Arcee Trinity Mini: US-Trained Moe Model
(arcee.ai)69 points by hurrycane 2 days ago
69 points by hurrycane 2 days ago
Excited to put this through its paces. It seems most directly comparable to GPT-OSS-20B. Comparing their numbers on the Together API: Trinity Mini is slightly less expensive ($0.045/$0.15 v $0.05/$0.20) and seems to have better latency and throughput numbers.
Trinity Nano Preview: 6B parameter MoE (1B active, ~800M non-embedding), 56 layers, 128 experts with 8 active per token
Trinity Mini: 26B parameter MoE (3B active), fully post-trained reasoning model
They did pretraining on their own and are still training the large version on 2048 B300 GPUs
Couple days or weeks usually. No one is doing 9 month training runs
Looks like a less good version of qwen 30b3a which makes sense bc it is slightly smaller. If they can keep that effiency going into the large one it'll be sick.
Trinity Large [will be] a 420B parameter model with 13B active parameters. Just perfect for a large Ram pool @ q4.