Klyrune-2B-General-1
Klyrune-2B-General-1 is the first general-purpose instruction-tuned checkpoint in the Klyrune-2B project. It is fine-tuned from OpenBMB's MiniCPM5-2B-Base to develop general instruction-following, knowledge, and problem-solving capabilities.
This release represents the project's general supervised fine-tuning (SFT) stage, not a coding-specialized or autonomous-agent model. It is a fine-tune of an existing base model, not a model pretrained from scratch or an official OpenBMB release.
The Klyrune-2B project was initiated and developed by Informatics Engineering students at the State Polytechnic of Jakarta (Politeknik Negeri Jakarta).
Evaluation Results
The following scores were obtained from the project's own evaluation of Klyrune-2B-General-1.
| Benchmark | Subject / Metric | Score |
|---|---|---|
| MATH-500 | Accuracy (full 500-question set) | 70.4% (352/500) |
| MMLU | College Mathematics | 63.0% |
| MMLU | High School Mathematics | 85.6% |
| MMLU | College Computer Science | 61.0% |
| MMLU | High School Computer Science | 79.0% |
| MMLU | Computer Security | 77.0% |
| MMLU | Formal Logic | 54.0% |
| MMLU | Abstract Algebra | 58.0% |
| HumanEval | Python code generation, pass@1 | 49.4% |
The benchmarks above focus primarily on mathematics and computer science. They do not, on their own, establish comprehensive performance across all general-purpose tasks or languages.
Base model: openbmb/MiniCPM5-2B-Base
Model repository: Jahirrrr/Klyrune-2B-General-1