深潮TechFlow
深潮TechFlow|Nov 19, 2026 15:26
[AI Research Project Marin Releases 57 Training Checkpoints for Mochi Model] DeepFlow TechFlow reported on November 19th that Open Athena announced on November 17th that the Mochi 535B-A23B model from the Marin project has completed its main pre-training phase and released 57 training checkpoints, 10 of which are from the long context expansion phase. Open Athena stated that these intermediate checkpoints have not yet undergone instruction fine-tuning or safety post-training and are intended for scientific research or further training use, and should not be used as a directly deployable tool. The team also stated that an investigation revealed data leakage in some commonly used pre-training benchmarks, which could affect the reportable evaluation results when fairly comparing Mochi with other models.
+2
Mentioned
Share To

Timeline

HotFlash

APP

X

Telegram

Facebook

Reddit

CopyLink

Hot Reads