Jamba Instruct: Complete Benchmark Performance Guide

other jamba-instruct 2024-03-28 Jamba Open Model License

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

Key Specifications

Vendorother
Versionjamba-instruct
Release Date2024-03-28
Context Window256000 tokens
Input Modalitiestext
Output Modalitiestext
LicenseJamba Open Model License
Documentationhttps://huggingface.co/models

Benchmark Performance

Benchmark Score Unit Evaluated At Notes Source
MMLU 76.2 % 2024-03-28 5-shot view
HUMANEVAL 66 pass@1 2024-03-28 — view
GSM8K 82.2 % 2024-03-28 0-shot CoT view
MATH 49 % 2024-03-28 0-shot CoT view
BBH 71.8 % 2024-03-28 3-shot CoT view
GPQA 37.4 % 2024-03-28 0-shot view
IFEVAL 79.2 % 2024-03-28 prompt_strict view
ARC 91.2 % 2024-03-28 challenge view
MUSR 56.9 % 2024-03-28 0-shot view
WINOGRANDE 78.3 % 2024-03-28 0-shot view

Pricing

Tier Price Currency
Input$0.5 / MtokUSD
Output$0.7 / MtokUSD
Cache Read$0 / MtokUSD
Cache Write$0 / MtokUSD

Source: https://huggingface.co/models · as of 2024-03-28

Compliance

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

Jamba Instruct

Model Overview

AI21 Labs Jamba Instruct 指令微调模型, 256K 上下文, 首个商用 Mamba-Transformer 混合架构。

Core Specifications

Vendor Version Release Date Context Window Input Modalities Output Modalities License
Other jamba-instruct 2024-03-28 256K text text Jamba Open Model License

Benchmark Performance

Benchmark Score Unit Notes
MMLU (Massive Multitask Language Understanding) 76.2 % 5-shot
HumanEval 66.0 pass@1 —
GSM8K (Grade School Math 8K) 82.2 % 0-shot CoT
MATH 49.0 % 0-shot CoT
BBH (BIG-Bench Hard) 71.8 % 3-shot CoT
GPQA 37.4 % 0-shot
IFEval 79.2 % prompt_strict
ARC 91.2 % challenge
MUSR 56.9 % 0-shot
WinoGrande 78.3 % 0-shot

Pricing

Input Output Cache Read Cache Write
— — — —

per million tokens

Strengths

  • Context window 256K.

Weaknesses

  • Proprietary, not self-hostable.

Use Cases

  • Long document summarization

References