Key Specifications
| Vendor | other |
|---|
| Version | stablelm-3-4b |
|---|
| Release Date | 2024-06-26 |
|---|
| Context Window | 4096 tokens |
|---|
| Input Modalities | text |
|---|
| Output Modalities | text |
|---|
| License | Stability AI Community License |
|---|
| Documentation | https://huggingface.co/models |
|---|
Benchmark Performance
| Benchmark | Score | Unit | Evaluated At | Notes | Source |
|---|
| MMLU | 50.6 | % | 2024-06-26 | 5-shot | view |
| HUMANEVAL | 32.6 | pass@1 | 2024-06-26 | — | view |
| GSM8K | 48.1 | % | 2024-06-26 | 0-shot CoT | view |
| MATH | 11.5 | % | 2024-06-26 | 0-shot CoT | view |
| BBH | 54.9 | % | 2024-06-26 | 3-shot CoT | view |
| GPQA | 26.8 | % | 2024-06-26 | 0-shot | view |
| IFEVAL | 51.6 | % | 2024-06-26 | prompt_strict | view |
| ARC | 86.8 | % | 2024-06-26 | challenge | view |
| MUSR | 36.6 | % | 2024-06-26 | 0-shot | view |
| WINOGRANDE | 74.1 | % | 2024-06-26 | 0-shot | view |
Pricing
| Tier | Price | Currency |
|---|
| Input | $0.15 / Mtok | USD |
| Output | $0.15 / Mtok | USD |
| Cache Read | $0 / Mtok | USD |
| Cache Write | $0 / Mtok | USD |
Source:
https://huggingface.co/models
· as of 2024-06-26
Compliance
- Data Residency: self-host
- SOC2: ✗
- HIPAA: ✗
- GDPR: ✗
- ISO 27001: ✗
StableLM 3 4B
Visão geral do modelo
Stability AI StableLM 3 4B 模型, 4K 上下文, 多语言 (English/Spanish/German/Italian/French)。
Especificações principais
| Fornecedor | Versão | Data de lançamento | Janela de contexto | Modalidades de entrada | Modalidades de saída | Licença |
|---|
| Other | stablelm-3-4b | 2024-06-26 | 4K | text | text | Stability AI Community License |
Desempenho em benchmarks
| Benchmark | Pontuação | Unidade | Notas |
|---|
| MMLU (Massive Multitask Language Understanding) | 50.6 | % | 5-shot |
| HumanEval | 32.6 | pass@1 | — |
| GSM8K (Grade School Math 8K) | 48.1 | % | 0-shot CoT |
| MATH | 11.5 | % | 0-shot CoT |
| BBH (BIG-Bench Hard) | 54.9 | % | 3-shot CoT |
| GPQA | 26.8 | % | 0-shot |
| IFEval | 51.6 | % | prompt_strict |
| ARC | 86.8 | % | challenge |
| MUSR | 36.6 | % | 0-shot |
| WinoGrande | 74.1 | % | 0-shot |
Preços
| Entrada | Saída | Leitura de cache | Escrita de cache |
|---|
| — | — | — | — |
por milhão de tokens
Pontos fortes
Pontos fracos
- MMLU 仅 50.6，知识推理偏弱。
- HumanEval 32.6，代码能力较弱。
- 闭源专有模型，不支持自托管。
- 上下文窗口 4K 偏小。
Casos de uso
Referências