Mirrors/text-generation-webui
1
0
Fork 0
mirror of https://github.com/oobabooga/text-generation-webui.git synced 2024-09-20 10:35:10 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions 1

Read more GGUF metadata (scale_linear and freq_base) (#3877)

Browse source
This commit is contained in:
Gennadij 2023-09-12 23:02:42 +03:00 committed by GitHub
parent 90fca6a77d
commit 460c40d8ab
WARNING! Although there is a key with this ID in the database it does not verify this commit! This commit is SUSPICIOUS.
GPG key ID: 4AEE18F83AFDEB23
3 changed files with 9 additions and 1 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

3
modules/metadata_gguf.py
View file

@ -71,6 +71,9 @@ def load_metadata(fname):
ti_data_count = struct.unpack("<Q", file.read(8))[0]
kv_data_count = struct.unpack("<Q", file.read(8))[0]
if GGUF_VERSION == 1:
raise Exception('You are using an outdated GGUF, please download a new one.')
for i in range(kv_data_count):
key_length = struct.unpack("<Q", file.read(8))[0]
key = file.read(key_length)

5
modules/models_settings.py
View file

@ -17,6 +17,7 @@ def get_fallback_settings():
'truncation_length': shared.settings['truncation_length'],
'n_ctx': 2048,
'rope_freq_base': 0,
'compress_pos_emb': 1,
}
@ -48,6 +49,10 @@ def get_model_metadata(model):
metadata = metadata_gguf.load_metadata(model_file)
if 'llama.context_length' in metadata:
model_settings['n_ctx'] = metadata['llama.context_length']
if 'llama.rope.scale_linear' in metadata:
model_settings['compress_pos_emb'] = metadata['llama.rope.scale_linear']
if 'llama.rope.freq_base' in metadata:
model_settings['rope_freq_base'] = metadata['llama.rope.freq_base']
return model_settings