Hudson Cooper
Hudson Cooper
It seems that these layers CAN accept float32 input, but their outputs are always float64 (as are computed gradients)
@riedgar-ms request for review :) If the manual rewinding done here can be achieved with mechanisms built into the parser, I think that would be better. But that code is...
I think I follow what you are saying re: hidden commit points that weren't meant to be hidden, but I'm not quite sure I track 100%. My thinking is as...
Ok -- I think I see. Here's a test that **should** pass but doesn't here: ```python grammar = gen(regex=r'a*', stop='@') + gen(regex=r'b*') matchstr = 'aaaaaabbbbb' match = grammar.match(matchstr, allow_partial=True) assert...
> @hudson-ai sorry, was out sick for a bit. Yes, that's what I meant by collapsing hidden commit points. In the example you gave I think we should just throw...
@slundberg thank you for taking a look and offering your thoughts. I'll take a crack at this when I have some free time :)
@slundberg I feared you were right that there is a fundamental ambiguity... So I naturally went on a walkabout and wrote an Earley recognizer in rust :laughing: I realized that...
_Always_ wrapping commit points in `optional` seems to break generation with stop tokens, but I don't quite understand why yet. Can you point me to anything to help me wrap...
Can you give a concrete example? E.g. is it okay if the model generates a substring of one of your sentences? Can one of the sentences be a substring of...
What about ``` result = "hello worl" ``` Is that an acceptable response?