import sys, json from lab_common import * from transformers import AutoModelForCausalLM, BitsAndBytesConfig name, prot, out = sys.argv[1], json.loads(sys.argv[2]), sys.argv[3] skip = ["lm_head"] + [f"model.layers.{i}" for i in prot] bq = BitsAndBytesConfig(load_in_4bit=True, bnb_4bit_quant_type="nf4", bnb_4bit_compute_dtype=torch.float16, bnb_4bit_use_double_quant=False, llm_int8_skip_modules=skip) m = AutoModelForCausalLM.from_pretrained(MID, token=HF, quantization_config=bq, dtype=torch.float16, device_map={"": 0}).eval() fp16, nf4, other = [], [], [] for i, ly in enumerate(m.model.layers): kinds = {type(mod).__name__ for mod in ly.modules() if isinstance(mod, torch.nn.Linear)} (fp16 if kinds == {"Linear"} else nf4 if kinds == {"Linear4bit"} else other).append(i) print("layer layout -> fp16:", fp16, "| nf4:", len(nf4), "layers | mixed/other:", other, flush=True) assert not other and fp16 == sorted(prot) and len(nf4) == len(m.model.layers) - len(prot), f"layout mismatch: wanted fp16 layers {sorted(prot)}, got {fp16}" m.save_pretrained(out); tok.save_pretrained(out) json.dump({"arm": name, "fp16_layers": sorted(prot), "other_layers": "NF4 (bitsandbytes, blocksize 64, no double quant, fp16 compute)"}, open(out + "/safecompress_layout.json", "w")) print("saved", out, flush=True)