StarChat2 15B v0.1: Complete Benchmark Performance Guide

other starchat2-15b-v0.1 2024-07-05 BigCode Open Model License

In-depth analysis of StarChat2 15B v0.1 performance on MMLU, HumanEval, GSM8K, MATH, BBH benchmarks, including pricing and use cases.

Key Specifications

Vendorother
Versionstarchat2-15b-v0.1
Release Date2024-07-05
Context Window8192 tokens
Input Modalitiestext
Output Modalitiestext
LicenseBigCode Open Model License
Documentationhttps://huggingface.co/models

Benchmark Performance

Benchmark Score Unit Evaluated At Notes Source
MMLU 74.9 % 2024-07-05 5-shot view
HUMANEVAL 69 pass@1 2024-07-05 — view
GSM8K 50.7 % 2024-07-05 0-shot CoT view
MATH 45.3 % 2024-07-05 0-shot CoT view
BBH 70.6 % 2024-07-05 3-shot CoT view
GPQA 25.9 % 2024-07-05 0-shot view
IFEVAL 63.3 % 2024-07-05 prompt_strict view
ARC 91.7 % 2024-07-05 challenge view
MUSR 44.4 % 2024-07-05 0-shot view
WINOGRANDE 75 % 2024-07-05 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 2024-07-05

Compliance

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

StarChat2 15B v0.1

Model Overview

HuggingFace StarChat2 15B 代码对话模型, 8K 上下文, 基于 StarCoder2 微调, 支持多语言代码生成。

Core Specifications

Vendor Version Release Date Context Window Input Modalities Output Modalities License
Other starchat2-15b-v0.1 2024-07-05 8K text text BigCode Open Model License

Benchmark Performance

Benchmark Score Unit Notes
MMLU (Massive Multitask Language Understanding) 74.9 % 5-shot
HumanEval 69.0 pass@1 —
GSM8K (Grade School Math 8K) 50.7 % 0-shot CoT
MATH 45.3 % 0-shot CoT
BBH (BIG-Bench Hard) 70.6 % 3-shot CoT
GPQA 25.9 % 0-shot
IFEval 63.3 % prompt_strict
ARC 91.7 % challenge
MUSR 44.4 % 0-shot
WinoGrande 75.0 % 0-shot

Pricing

Input Output Cache Read Cache Write
— — — —

per million tokens

Strengths

  • Reliable general-purpose model.

Weaknesses

  • Proprietary, not self-hostable.
  • Context window 8K is limited.

Use Cases

  • Code generation and debugging

References