Mirrors/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggerganov/llama.cpp.git synced 2025-01-12 21:37:19 +01:00
Code Issues Packages Projects Releases Wiki Activity

Fix python stuff (#109)

Browse Source
This commit is contained in:
Georgi Gerganov 2023-03-19 19:33:18 +02:00
parent 467b149761
commit c1c7026b47
No known key found for this signature in database
GPG Key ID: 449E073F9DC10735
1 changed files with 14 additions and 23 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

37
convert-pth-to-ggml.py
View File

@ -32,7 +32,7 @@ def parse_args():
return parser.parse_args() return parser.parse_args()
def get_n_parts(dim): def get_n_parts(dim):
mappings = {4096: 1, 5120: 2, 6656: 4, 8192: 8} mappings = {4096: 1, 5120: 2, 6656: 4, 8192: 8}
n_parts = mappings.get(dim) n_parts = mappings.get(dim)
if n_parts is None: if n_parts is None:
@ -43,7 +43,7 @@ def get_n_parts(dim):
return n_parts return n_parts
def load_hparams_and_tokenizer(dir_model): def load_hparams_and_tokenizer(dir_model):
fname_hparams = f"{dir_model}/params.json" fname_hparams = f"{dir_model}/params.json"
fname_tokenizer = f"{dir_model}/../tokenizer.model" fname_tokenizer = f"{dir_model}/../tokenizer.model"
@ -57,7 +57,7 @@ def load_hparams_and_tokenizer(dir_model):
return hparams, tokenizer return hparams, tokenizer
def write_header(fout, hparams, ftype): def write_header(fout, hparams, ftype):
keys = ["vocab_size", "dim", "multiple_of", "n_heads", "n_layers"] keys = ["vocab_size", "dim", "multiple_of", "n_heads", "n_layers"]
values = [ values = [
0x67676d6c, # magic: ggml in hex 0x67676d6c, # magic: ggml in hex
@ -88,26 +88,17 @@ def write_tokens(fout, tokenizer):
def process_and_write_variables(fout, model, ftype): def process_and_write_variables(fout, model, ftype):
for name, data in model.items(): for name, datao in model.items():
if name.endswith("freqs"): if name.endswith("freqs"):
continue continue
shape = data.shape
print(f"Processing variable: {name} with shape: {shape} and type: {data.dtype}\n")
data = np.squeeze(data)
n_dims = len(shape)
# for efficiency - transpose some matrices shape = datao.shape
# "model/h.*/attn/c_attn/w"
# "model/h.*/attn/c_proj/w" print(f"Processing variable: {name} with shape: {shape} and type: {datao.dtype}")
# "model/h.*/mlp/c_fc/w"
# "model/h.*/mlp/c_proj/w" data = datao.numpy().squeeze()
#if name.endswith(("/attn/c_attn/w", "/attn/c_proj/w", "/mlp/c_fc/w", "/mlp/c_proj/w")): n_dims = len(shape)
# print("Transposing")
# data = data.transpose()
# default type is fp16 # default type is fp16
ftype_cur = 1 ftype_cur = 1
@ -122,8 +113,8 @@ def process_and_write_variables(fout, model, ftype):
for dim in reversed(data.shape): for dim in reversed(data.shape):
fout.write(struct.pack("i", dim)) fout.write(struct.pack("i", dim))
fout.write(sname) fout.write(sname)
# data # data output to file
data.tofile(fout) data.tofile(fout)
def main(): def main():
@ -139,7 +130,7 @@ def main():
for p in range(n_parts): for p in range(n_parts):
print(f"Processing part {p}\n") print(f"Processing part {p}\n")
fname_model = f"{dir_model}/consolidated.0{p}.pth" fname_model = f"{dir_model}/consolidated.0{p}.pth"
fname_out = f"{dir_model}/ggml-model-{ftype_str[ftype]}.bin{'' if p == 0 else '.' + str(p)}" fname_out = f"{dir_model}/ggml-model-{ftype_str[ftype]}.bin{'' if p == 0 else '.' + str(p)}"