Klyra-64M-Reasoning-Fundamentals
Klyra-64M-Reasoning-Fundamentals is the current recommended Klyra-64M checkpoint.
It starts from Klyra-64M-Instruct and adds a conservative reasoning-focused SFT stage designed to improve basic reasoning while retaining general instruction capability.
Training
Training set:
- 18,000 reasoning examples
- 12,000 general-instruction replay examples
- 30,000 total examples
Focus areas include arithmetic, short multi-step reasoning, unit-rate problems, basic logic, science fundamentals, and simple algorithms.
Benchmark
| ARC-Easy | PIQA | OpenBookQA | HellaSwag | Social-IQA | Mean |
|---|---|---|---|---|---|
| 32.37 | 58.98 | 28.60 | 28.73 | 35.93 | 36.92 |
This checkpoint currently has the highest five-task mean among the released Klyra checkpoints.
Recommended Use
Use this checkpoint for:
- general Klyra inference,
- reasoning experiments,
- downstream fine-tuning,
- further RL or distillation research.
About Klyra
Klyra-64M is a compact language-model research project initiated and developed by a student of Informatics Engineering at Politeknik Negeri Jakarta (State Polytechnic of Jakarta).
The model uses a MiniMind-compatible decoder-only architecture and was trained through a staged pipeline from random initialization.
Main project: Jahirrrr/Klyra-64M
Architecture
| Component | Value |
|---|---|
| Unique runtime parameters | ~63.9M |
| Transformer layers | 8 |
| Hidden size | 768 |
| Attention heads | 8 |
| KV heads | 4 |
| Vocabulary | ~6.4K |
| Context length | 1,024 |
| MoE | No |
Parameter note: Klyra uses tied input/output embeddings at runtime. Some exported
safetensorsfiles may contain bothmodel.embed_tokens.weightandlm_head.weightas separate serialized tensors. The intended runtime model has 63,912,192 unique parameters after weight tying.