| import argparse,json,struct,sys |
| from pathlib import Path |
| import numpy as np,torch |
| ROOT=Path(__file__).resolve().parents[1];sys.path.insert(0,str(ROOT)) |
| from src.model import EXLLM,EXLLMConfig |
| MAGIC=b'EXLLM8\0\0' |
| def load_bin(path,manifest_path): |
| man=json.loads(Path(manifest_path).read_text(encoding='utf-8'));m=EXLLM(EXLLMConfig(**man['config']));vals={} |
| with open(path,'rb') as f: |
| if f.read(8)!=MAGIC:raise ValueError('bad magic') |
| ver,n=struct.unpack('<II',f.read(8)); |
| if ver!=1:raise ValueError('unsupported format') |
| for _ in range(n): |
| nl,nd,qt=struct.unpack('<HBB',f.read(4));name=f.read(nl).decode();dims=[struct.unpack('<I',f.read(4))[0] for _ in range(nd)];ns,nb=struct.unpack('<II',f.read(8));sc=np.frombuffer(f.read(ns*4),dtype='<f4').copy() if ns else np.array([],dtype=np.float32);raw=f.read(nb) |
| if qt==1:a=np.frombuffer(raw,dtype=np.int8).reshape(dims).astype(np.float32)*sc[:,None] |
| elif qt==2:a=np.frombuffer(raw,dtype='<f2').reshape(dims).astype(np.float32) |
| else:raise ValueError(qt) |
| vals[name]=torch.from_numpy(a.copy()) |
| sd=m.state_dict();aliases=man.get('aliases',{}) |
| for k in sd: |
| src=aliases.get(k,k) |
| if src in vals:sd[k]=vals[src] |
| m.load_state_dict(sd);m.eval();return m,man |
| def main(): |
| ap=argparse.ArgumentParser();ap.add_argument('--bin',default='weights/EXLLM-v1.1-5m-int8.bin');ap.add_argument('--manifest',default='weights/EXLLM-v1.1-5m-int8.manifest.json');ap.add_argument('--out',default='weights/EXLLM-v1.1-5m-int8-dequant-test.pt');a=ap.parse_args();m,man=load_bin(ROOT/a.bin,ROOT/a.manifest);torch.save({'model':m.state_dict(),'config':m.cfg.__dict__,'meta':{'name':'EXLLM','developer':'ToTo','version':'1.1.0-5m','quantized_source':a.bin}},ROOT/a.out);print(a.out) |
| if __name__=='__main__':main() |
|
|
