ARKA Advanced SLM Base v2
ARKA stands for Academic Research Knowledge Assistant.
This repository contains a clean consolidated ARKA model.
It contains the final learned model weights and the files required to load the model with Hugging Face Transformers.
Model
- Parameters: 124,439,808
- Source snapshot:
Abhayn01/arka-advanced-slm-identity-squad-v2
Purpose
This repository is intended to serve as a clean base for future ARKA training and experimentation.
Loading
Use AutoTokenizer and AutoModelForCausalLM from Transformers.
Clean Base Policy
This repository intentionally does not contain:
- optimizer states
- training checkpoints
- recovery checkpoints
- scheduler states
- AMP scaler states
- milestone checkpoints
- historical training-state files
Only the consolidated learned model weights are retained.
Status
This is a development base model. Further factuality, hallucination, reasoning and retention evaluation is recommended.