tqcq/llama.cpp
0
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-08-17 21:51:27 -04:00
Code Issues Packages Projects Releases Wiki Activity

imatrix : fix conversion problems

Browse Source
This commit is contained in:
Francis Couture-Harpin
2024-09-08 10:04:01 -04:00
parent 3de9300c37
commit c8ab6a3ba3
3 changed files with 14 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
convert_legacy_imatrix_to_gguf.py
View File

@@ -64,10 +64,11 @@ class IMatrixReader:
data = self._get(np.float32, nval)
assert name not in self.entries, f"duplicated name: {name!r}"
self.entries[name] = IMatrixEntry(data, np.array([ncall * self.chunk_size], dtype=np.float32))
self.entries[name] = IMatrixEntry(data * np.float32(self.chunk_size), np.array([ncall * self.chunk_size], dtype=np.float32))
self.chunk_count = self._get(np.int32).item()
self.dataset = self._get(np.uint8, self._get(np.int32).item()).tobytes().decode("utf-8")
dataset_len = self._get(np.int32).item()
self.dataset = self._get(np.uint8, dataset_len).tobytes().decode("utf-8")
def to_writer(self, outfile: Path) -> IMatrixWriter:
writer = IMatrixWriter(path=outfile, arch="")
@@ -110,6 +111,9 @@ if __name__ == "__main__":
input_file: Path = args.imatrix
if input_file.suffix != ".gguf":
args.outfile = input_file.with_suffix(".gguf")
if args.outfile.exists():
logger.error(f"default file exists, specify with --outfile to overwrite: {args.outfile}")
exit(1)
writer = IMatrixReader(args.imatrix).to_writer(args.outfile)