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

convert-falcon-hf-to-gguf.py : update ref

Browse Source
This commit is contained in:
klosax
2023-08-19 01:08:39 +02:00
committed by GitHub
parent 1d80eea574
commit 2c8055b65b
1 changed files with 2 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
convert-falcon-hf-to-gguf.py
View File

@@ -193,11 +193,8 @@ if Path(dir_model + "/tokenizer.json").is_file():
tensor_map = gguf.get_tensor_name_map(ARCH,block_count)
# params for qkv transform
if "n_head_kv" in hparams:
n_head_kv = hparams["n_head_kv"]
else:
n_head_kv = 1
n_head = hparams["n_head"]
n_head_kv = hparams["n_head_kv"] if "n_head_kv" in hparams else 1
head_dim = hparams["hidden_size"] // n_head
# tensor info
@@ -231,7 +228,7 @@ for part_name in part_names:
# So we rearrange them here,, so that we have n_head query weights
# followed by n_head_kv key weights followed by n_head_kv value weights,
# in contiguous fashion.
# ref: https://github.com/jploski/ggml/blob/d5295b477fb36c69468c3fecb0393a8d7980b7c8/examples/falcon/convert-hf-to-ggml.py#L107-L123
# ref: https://github.com/jploski/ggml/blob/falcon40b/examples/falcon/convert-hf-to-ggml.py
if "query_key_value" in name:
qkv = data.view(n_head_kv, n_head // n_head_kv + 2, head_dim, head_dim * n_head)