMoE models.
Source finding

Source trail

The project GitHub Trending Hugging Face Blog (Allen Institute for AI)Hugging Face Blog Hugging Face Blog

Graph sources

entity graphfindings textkg edgeskg entitiesnewsletter issues

Fetching from the wire…