How to use squaredcuber/holo-gamedev-lora-arm-a with PEFT:
from peft import PeftModel from transformers import AutoModelForCausalLM base_model = AutoModelForCausalLM.from_pretrained("/workspace/model") model = PeftModel.from_pretrained(base_model, "squaredcuber/holo-gamedev-lora-arm-a")
8x DDP SFT on the cybus game-dev + grounding mix, 185 steps @ 7168 seq, eval_loss 0.4275. Load on top of Hcompany/Holo-3.1-35B-A3B.
Hcompany/Holo-3.1-35B-A3B
Base model
