Atlas

Models

← All models

MPT-7B

MosaicML · 2023-05-05 · 6.6B parameters

MPT-7B is MosaicML's dense decoder-only base transformer released on May 5, 2023; the published checkpoint contains 6,649,286,656 parameters. MosaicML reports that it was trained from scratch on 1 trillion tokens of text and code using ALiBi and the Lion optimizer. Its 2,048-token base checkpoint was released under Apache-2.0, while the Instruct, Chat, and StoryWriter releases were separate fine-tuned artifacts.

Benchmark scores

BenchmarkScore
BBH (Open LLM Leaderboard v2)33.0
Capability87.1
Epoch Capabilities Index (ECI)93.0
GPQA (Open LLM Leaderboard v2)26.0
IFEval (Open LLM Leaderboard v2)21.5
LAMBADA70.0
MATH Level 5 (Open LLM Leaderboard v2)1.6
MMLU-Pro (Open LLM Leaderboard v2)12.1
MuSR (Open LLM Leaderboard v2)36.7
Loading Atlas data…