Phi-3.5 Mini: Complete Benchmark Performance Guide

other phi-3-5-mini 2024-08-16 MIT

In-depth analysis of Phi-3.5 Mini performance on MMLU, HumanEval, GSM8K, MATH, BBH benchmarks, including pricing and use cases.

Key Specifications

Vendorother
Versionphi-3-5-mini
Release Date2024-08-16
Context Window128000 tokens
Input Modalitiestext
Output Modalitiestext
LicenseMIT
Documentationhttps://huggingface.co/models

Benchmark Performance

BenchmarkScoreUnitEvaluated AtNotesSource
MMLU55.7%2024-08-165-shotview
HUMANEVAL44pass@12024-08-16—view
GSM8K22.1%2024-08-160-shot CoTview
MATH9.6%2024-08-160-shot CoTview
BBH47.5%2024-08-163-shot CoTview
GPQA24.4%2024-08-160-shotview
IFEVAL48.7%2024-08-16prompt_strictview
ARC81.5%2024-08-16challengeview
MUSR31.2%2024-08-160-shotview
WINOGRANDE68.6%2024-08-160-shotview

Pricing

TierPriceCurrency
Input$0.1 / MtokUSD
Output$0.1 / MtokUSD
Cache Read$0 / MtokUSD
Cache Write$0 / MtokUSD

Source: https://huggingface.co/models · as of 2024-08-16

Compliance

  • Data Residency: self-host
  • SOC2: ✗
  • HIPAA: ✗
  • GDPR: ✗
  • ISO 27001: ✗

Phi-3.5 Mini

Model Overview

Microsoft Phi-3.5 Mini 3.8B 模型, 128K 上下文, 改进多语言与长上下文能力, 适合边缘部署。

Core Specifications

VendorVersionRelease DateContext WindowInput ModalitiesOutput ModalitiesLicense
Otherphi-3-5-mini2024-08-16128KtexttextMIT

Benchmark Performance

BenchmarkScoreUnitNotes
MMLU (Massive Multitask Language Understanding)55.7%5-shot
HumanEval44.0pass@1—
GSM8K (Grade School Math 8K)22.1%0-shot CoT
MATH9.6%0-shot CoT
BBH (BIG-Bench Hard)47.5%3-shot CoT
GPQA24.4%0-shot
IFEval48.7%prompt_strict
ARC81.5%challenge
MUSR31.2%0-shot
WinoGrande68.6%0-shot

Pricing

InputOutputCache ReadCache Write
————

per million tokens

Strengths

  • Context window 128K.

Weaknesses

  • MMLU 55.7, weak knowledge reasoning.
  • HumanEval 44.0, coding weak.
  • Proprietary, not self-hostable.

Use Cases

  • Long document summarization

References