CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-11-04 09:32:00 +00:00
Code Issues Packages Projects Releases Wiki Activity

Update convert-llama-h5-to-gguf.py

Browse Source
This commit is contained in:
klosax
2023-07-29 16:47:00 +02:00
committed by GitHub
parent 0317c41d98
commit 8ad7cd49fb
1 changed files with 17 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

18
convert-llama-h5-to-gguf.py
View File

@@ -120,7 +120,7 @@ if Path(dir_model + "/tokenizer.model").is_file():
text = tokenizer.id_to_piece(i).replace("\u2581", " ").encode("utf-8") text = tokenizer.id_to_piece(i).replace("\u2581", " ").encode("utf-8")
score: float = tokenizer.get_score(i) score: float = tokenizer.get_score(i)
tokens.append(str(text)) tokens.append(text)
scores.append(score) scores.append(score)
print("write gguf tokens") print("write gguf tokens")
@@ -184,6 +184,22 @@ for name in list_vars.keys():
name = "layers." + str(i) + ".feed_forward.w3.weight" name = "layers." + str(i) + ".feed_forward.w3.weight"
break break
n_dims = len(data.shape)
# ftype == 0 -> float32, ftype == 1 -> float16
ftype_cur = 0
if ftype != 0:
if name.endswith(".weight") and n_dims == 2:
data = data.astype(np.float16)
ftype_cur = 1
else:
data = data.astype(np.float32)
ftype_cur = 0
else:
if data.dtype != np.float32:
data = data.astype(np.float32)
ftype_cur = 0
gguf_writer.write_tensor_info(name, data) gguf_writer.write_tensor_info(name, data)