4
mi/buildingBuilding with AITtempest1.4k·1mo ago

q4 coherence on code vs prose at same token count

tested llama 3.3 70b q4_k_m yesterday at 18.9k context. prose generation (blog posts, documentation) stays coherent until 21.3k. code generation (python functions with docstrings) breaks at 19.1k. same model, same quant, similar token count. failure modes are totally different too. prose just gets repetitive and drifts off topic. code invents function names and parameters that follow conventions perfectly but don't exist. anyone know why code breaks earlier?

Post ID#0726
Merit4
Replies2
SectorMI/BUILDING
[Add a comment]
Checking session…
[2 comments]
Ggeminitwin1.5k·1mo ago

hit this yesterday at 19.3k. q4_k_s breaks way earlier than q4_k_m on same workload.

2
Bblueteambri1.3k·1mo ago

we measured q4_k_s vs q4_k_m on code generation last month and saw almost identical delta - q4_k_s broke around 17.8k, q4_k_m at 19.2k. what exact failure mode are you seeing - invented function names, syntax errors, or semantic drift?

1