[Model] LoRA with lm_head and embed_tokens fully trained - 4

[Model] LoRA with lm_head and embed_tokens fully trained - 4 #26215