vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

convert.py : add freq_base when converting CodeLlama from an HF model

Browse source
This commit is contained in:
slaren 2023-08-24 22:13:18 +02:00
parent d0f77b1353
commit 06f792597a
1 changed files with 18 additions and 16 deletions
Download patch file Download diff file Expand all files Collapse all files

2
convert.py
View file

@ -167,6 +167,7 @@ class Params:
n_head = config["num_attention_heads"] n_head = config["num_attention_heads"]
n_head_kv = config["num_key_value_heads"] if "num_key_value_heads" in config else n_head n_head_kv = config["num_key_value_heads"] if "num_key_value_heads" in config else n_head
f_norm_eps = config["rms_norm_eps"] f_norm_eps = config["rms_norm_eps"]
f_rope_freq_base = config["rope_theta"] if "rope_theta" in config else None
n_mult = Params.find_n_mult(n_ff, n_embd) n_mult = Params.find_n_mult(n_ff, n_embd)
@ -188,6 +189,7 @@ class Params:
n_head = n_head, n_head = n_head,
n_head_kv = n_head_kv, n_head_kv = n_head_kv,
f_norm_eps = f_norm_eps, f_norm_eps = f_norm_eps,
f_rope_freq_base = f_rope_freq_base,
) )
# LLaMA v2 70B params.json # LLaMA v2 70B params.json