Code Llama 34B: Complete Benchmark Performance Guide

meta code-llama-34b 2023-08-24 Llama 2 Community License

In-depth analysis of Code Llama 34B performance on MMLU, HumanEval, GSM8K, MATH, BBH benchmarks, including pricing and use cases.

Key Specifications

Vendormeta
Versioncode-llama-34b
Release Date2023-08-24
Context Window16000 tokens
Input Modalitiestext
Output Modalitiestext
LicenseLlama 2 Community License
Documentationhttps://llama.meta.com/docs/

Benchmark Performance

BenchmarkScoreUnitEvaluated AtNotesSource
MMLU74.5%2023-08-245-shotview
HUMANEVAL71.7pass@12023-08-24—view
GSM8K59.8%2023-08-240-shot CoTview
MATH44.7%2023-08-240-shot CoTview
BBH59.1%2023-08-243-shot CoTview
GPQA27%2023-08-240-shotview
IFEVAL58.7%2023-08-24prompt_strictview
ARC88.7%2023-08-24challengeview
MUSR44.9%2023-08-240-shotview
WINOGRANDE80%2023-08-240-shotview

Pricing

TierPriceCurrency
Input$0.5 / MtokUSD
Output$0.5 / 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 34B

모델 개요

Meta Code Llama 34B 代码专用开源模型, 16K 上下文, 中等规模, 平衡代码生成性能与资源消耗。

핵심 사양

공급업체버전출시일컨텍스트 창입력 모달리티출력 모달리티라이선스
Metacode-llama-34b2023-08-2416KtexttextLlama 2 Community License

벤치마크 성능

벤치마크점수단위비고
MMLU (Massive Multitask Language Understanding)74.5%5-shot
HumanEval71.7pass@1—
GSM8K (Grade School Math 8K)59.8%0-shot CoT
MATH44.7%0-shot CoT
BBH (BIG-Bench Hard)59.1%3-shot CoT
GPQA27.0%0-shot
IFEval58.7%prompt_strict
ARC88.7%challenge
MUSR44.9%0-shot
WinoGrande80.0%0-shot

가격

입력출력캐시 읽기캐시 쓰기
————

백만 토큰당

강점

  • 可靠的通用模型。

약점

  • 闭源专有模型，不支持自托管。
  • 上下文窗口 16K 偏小。

사용 사례

  • 代码生成与调试

참고문헌