Qelm Theoretical Beta 01

[Inserian]

This release focus's on fixing previous run issues as well as new variables to create more in-depth encoded models for larger language tasks.

Entropy Handling Added:
• Introduced an entropy factor in the QuantumChannel class with a new _apply_entropy_mixing method that applies random RY rotations based on the specified entropy factor.

Sub-bit Encoding
• Added new sub-bit encoding techniques and layers to better control input and output.
• Increased channel complexity to better encode and decode sub-bit layers.

Statevector Saving Fix:
• Added the helper function ensure_single_statevector(circuit) to remove duplicate "save_statevector" instructions.
• Updated all simulation calls (in methods such as decode, decode_subbit, and simulate) to call this helper for consistent statevector retrieval.

Transformer Block Output Adjustment:
• Modified the output of the QuantumTransformerBlock.forward method so that if it returns a scalar value, it is replicated into a full vector of length equal to the embedding dimension. This change is meant to preserve the representation required for next-token prediction.

Parameter Management Improvements:
• Revised the implementations of get_all_parameters and set_all_parameters in the attention, feed-forward, and transformer block classes to ensure correct concatenation, reshaping, and assignment of quantum circuit parameters.

Logging Enhancements and Error-Handling Refinements:
• Improved error messages and logging throughout the script (ensuring that the logging module is properly imported and used) to provide more helpful debugging information.
• Added extra error-checks (such as verifying input token ID ranges) and clearer inline comments without altering the overall architecture.

Fixed duplicate issues and calls.

Fixed chat-ui tokenization errors.

Full Changelog: V2-Alpha...Theoretical-Beta-01

---

This discussion was created from the release Qelm Theoretical Beta 01.

[Inserian]

Future release will include Quantum's connection so this can be run on IBM's QPU's.

[Inserian]

If you are having problems with cmd popping up you can build it from scratch using pyinstaller using the no window cmd, however I will release a fixed version shortly.