Klyra-64M-RL-OPD
Klyra-64M-RL-OPD is an experimental reinforcement-learning and policy-distillation checkpoint.
The student starts from Klyra-64M-Reasoning-Fundamentals and is distilled from three specialist branches:
- Reasoning RL
- Instruction-Following RL
- General-Capability RL
The final Online Policy Distillation stage was run for 500 steps.
Benchmark
| ARC-Easy | PIQA | OpenBookQA | HellaSwag | Social-IQA | Mean |
|---|---|---|---|---|---|
| 32.41 | 59.36 | 27.80 | 28.39 | 35.98 | 36.79 |
This checkpoint records Klyra's strongest PIQA and Social-IQA scores, but its overall five-task mean remains slightly below Klyra-64M-Reasoning-Fundamentals.
Status
Experimental.
Use this checkpoint mainly for RL, distillation, and ablation research. For general use, Klyra-64M-Reasoning-Fundamentals is currently recommended.
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.