Phi-3 Medium: Complete Benchmark Performance Guide

other phi-3-medium 2024-05-21 MIT

In-depth analysis of Phi-3 Medium performance on MMLU, HumanEval, GSM8K, MATH, BBH benchmarks, including pricing and use cases.

Key Specifications

Vendorother
Versionphi-3-medium
Release Date2024-05-21
Context Window4096 tokens
Input Modalitiestext
Output Modalitiestext
LicenseMIT
Documentationhttps://huggingface.co/models

Benchmark Performance

Benchmark Score Unit Evaluated At Notes Source
MMLU 62.1 % 2024-05-21 5-shot view
HUMANEVAL 53.9 pass@1 2024-05-21 — view
GSM8K 59 % 2024-05-21 0-shot CoT view
MATH 34.3 % 2024-05-21 0-shot CoT view
BBH 71.4 % 2024-05-21 3-shot CoT view
GPQA 25.6 % 2024-05-21 0-shot view
IFEVAL 67.4 % 2024-05-21 prompt_strict view
ARC 90.1 % 2024-05-21 challenge view
MUSR 43.9 % 2024-05-21 0-shot view
WINOGRANDE 72.1 % 2024-05-21 0-shot view

Pricing

Tier Price Currency
Input$0.2 / MtokUSD
Output$0.2 / MtokUSD
Cache Read$0 / MtokUSD
Cache Write$0 / MtokUSD

Source: https://huggingface.co/models · as of 2024-05-21

Compliance

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

Phi-3 Medium

Model Overview

Microsoft Phi-3 Medium 14B 模型, 4K 上下文 (可扩展 128K), 在 14B 规模上接近 GPT-3.5 性能。

Core Specifications

Vendor Version Release Date Context Window Input Modalities Output Modalities License
Other phi-3-medium 2024-05-21 4K text text MIT

Benchmark Performance

Benchmark Score Unit Notes
MMLU (Massive Multitask Language Understanding) 62.1 % 5-shot
HumanEval 53.9 pass@1 —
GSM8K (Grade School Math 8K) 59.0 % 0-shot CoT
MATH 34.3 % 0-shot CoT
BBH (BIG-Bench Hard) 71.4 % 3-shot CoT
GPQA 25.6 % 0-shot
IFEval 67.4 % prompt_strict
ARC 90.1 % challenge
MUSR 43.9 % 0-shot
WinoGrande 72.1 % 0-shot

Pricing

Input Output Cache Read Cache Write
— — — —

per million tokens

Strengths

  • Reliable general-purpose model.

Weaknesses

  • Proprietary, not self-hostable.
  • Context window 4K is limited.

Use Cases

  • General chat and Q&A

References