I’ve been using airoboros-l2-70b for writing fiction, and while overall I’d describe the results as excellent and better than any llama1 model I’ve used, it doesn’t seem to be living up to the promise of 4k token sequence length.
Around 2500 tokens output quality degrades rapidly, and either starts repeating previous text verbatim, or becomes incoherent (grammar, punctuation and capitalization disappear, becomes salad of vaguely related words)
Any other experiences with llama2 and long context? Does the base model work better? Are other fine tunes behaving similarly? I’ll try myself eventually, but the 70b models are chunky downloads, and experimentation takes a while at 1 t/s.
(I’m using GGML Q4_K_M on kobold.cpp, with rope scaling off like you’re supposed to do with llama2)
I was unaware that the smaller context models exhibited the same effect. It does seem logical that broad important information and conclusions is naturally put at the ends of a sentence by us. I haven’t read the paper yet, but wonder if the training set - our communication - also contains more information at the ends, so the effect isn’t caused by the algorithm, but by the data. I’ll give the paper a read, thx…