1ul1/sma-llm: sma-llm: Full-Stack Speech Memory Assistant with LLM Benchmarking
Year: 2026
Type: software
Abstract: This release archives the full sma-llm project, including: A responsive PySide6 chat UI in a concurrent architecture coordinating worker threads for LLM streaming, Speech-to-Text and Text-to-Speech, with signal-based graceful cleanup upon exit A unified interface for MLC-LLM and PyTorch backends Integrated benchmarking (TTFT, throughput, upload time, RAM usage) Commands to export conversations to PDF, view live model statistics and adjust the context memory window Bash scripts to check RAM usage, automate execution and create a RAM disk volume for saving the audio files The technical report focuses on the performance comparison between MLC-LLM and PyTorch (with MPS) inference on Apple Silicon.
Author: 1ul1
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7125591920
