Llama 2 70B: Complete Benchmark Performance Guide

meta 2-70b 2023-07-18 Llama 2 Community License

In-depth analysis of Llama 2 70B performance on MMLU, HumanEval, GSM8K, MATH, BBH benchmarks, including pricing and use cases.

Key Specifications

Vendormeta
Version2-70b
Release Date2023-07-18
Context Window4096 tokens
Input Modalitiestext
Output Modalitiestext
LicenseLlama 2 Community License
Documentationhttps://llama.meta.com/docs/

Benchmark Performance

BenchmarkScoreUnitEvaluated AtNotesSource
MMLU53%2023-07-185-shotview
HUMANEVAL50.6pass@12023-07-18—view
GSM8K51.8%2023-07-180-shot CoTview
MATH22.4%2023-07-180-shot CoTview
BBH62.5%2023-07-183-shot CoTview
GPQA27.2%2023-07-180-shotview
IFEVAL57.1%2023-07-18prompt_strictview
ARC80.3%2023-07-18challengeview
MUSR36.1%2023-07-180-shotview
WINOGRANDE73.4%2023-07-180-shotview

Pricing

TierPriceCurrency
Input$0.9 / MtokUSD
Output$0.9 / MtokUSD
Cache Read$0 / MtokUSD
Cache Write$0 / MtokUSD

Source: https://ai.meta.com/blog/ · as of 2023-07-18

Compliance

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

Llama 2 70B

Model Overview

Meta Llama 2 70B 开源模型, 4K 上下文, 在开源模型中领先, 商用许可, 适合企业自托管。

Core Specifications

VendorVersionRelease DateContext WindowInput ModalitiesOutput ModalitiesLicense
Meta2-70b2023-07-184KtexttextLlama 2 Community License

Benchmark Performance

BenchmarkScoreUnitNotes
MMLU (Massive Multitask Language Understanding)53.0%5-shot
HumanEval50.6pass@1—
GSM8K (Grade School Math 8K)51.8%0-shot CoT
MATH22.4%0-shot CoT
BBH (BIG-Bench Hard)62.5%3-shot CoT
GPQA27.2%0-shot
IFEval57.1%prompt_strict
ARC80.3%challenge
MUSR36.1%0-shot
WinoGrande73.4%0-shot

Pricing

InputOutputCache ReadCache Write
————

per million tokens

Strengths

  • Reliable general-purpose model.

Weaknesses

  • MMLU 53.0, weak knowledge reasoning.
  • Proprietary, not self-hostable.
  • Context window 4K is limited.

Use Cases

  • General chat and Q&A

References