CrowdGPT v2.6: crowdsourced LLM training adds 4.5B tokens in 6 days
Vxtzq1 · reddit · 2026-11-01
CrowdGPT is an open-source project aiming to train an LLM collaboratively without datacenters or millions in funding, targeting a 1B-parameter model as proof of concept. Version 2.6 trained on 4.5B new tokens in 6 days—what the author previously achieved in two months on a personal GPU. The update adds server-side validation, Byzantine cross-checks (3 clients per data shard) that cap malicious contributions at 2-3% per round, cryptographic signing of client releases (patching an arbitrary code execution risk in auto-update), activated Engrams for better fact recall, and a collaborative dataset submission pipeline. The author also released CrowdBench, an open collaborative benchmark, and invites data-poisoning experiments.
More from Research
- Stanford's VERDICT clinical trial matching paper wins EMNLP 2026 Outstanding Paper Award — stanfordnlp · 2026-11-01
- Study of 3,172 US Chinese restaurants reveals systematic English–Chinese name gaps — stanfordnlp · 2026-11-01
- Stanford's GUIDE generates personalized mental health interventions at runtime, RCT shows reduced stress — stanfordnlp · 2026-11-01
- Bill Gates podcast interviews AI researcher Yejin Choi on AI training systems — stanfordnlp · 2026-11-01
- Researchers debate LLM layerwise picture: where does a model's persona live? — maxsloef · 2026-11-01
- QwenPU runs Karpathy's stories260K on an NPU compiled inside Qwen 3.5-4B's weights — LatentSpacer · 2026-11-01