transformers
8edd87b6 - 🚨 EP: fix EP router contract for many models + honor FP8 scale format (#46818)

Commit
2 days ago
🚨 EP: fix EP router contract for many models + honor FP8 scale format (#46818) * honor the quant config's scale format and refuse * fix fp4 specific * strict * deeper ep fix * test * style * add assertion * more ep plans * fold tp+ep checks and ep assert into one helper * style * rasie propper error upon ep request with no ep plan * address anton's comments and make more modular * fix repo * more modular * more modular dsv2 topK router * modular phimoe router * fix * reverting phimoe changes * last modular attempt * correct fix ? * add BC variation just in case * clearer message * post init workaround? * remove the warning * fix CI * ci * fixup modular so we don't need to mess up more attribute maps * fix glm4v moe * fix exaone --------- Co-authored-by: vasqu <antonprogamer@gmail.com> Co-authored-by: Anton Vlasjuk <73884904+vasqu@users.noreply.github.com>
Parents
Loading