The top 10 hugging face models & datasets measured by RiseFinder over the 7d window ending Week 44 · 26 October 2026.
FDE-Gym is a benchmark dataset for evaluating AI agents on forward-deployed engineering work. It tests whether an agent can enter a customer's running environment, identify what matters there, make job decisions, and ship a system that holds up in production. The dataset contains the original tasks for 20 scenarios, each with an engagement.json file holding the scenario title and related details.
Hugging Face dataset
Open source ↗LiquidAI/LFM2.5-24B-A2B is a reasoning model built for fast, private agents that operate on-device. It is characterized as a 24B-parameter model that utilizes only 2B active parameters. This model represents the largest and final iteration within the LFM2.5 family. Its design focuses on providing efficient, private AI capabilities for local deployment.
Hugging Face model
Open source ↗MiniCPM-V 4.7 is a vision-language model from the MiniCPM-V series, released in a 1B variant. It handles single-image, multi-image, and video understanding, and combines foundation capabilities with efficient visual encoding and on-device inference, meaning it can run on local devices rather than only in the cloud.
Hugging Face model
Open source ↗This is an experimental character-replacement LoRA. It was trained for 1,000 updates and aims to replace a selected character using an image reference. The process is designed to preserve the source scene during the replacement. In local comparisons, it often preserved the background more closely than the base model.
Hugging Face model
Open source ↗Vyber07/cyber-security is a cybersecurity knowledge base dataset intended for AI training. It covers cybersecurity domains including offensive red teaming, bug bounty exploitation, defensive SOC operations, digital forensics, and AI/LLM security. The dataset is 16 GB in size, contains 507 files, and spans more than 30 domains drawn from more than 15 platforms.
Hugging Face dataset
Open source ↗jialinyyzz/humanizer is a 12B language model that rewrites AI-written drafts such as emails, essays, reports, and forum posts in English and Chinese so they read like a person wrote them. It is trained to keep every number, unit, date, name, and quote from the original text and to add nothing. The model runs locally, and no AI detector was used anywhere in its training.
Hugging Face model
Open source ↗Cactus-Compute/whistle is a speech-to-text model designed for mobiles, wearables, robots, smart home devices, automotive systems, and microcontrollers. The whole model is a single 16.9 MB file. It runs on the same CPU engine as Needle, from the same container and the same quantisation, with no dependencies and no GPU.
Hugging Face model
Open source ↗vllm-sr/d3 is a 27B multimodal foundation decision model, part of Decision 3.0, the decision models of vLLM Semantic Router. It takes an input in text or JSON form, optionally with images and videos, together with questions to answer. It can pick one of several options, say yes or no, or rate on a scale, and answers all of the questions it is given.
Hugging Face model
Open source ↗NJU-HARD is a deduplicated, full-resolution dataset for visual question answering. It is a release of the HARD visual question answering data. The dataset contains 1,563 valid VQA records across 8 task types. These records utilize 854 original aerial images, and only images referenced by these valid questions are included.
Hugging Face dataset
Open source ↗markov-ai/cad-1000-hours is a dataset of 509 end-to-end, single-display Windows CAD task recordings across seven CAD software families, totaling 1,018.1229 hours. Each task includes a task prompt, reference input files, submitted CAD deliverables, task-specific evaluation rubrics, a screen recording video, frame-synchronized events, metadata, and frame-level task narration. The dataset contains 5,455 public task objects totaling 207.18 GiB, and every task includes all seven required root files plus at least one input and one output.
Hugging Face dataset
Open source ↗Get early signals on the websites, apps, repositories and startups gaining momentum. Join the list for Risefinder email briefings.
By subscribing, you ask us to store your email and send Risefinder updates. You can unsubscribe at any time. Privacy.
Want a sample first? Read past daily emails.