Key Specifications

Vendorother
Versionstarcoder2-3b
Release Date2024-02-28
Context Window16384 tokens
Input Modalitiestext
Output Modalitiestext
LicenseBigCode Open Model License
Documentationhttps://huggingface.co/models

Benchmark Performance

Benchmark Score Unit Evaluated At Notes Source
MMLU 60.6 % 2024-02-28 5-shot view
HUMANEVAL 82.1 pass@1 2024-02-28 view
GSM8K 48.2 % 2024-02-28 0-shot CoT view
MATH 37.4 % 2024-02-28 0-shot CoT view
BBH 62.1 % 2024-02-28 3-shot CoT view
GPQA 36.2 % 2024-02-28 0-shot view
IFEVAL 55.2 % 2024-02-28 prompt_strict view
ARC 87.5 % 2024-02-28 challenge view
MUSR 38.4 % 2024-02-28 0-shot view
WINOGRANDE 76.9 % 2024-02-28 0-shot view

Pricing

Tier Price Currency
Input$0.12 / MtokUSD
Output$0.12 / MtokUSD
Cache Read$0 / MtokUSD
Cache Write$0 / MtokUSD

Source: https://huggingface.co/models · as of 2024-02-28

Compliance

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

StarCoder2 3B

Model Overview

BigCode StarCoder2 3B 轻量代码模型, 16K 上下文, 3B 参数, 适合边缘设备代码补全任务。

Core Specifications

Vendor Version Release Date Context Window Input Modalities Output Modalities License
Other starcoder2-3b 2024-02-28 16K text text BigCode Open Model License

Benchmark Performance

Benchmark Score Unit Notes
MMLU (Massive Multitask Language Understanding) 60.6 % 5-shot
HumanEval 82.1 pass@1
GSM8K (Grade School Math 8K) 48.2 % 0-shot CoT
MATH 37.4 % 0-shot CoT
BBH (BIG-Bench Hard) 62.1 % 3-shot CoT
GPQA 36.2 % 0-shot
IFEval 55.2 % prompt_strict
ARC 87.5 % challenge
MUSR 38.4 % 0-shot
WinoGrande 76.9 % 0-shot

Pricing

Input Output Cache Read Cache Write

per million tokens

Strengths

  • HumanEval 82.1, excellent code generation.

Weaknesses

  • Proprietary, not self-hostable.
  • Context window 16K is limited.

Use Cases

  • Code generation and debugging

References