Code Llama 7B: Complete Benchmark Performance Guide

meta code-llama-7b 2023-08-24 Llama 2 Community License

In-depth analysis of Code Llama 7B performance on MMLU, HumanEval, GSM8K, MATH, BBH benchmarks, including pricing and use cases.

Key Specifications

Vendormeta
Versioncode-llama-7b
Release Date2023-08-24
Context Window16000 tokens
Input Modalitiestext
Output Modalitiestext
LicenseLlama 2 Community License
Documentationhttps://llama.meta.com/docs/

Benchmark Performance

Benchmark Score Unit Evaluated At Notes Source
MMLU 58.5 % 2023-08-24 5-shot view
HUMANEVAL 76.4 pass@1 2023-08-24 — view
GSM8K 52.2 % 2023-08-24 0-shot CoT view
MATH 47.1 % 2023-08-24 0-shot CoT view
BBH 66.9 % 2023-08-24 3-shot CoT view
GPQA 29.5 % 2023-08-24 0-shot view
IFEVAL 61.2 % 2023-08-24 prompt_strict view
ARC 90.1 % 2023-08-24 challenge view
MUSR 42.9 % 2023-08-24 0-shot view
WINOGRANDE 76.3 % 2023-08-24 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://ai.meta.com/blog/ · as of 2023-08-24

Compliance

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

Code Llama 7B

Model Overview

Meta Code Llama 7B 代码专用开源模型, 16K 上下文, 7B 参数, 适合本地代码补全与生成。

Core Specifications

Vendor Version Release Date Context Window Input Modalities Output Modalities License
Meta code-llama-7b 2023-08-24 16K text text Llama 2 Community License

Benchmark Performance

Benchmark Score Unit Notes
MMLU (Massive Multitask Language Understanding) 58.5 % 5-shot
HumanEval 76.4 pass@1 —
GSM8K (Grade School Math 8K) 52.2 % 0-shot CoT
MATH 47.1 % 0-shot CoT
BBH (BIG-Bench Hard) 66.9 % 3-shot CoT
GPQA 29.5 % 0-shot
IFEval 61.2 % prompt_strict
ARC 90.1 % challenge
MUSR 42.9 % 0-shot
WinoGrande 76.3 % 0-shot

Pricing

Input Output Cache Read Cache Write
— — — —

per million tokens

Strengths

  • Reliable general-purpose model.

Weaknesses

  • MMLU 58.5, weak knowledge reasoning.
  • Proprietary, not self-hostable.
  • Context window 16K is limited.

Use Cases

  • Code generation and debugging

References