Hello!
By popular demand I am planning a fine-tune of https://huggingface.co/dreamgen/opus-v0-7b on top of Yi-34B and wonder whether to use the 200K as the base.
The regular Yi-34B seems slightly better than Yi-34B-200K on standard benchmarks, but I wonder how it “feels” and whether the loss of performance on short context is worth it, given that the regular version can be used up to 32K tokens.
Did anyone try an analysis of these 2 models on various sequence lengths (<4K, <8K, <16K, etc.)?
Yi is not trustable on standard benchmarks because they are easy to game by including them in training data and the LKF gang who built this has a high pressure to justify their 1 billion dollar valuation and continue to milk investors.
The only way to really evaluate this is on some hidden benchmark never seen before and / or rigorous qualitative experiments.
Until then, I’m not holding my breath.
I believe they said they’re going to release training data. We’ll see. That’s about the only way to easily verify what made it in.