TechPrufer
Get started
Hugging Face
Hugging Face logoHugging Face · Feature trail

Ai Ml

2 posts · Aug 2026 – Oct 2026

What this trail covers

Hugging Face now supports specialized multimodal models, including the LFM3 (Large Foundation Model 3) family, which offers a 70B parameter model with a 128k context window. This enables advanced capabilities in multimodal reasoning, allowing for the creation of specialized models for tasks like visual question answering, image captioning, and complex multimodal reasoning. The LFM3 models are designed for efficient fine-tuning and deployment, making them accessible for a wide range of applications.

All Hugging Face trails
  1. OlmoEarth v1.3: State-of-the-art embeddings for the whole planetThis post introduces OlmoEarth v1.3, a new version of a satellite embedding model. It details the model's architecture (vision transformer, attention pooler, linear projection), its input data (Sentinel-1, Sentinel-2, Landsat time series), and its output (128-dimensional vectors per 10m pixel). The post highlights the model's performance improvements over competitors and previous versions, its efficiency gains in training cost and carbon emissions, and the availability of precomputed global embeddings. It also introduces the OlmoEarth Embeddings Explorer as a tool for interactive analysis and map customization using these embeddings, and discusses the provenance of the embeddings, linking back to source imagery.
  2. Build specialized multimodal models with LFM3This post introduces LFM3 (Large Foundation Model 3), a new family of multimodal models from Liquid AI. LFM3 includes a 70B parameter model with a 128k context window, designed for specialized multimodal reasoning tasks. The post highlights the model's capabilities in visual question answering, image captioning, and complex multimodal reasoning, and discusses its potential for fine-tuning and deployment. It also mentions the collaborative effort behind LFM3, involving multiple researchers and engineers.