PrimeIntellect / Qwen3-4B-Instruct-2507-untied

Untied copy of Qwen/Qwen3-4B-Instruct-2507. Same weights; lmhead.weight is stored as a copy of the input embedding and tiewordembeddings is false, so trainers that don't support tied LM heads (e.g. prime-rl) can load it. Made with tools/untiewordembeddings.py from prime-rl.