Llama 3.2 3B: Complete Benchmark Performance Guide

meta 3.2-3b 2024-09-25 Llama 3.2 Community License

In-depth analysis of Llama 3.2 3B performance on MMLU, HumanEval, GSM8K, MATH, BBH benchmarks, including pricing and use cases.

Key Specifications

Vendormeta
Version3.2-3b
Release Date2024-09-25
Context Window128000 tokens
Input Modalitiestext
Output Modalitiestext
LicenseLlama 3.2 Community License
Documentationhttps://llama.meta.com/docs/

Benchmark Performance

BenchmarkScoreUnitEvaluated AtNotesSource
MMLU52.5%2024-09-255-shotview
HUMANEVAL42.5pass@12024-09-25—view
GSM8K46.4%2024-09-250-shot CoTview
MATH10%2024-09-250-shot CoTview
BBH46%2024-09-253-shot CoTview
GPQA18.8%2024-09-250-shotview
IFEVAL48.6%2024-09-25prompt_strictview
ARC76.3%2024-09-25challengeview
MUSR29.3%2024-09-250-shotview
WINOGRANDE67.8%2024-09-250-shotview

Pricing

TierPriceCurrency
Input$0.15 / MtokUSD
Output$0.15 / MtokUSD
Cache Read$0 / MtokUSD
Cache Write$0 / MtokUSD

Source: https://ai.meta.com/blog/ · as of 2024-09-25

Compliance

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

Llama 3.2 3B

모델 개요

Meta Llama 3.2 3B 轻量开源模型, 128K 上下文, 3B 参数, 适合移动设备与边缘部署。

핵심 사양

공급업체버전출시일컨텍스트 창입력 모달리티출력 모달리티라이선스
Meta3.2-3b2024-09-25128KtexttextLlama 3.2 Community License

벤치마크 성능

벤치마크점수단위비고
MMLU (Massive Multitask Language Understanding)52.5%5-shot
HumanEval42.5pass@1—
GSM8K (Grade School Math 8K)46.4%0-shot CoT
MATH10.0%0-shot CoT
BBH (BIG-Bench Hard)46.0%3-shot CoT
GPQA18.8%0-shot
IFEval48.6%prompt_strict
ARC76.3%challenge
MUSR29.3%0-shot
WinoGrande67.8%0-shot

가격

입력출력캐시 읽기캐시 쓰기
————

백만 토큰당

강점

  • 可靠的通用模型。

약점

  • MMLU 仅 52.5，知识推理偏弱。
  • HumanEval 42.5，代码能力较弱。
  • 闭源专有模型，不支持自托管。

사용 사례

  • 通用对话与问答

참고문헌