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

convert.py : try to determine n_ctx automatically for CodeLlama (#2770)

Browse source
This commit is contained in:
slaren 2023-08-24 21:10:39 +02:00 committed by GitHub
parent 0d3094f0c7
commit d0f77b1353
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
1 changed files with 11 additions and 1 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

12
convert.py
View file

@ -200,13 +200,23 @@ class Params:
n_embd = config["dim"]
n_layer = config["n_layers"]
n_mult = config["multiple_of"]
n_ctx = 2048 if config["norm_eps"] == 1e-06 else 4096 # hack to determine LLaMA v1 vs v2
n_ff = -1
n_head = config["n_heads"]
n_head_kv = config["n_kv_heads"] if "n_kv_heads" in config else n_head
f_norm_eps = config["norm_eps"]
f_rope_freq_base = config["rope_theta"] if "rope_theta" in config else None
# hack to determine LLaMA v1 vs v2 vs CodeLlama
if f_rope_freq_base and f_rope_freq_base == 1000000:
# CodeLlama
n_ctx = 16384
elif config["norm_eps"] == 1e-05:
# LLaMA v2
n_ctx = 4096
else:
# LLaMA v1
n_ctx = 2048
if n_vocab == -1:
n_vocab = model["tok_embeddings.weight"].shape[0]