Llama 2 13B: Complete Benchmark Performance Guide

meta 2-13b 2023-07-18 Llama 2 Community License

In-depth analysis of Llama 2 13B performance on MMLU, HumanEval, GSM8K, MATH, BBH benchmarks, including pricing and use cases.

Key Specifications

Vendormeta
Version2-13b
Release Date2023-07-18
Context Window4096 tokens
Input Modalitiestext
Output Modalitiestext
LicenseLlama 2 Community License
Documentationhttps://llama.meta.com/docs/

Benchmark Performance

Benchmark Score Unit Evaluated At Notes Source
MMLU 56.4 % 2023-07-18 5-shot view
HUMANEVAL 40.7 pass@1 2023-07-18 — view
GSM8K 34.6 % 2023-07-18 0-shot CoT view
MATH 25.9 % 2023-07-18 0-shot CoT view
BBH 62.1 % 2023-07-18 3-shot CoT view
GPQA 22.3 % 2023-07-18 0-shot view
IFEVAL 49.9 % 2023-07-18 prompt_strict view
ARC 89.5 % 2023-07-18 challenge view
MUSR 33.8 % 2023-07-18 0-shot view
WINOGRANDE 76.4 % 2023-07-18 0-shot view

Pricing

Tier Price Currency
Input$0.3 / MtokUSD
Output$0.3 / 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 13B

Model Overview

Meta Llama 2 13B 中型开源模型, 4K 上下文, 13B 参数, 平衡性能与资源消耗, 适合中等规模部署。

Core Specifications

Vendor Version Release Date Context Window Input Modalities Output Modalities License
Meta 2-13b 2023-07-18 4K text text Llama 2 Community License

Benchmark Performance

Benchmark Score Unit Notes
MMLU (Massive Multitask Language Understanding) 56.4 % 5-shot
HumanEval 40.7 pass@1 —
GSM8K (Grade School Math 8K) 34.6 % 0-shot CoT
MATH 25.9 % 0-shot CoT
BBH (BIG-Bench Hard) 62.1 % 3-shot CoT
GPQA 22.3 % 0-shot
IFEval 49.9 % prompt_strict
ARC 89.5 % challenge
MUSR 33.8 % 0-shot
WinoGrande 76.4 % 0-shot

Pricing

Input Output Cache Read Cache Write
— — — —

per million tokens

Strengths

  • Reliable general-purpose model.

Weaknesses

  • MMLU 56.4, weak knowledge reasoning.
  • HumanEval 40.7, coding weak.
  • Proprietary, not self-hostable.
  • Context window 4K is limited.

Use Cases

  • General chat and Q&A

References