Flan-T5 XL: Complete Benchmark Performance Guide

other flan-t5-xl 2022-12-07 Apache 2.0

In-depth analysis of Flan-T5 XL performance on MMLU, HumanEval, GSM8K, MATH, BBH benchmarks, including pricing and use cases.

Key Specifications

Vendorother
Versionflan-t5-xl
Release Date2022-12-07
Context Window4096 tokens
Input Modalitiestext
Output Modalitiestext
LicenseApache 2.0
Documentationhttps://huggingface.co/models

Benchmark Performance

Benchmark Score Unit Evaluated At Notes Source
MMLU 60.4 % 2022-12-07 5-shot view
HUMANEVAL 34 pass@1 2022-12-07 — view
GSM8K 46.5 % 2022-12-07 0-shot CoT view
MATH 27.8 % 2022-12-07 0-shot CoT view
BBH 52.5 % 2022-12-07 3-shot CoT view
GPQA 22.6 % 2022-12-07 0-shot view
IFEVAL 55.7 % 2022-12-07 prompt_strict view
ARC 88 % 2022-12-07 challenge view
MUSR 40.2 % 2022-12-07 0-shot view
WINOGRANDE 75.5 % 2022-12-07 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://huggingface.co/models · as of 2022-12-07

Compliance

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

Flan-T5 XL

Model Overview

Google Flan-T5 XL 3B 指令微调模型, 4K 上下文, 基于多任务指令微调, 适合零样本泛化任务。

Core Specifications

Vendor Version Release Date Context Window Input Modalities Output Modalities License
Other flan-t5-xl 2022-12-07 4K text text Apache 2.0

Benchmark Performance

Benchmark Score Unit Notes
MMLU (Massive Multitask Language Understanding) 60.4 % 5-shot
HumanEval 34.0 pass@1 —
GSM8K (Grade School Math 8K) 46.5 % 0-shot CoT
MATH 27.8 % 0-shot CoT
BBH (BIG-Bench Hard) 52.5 % 3-shot CoT
GPQA 22.6 % 0-shot
IFEval 55.7 % prompt_strict
ARC 88.0 % challenge
MUSR 40.2 % 0-shot
WinoGrande 75.5 % 0-shot

Pricing

Input Output Cache Read Cache Write
— — — —

per million tokens

Strengths

  • Reliable general-purpose model.

Weaknesses

  • HumanEval 34.0, coding weak.
  • Proprietary, not self-hostable.
  • Context window 4K is limited.

Use Cases

  • General chat and Q&A

References