decima
4 TagsUpdated Apache-2.0by A. M. Madani
A. M. Madani's Decima: multilingual encoders (mmBERT-base, multilingual-e5-small) with a late-interaction scorer that reads every option against the state, so option order never changes the answer, and an ordinal head for scores. A general model, one for coding-agent decisions, and a small one that is the fastest on a CPU.
decisionmultilingualfine-tuned122m321m
Tags
| Name |
|---|
| decimalatestSame as decima:base.1.3 GB · 512 ctx · 100+ languages |
| decima:agentDecima-agent 2.1: decima-base fine-tuned for coding-agent decisions (secret and command gates, tool and model choice); states up to 2,048 tokens.1.3 GB · 2048 ctx · 100+ languages |
| decima:baseDecima-base 2.0 (mmBERT-base, 321M), fp32: 0.495 on typed decisions; 15 ms for five questions on an RTX 4090.1.3 GB · 512 ctx · 100+ languages |
| decima:smallDecima-small 1.1 (multilingual-e5-small, 122M), fp32: 0.432 on typed decisions; 146 ms for five questions on a CPU, the fastest there.510 MB · 512 ctx · 100+ languages |
Tags without a suffix load fp16 on a CUDA GPU and fp32 on CPU; add -fp16 or -fp32 to pin one precision.