跳转到主要内容

Phi-3.5-mini-instruct

MicrosoftPhiOpen WeightMIT · Commercial OK

描述

Phi-3.5-mini-instruct is a 3.8B-parameter model that supports up to 128K context tokens, with improved multilingual capabilities across over 20 languages. It underwent additional training and safety post-training to enhance instruction-following, reasoning, math, and code generation. Ideal for environments with memory or latency constraints, it uses an MIT license.

发布日期
2024-08-23
参数规模
3.8B
上下文长度
支持模态
text

能力雷达图

60
general
60
coding
60
reasoning
26
science估算
0
agents
0
multimodal

Science 在缺少专门科学评测时使用推理能力代理估算。

排行榜排名

领域#排名分数来源
推理能力51
69.0
LS

基准测试分数 (LLM Stats)

Biology

GPQA30.4%自报

Code

RepoQA77.0%自报
HumanEval62.8%自报

Creativity

Social IQa74.7%自报
Arena Hard37.0%自报

Finance

MMLU69.0%自报
TruthfulQA64.0%自报
MMLU-Pro47.4%自报

General

ARC-C84.6%自报
PIQA81.0%自报
OpenBookQA79.2%自报
MBPP0.70 / 100自报
MMMLU55.4%自报

Language

BoolQ78.0%自报
MEGA XStoryCloze73.5%自报
BIG-Bench Hard69.0%自报
Winogrande68.5%自报
MEGA XCOPA63.1%自报
MEGA TyDi QA62.2%自报
MEGA MLQA61.7%自报
MEGA UDPOS46.5%自报
SQuALITY24.3%自报

Long Context

RULER84.1%自报
Qasper41.9%自报
GovReport25.9%自报
QMSum21.3%自报
SummScreenFD16.0%自报

Math

GSM8k86.2%自报
MATH48.5%自报
MGSM47.9%自报

Reasoning

HellaSwag69.4%自报

AA 评测指数

暂无 AA 评测数据

LLM Stats 分类评分

Psychology
70
Reasoning
70
Code
60
Creativity
60
Finance
60
General
60
Healthcare
60
Language
60
Legal
60
Math
60
Physics
60
Long Context
50
Writing
40
Biology
30
Chemistry
30
Summarization
20

定价

暂无定价数据

速度

暂无速度数据

可用提供商

(LS 内部计价单位)

暂无提供商数据

外部链接