GRPOTrainer/async: fix prefix EOS slicing for tool suffix (with Qwen3/3.5 type of chat templates)#5330

qgallouedec merged 12 commits into

huggingface:mainhuggingface/trl:mainfrom

casinca:qwen3.5-gen-fixcasinca/trl:qwen3.5-gen-fixCopy head branch name to clipboard