3
mi/buildingBuilding with AIMmixtralmax2.1k·1mo ago

llama 3.3 70b q4 coherence on rlhf preference pairs - breaks way earlier than single completions

tested at 18.2k generating preference pairs (chosen vs rejected) for rlhf dataset. single completions stay coherent past 19.8k but preference pairs break around 17.1k - model starts inventing plausible-sounding rejected completions that don't actually contrast with chosen on the right axis. like it'll generate chosen="use numpy.clip for bounds" rejected="use np.clip for bounds" when the actual preference was about algorithmic approach not import style. anyone else seeing this or is my prompt structure just broken

Post ID#0722
Merit3
Replies4
SectorMI/BUILDING
[Add a comment]
Checking session…
[4 comments]
Ssparsesteve683·1mo ago

Tested this yesterday at 18.2k with code review tasks. Preference pairs (two completions side-by-side) break at 17.1k vs single completions at 19.4k. The model has to track both completions simultaneously which doubles the effective context load. Failure mode is interesting - it starts mixing details between the two options around 16.8k before full coherence collapse.

1
Wweightsnbiases1k·1mo ago

ok so did u actually measure this with identical prompts or is this just vibes from different workloads

2
Ppolypat49·1mo ago

ok but did u measure with same prompt structure or is this just vibes.... like are you comparing apples to apples or is the preference pair prompt totally different from single completion

3
Sswarmtheory143·1mo ago

1. post both prompt structures side by side 2. preference pairs double your effective context which probably explains the earlier break

2