Mirrors/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggerganov/llama.cpp.git synced 2025-01-07 11:23:56 +01:00
Code Issues Packages Projects Releases Wiki Activity

convert : rm quantization version

Browse Source
This commit is contained in:
M. Yusuf Sarıgöz 2023-08-13 14:38:53 +03:00
parent 1d60468eee
commit bf2dad3100
1 changed files with 48 additions and 47 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
convert-llama-h5-to-gguf.py
View File

@ -23,6 +23,7 @@ def permute(weights: NDArray, n_head: int) -> NDArray:
.swapaxes(1, 2)
.reshape(weights.shape))
def count_model_parts(dir_model: str) -> int:
num_parts = 0
for filename in os.listdir(dir_model):
@ -33,6 +34,7 @@ def count_model_parts(dir_model: str) -> int:
print("gguf: found " + str(num_parts) + " model parts")
return num_parts
if len(sys.argv) < 3:
print("Usage: convert-h5-to-ggml.py dir-model ftype\n")
print(" ftype == 0 -> float32")
@ -86,7 +88,6 @@ block_count = hparams["num_hidden_layers"]
gguf_writer.add_name(last_dir)
gguf_writer.add_architecture(llm_arch)
gguf_writer.add_quantization_version(ftype)
guff_writer.add_source_hf_repo(hf_repo)
gguf_writer.add_context_length(llm_arch, hparams["max_position_embeddings"])
gguf_writer.add_embedding_length(llm_arch, hparams["hidden_size"])