Qwen2 57B: Complete Benchmark Performance Guide

alibaba 2-57b 2024-06-07 Qwen License

In-depth analysis of Qwen2 57B performance on MMLU, HumanEval, GSM8K, MATH, BBH benchmarks, including pricing and use cases.

Key Specifications

Vendoralibaba
Version2-57b
Release Date2024-06-07
Context Window131072 tokens
Input Modalitiestext
Output Modalitiestext
LicenseQwen License
Documentationhttps://qwenlm.github.io/

Benchmark Performance

Benchmark Score Unit Evaluated At Notes Source
MMLU 77.7 % 2024-06-07 5-shot view
HUMANEVAL 67.6 pass@1 2024-06-07 — view
GSM8K 83.3 % 2024-06-07 0-shot CoT view
MATH 54.8 % 2024-06-07 0-shot CoT view
BBH 72.4 % 2024-06-07 3-shot CoT view
GPQA 39.4 % 2024-06-07 0-shot view
IFEVAL 71.2 % 2024-06-07 prompt_strict view
ARC 94.6 % 2024-06-07 challenge view
MUSR 50.5 % 2024-06-07 0-shot view
WINOGRANDE 78.1 % 2024-06-07 0-shot view

Pricing

Tier Price Currency
Input$0.4 / MtokUSD
Output$0.7 / MtokUSD
Cache Read$0 / MtokUSD
Cache Write$0 / MtokUSD

Source: https://help.aliyun.com/zh/model-studio/getting-started/models · as of 2024-06-07

Compliance

  • Data Residency: CN
  • SOC2: ✗
  • HIPAA: ✗
  • GDPR: ✗
  • ISO 27001: ✗

Qwen2 57B

Model Overview

阿里巴巴 Qwen2 57B MoE 开源模型, 131K 上下文, 总参 57B/活跃 14B, 性能与效率平衡。

Core Specifications

Vendor Version Release Date Context Window Input Modalities Output Modalities License
Alibaba 2-57b 2024-06-07 131K text text Qwen License

Benchmark Performance

Benchmark Score Unit Notes
MMLU (Massive Multitask Language Understanding) 77.7 % 5-shot
HumanEval 67.6 pass@1 —
GSM8K (Grade School Math 8K) 83.3 % 0-shot CoT
MATH 54.8 % 0-shot CoT
BBH (BIG-Bench Hard) 72.4 % 3-shot CoT
GPQA 39.4 % 0-shot
IFEval 71.2 % prompt_strict
ARC 94.6 % challenge
MUSR 50.5 % 0-shot
WinoGrande 78.1 % 0-shot

Pricing

Input Output Cache Read Cache Write
— — — —

per million tokens

Strengths

  • Mixture-of-Experts architecture.

Weaknesses

  • Proprietary, not self-hostable.

Use Cases

  • General chat and Q&A

References