At this time, this repo is unfinished. Although passing our own smoke tests and various other metrics, the platform is still not able to get anything but mostly-garbled output out of the sample QWEN LLM. At this time, I believe v0.8b begins to solve this - the LLM needs to be re-trained for the HDGL / Analog-over-digital framework. But at this time, just like solving for primes, there’s no way of knowing if it works until a massive amount of compute is thrown at it. Such as it is, bigger fish to fry, right now…
