Llama 3.2 1B: Complete Benchmark Performance Guide

meta 3.2-1b 2024-09-25 Llama 3.2 Community License

In-depth analysis of Llama 3.2 1B performance on MMLU, HumanEval, GSM8K, MATH, BBH benchmarks, including pricing and use cases.

Key Specifications

Vendormeta
Version3.2-1b
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
MMLU48.6%2024-09-255-shotview
HUMANEVAL29.9pass@12024-09-25—view
GSM8K48.9%2024-09-250-shot CoTview
MATH11.4%2024-09-250-shot CoTview
BBH38.8%2024-09-253-shot CoTview
GPQA22.6%2024-09-250-shotview
IFEVAL44.1%2024-09-25prompt_strictview
ARC75.9%2024-09-25challengeview
MUSR26.6%2024-09-250-shotview
WINOGRANDE64%2024-09-250-shotview

Pricing

TierPriceCurrency
Input$0.1 / MtokUSD
Output$0.1 / 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 1B

Ikhtisar Model

Meta Llama 3.2 1B 超轻量开源模型, 128K 上下文, 1B 参数, 适合资源受限设备与离线部署。

Spesifikasi Inti

VendorVersiTanggal RilisJendela KonteksModalitas InputModalitas OutputLisensi
Meta3.2-1b2024-09-25128KtexttextLlama 3.2 Community License

Kinerja Benchmark

BenchmarkSkorSatuanCatatan
MMLU (Massive Multitask Language Understanding)48.6%5-shot
HumanEval29.9pass@1—
GSM8K (Grade School Math 8K)48.9%0-shot CoT
MATH11.4%0-shot CoT
BBH (BIG-Bench Hard)38.8%3-shot CoT
GPQA22.6%0-shot
IFEval44.1%prompt_strict
ARC75.9%challenge
MUSR26.6%0-shot
WinoGrande64.0%0-shot

Harga

InputOutputBaca CacheTulis Cache
————

per juta token

Kelebihan

  • 可靠的通用模型。

Kekurangan

  • MMLU 仅 48.6，知识推理偏弱。
  • HumanEval 29.9，代码能力较弱。
  • 闭源专有模型，不支持自托管。

Kasus Penggunaan

  • 通用对话与问答

Referensi