Atlas

Models

← All models

Yi 1.5 6B Chat

01.AI · 2024-05-13 · 6.1B parameters

Yi-1.5-6B-Chat is 01.AI's open-weight conversational checkpoint in the Yi-1.5 series, open-sourced on May 13, 2024. It was produced by continuing training from the bilingual English/Chinese Yi foundation on a corpus reported as 500B tokens, followed by fine-tuning on 3M diverse samples; 01.AI reports improvements in coding, mathematics, reasoning, and instruction following. This exact checkpoint is configured for a 4,096-token context window.

Benchmark scores

BenchmarkScore
BBH (Open LLM Leaderboard v2)45.7
BigCodeBench Complete33.9
BigCodeBench Instruct25.6
BigCodeBench-Hard Complete6.8
BigCodeBench-Hard Instruct4.1
Capability105.5
EQ-Bench v259.5
EQ-Bench v2 + MAGI-Hard Combined52.8
GPQA (Open LLM Leaderboard v2)30.2
IFEval (Open LLM Leaderboard v2)51.5
MAGI-Hard46.2
MATH Level 5 (Open LLM Leaderboard v2)16.2
MEDIC (closed-ended)50.5
MEDIC (open-ended Elo)1461
MMLU-Pro (Open LLM Leaderboard v2)31.9
MuSR (Open LLM Leaderboard v2)43.9
Loading Atlas data…