An open-source sandbox for PKLM Core is now available on GitHub. The project provides lightweight Hugging Face logits processing with zero-drift linguistic containment, designed for local pipeline integration.
Developers can integrate the sandbox processor with a few lines of code:
from transformers import AutoModelForCausalLM, AutoTokenizer
from pklm_sandbox.processor import PKLMSandboxProcessor
processor = PKLMSandboxProcessor(blocked_token_ids=[50256])The public repository offers a basic developer-tier wrapper intended for testing and evaluation. The project is distributed under the MIT License.
For high-stakes enterprise environments, the repository notes that custom token-level constraint modeling, heavy Paninian karaka rule engines, low-latency compiled binaries, and production SLAs are available. Organizations requiring these capabilities should contact the team directly at the provided email address.
The sandbox features zero-drift linguistic containment, which prevents unintended language drift during model operation. Token processing occurs at the Hugging Face logits level, allowing developers to block specific token IDs during generation. The MIT-licensed code serves as a foundation for further customization in research and development workflows.