Mirrors/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggerganov/llama.cpp.git synced 2025-01-12 05:17:21 +01:00
Code Issues Packages Projects Releases Wiki Activity

convert : fix F32 ftype not being saved (#3048)

Browse Source
This commit is contained in:
Cebtenzzre 2023-09-07 14:27:42 -04:00 committed by GitHub
parent 00d62adb79
commit 6336d834ec
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 6 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

12
convert.py
View File

@ -266,7 +266,7 @@ class Params:
f_rope_freq_base = config["rope_theta"] if "rope_theta" in config else None f_rope_freq_base = config["rope_theta"] if "rope_theta" in config else None
# hack to determine LLaMA v1 vs v2 vs CodeLlama # hack to determine LLaMA v1 vs v2 vs CodeLlama
if f_rope_freq_base and f_rope_freq_base == 1000000: if f_rope_freq_base == 1000000:
# CodeLlama # CodeLlama
n_ctx = 16384 n_ctx = 16384
elif config["norm_eps"] == 1e-05: elif config["norm_eps"] == 1e-05:
@ -841,9 +841,9 @@ class OutputFile:
name = "LLaMA" name = "LLaMA"
# TODO: better logic to determine model name # TODO: better logic to determine model name
if (params.n_ctx == 4096): if params.n_ctx == 4096:
name = "LLaMA v2" name = "LLaMA v2"
elif params.path_model: elif params.path_model is not None:
name = str(params.path_model.parent).split('/')[-1] name = str(params.path_model.parent).split('/')[-1]
self.gguf.add_name (name) self.gguf.add_name (name)
@ -856,13 +856,13 @@ class OutputFile:
self.gguf.add_head_count_kv (params.n_head_kv) self.gguf.add_head_count_kv (params.n_head_kv)
self.gguf.add_layer_norm_rms_eps (params.f_norm_eps) self.gguf.add_layer_norm_rms_eps (params.f_norm_eps)
if params.f_rope_freq_base: if params.f_rope_freq_base is not None:
self.gguf.add_rope_freq_base(params.f_rope_freq_base) self.gguf.add_rope_freq_base(params.f_rope_freq_base)
if params.f_rope_scale: if params.f_rope_scale is not None:
self.gguf.add_rope_scale_linear(params.f_rope_scale) self.gguf.add_rope_scale_linear(params.f_rope_scale)
if params.ftype: if params.ftype is not None:
self.gguf.add_file_type(params.ftype) self.gguf.add_file_type(params.ftype)
def add_meta_vocab(self, vocab: Vocab) -> None: def add_meta_vocab(self, vocab: Vocab) -> None: