Instructions to use api-service-sac/s1-code-v1 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Laya
How to use api-service-sac/s1-code-v1 with Laya:
# No code snippets available yet for this library. # To use this model, check the repository files and the library's documentation. # Want to help? PRs adding snippets are welcome at: # https://github.com/huggingface/huggingface.js
- Notebooks
- Google Colab
- Kaggle
s1-code v1
First version of s1-code. For use, prefer v3. A System One decision model (322M, CPU) that returns the probability that a Python function answers a search. Fine-tuned from convaiinnovations/laya (Apache 2.0).
Same input format as v3: the state is the file path, the function name, an empty line and the first 1,500 characters
of the source; the question is the noul This code answers the search: <your search>.
Results (new held-out test, 197 questions, 25 candidates from Qwen3-Embedding)
| System | Top 1 | English | Spanish |
|---|---|---|---|
| s1-code v1 alone | 144 | 68 % | 78 % |
| s1-code v1 + Qwen3-Embedding (fused) | 167 | 79 % | 91 % |
| Qwen3-Embedding alone | 152 | 76 % | 78 % |
Trained on 30,068 generated questions (59 % Spanish, 41 % English) for functions of 306 public Python repositories with permissive licences, with granite-mined negatives. Stronger in Spanish than in English, which v2 and v3 corrected.
Licence
Apache 2.0. Fine-tuned from Laya by Convai Innovations.
- Downloads last month
- 5
Model tree for api-service-sac/s1-code-v1
Base model
convaiinnovations/laya