summaryrefslogtreecommitdiff
path: root/convert-baichuan-hf-to-gguf.py
diff options
context:
space:
mode:
authorcebtenzzre <cebtenzzre@gmail.com>2023-11-01 18:04:33 -0400
committerGitHub <noreply@github.com>2023-11-01 18:04:33 -0400
commit898aeca90a9bb992f506234cf3b8b7f7fa28a1df (patch)
tree125f8a9b466efd4534ecd3e64419ece001c86a7d /convert-baichuan-hf-to-gguf.py
parentc43c2da8afacaddfe51c09b21dbd9922cd0ea46b (diff)
llama : implement YaRN RoPE scaling (#2268)
Co-authored-by: cebtenzzre <cebtenzzre@gmail.com> Co-authored-by: Jeffrey Quesnelle <jquesnelle@gmail.com>
Diffstat (limited to 'convert-baichuan-hf-to-gguf.py')
-rwxr-xr-xconvert-baichuan-hf-to-gguf.py3
1 files changed, 2 insertions, 1 deletions
diff --git a/convert-baichuan-hf-to-gguf.py b/convert-baichuan-hf-to-gguf.py
index 5ee99be7..67ccbe99 100755
--- a/convert-baichuan-hf-to-gguf.py
+++ b/convert-baichuan-hf-to-gguf.py
@@ -163,7 +163,8 @@ gguf_writer.add_layer_norm_rms_eps(hparams["rms_norm_eps"])
if "rope_scaling" in hparams and hparams["rope_scaling"] != None and "factor" in hparams["rope_scaling"]:
if "type" in hparams["rope_scaling"]:
if hparams["rope_scaling"]["type"] == "linear":
- gguf_writer.add_rope_scale_linear(hparams["rope_scaling"]["factor"])
+ gguf_writer.add_rope_scaling_type(gguf.RopeScalingType.LINEAR)
+ gguf_writer.add_rope_scaling_factor(hparams["rope_scaling"]["factor"])
# TOKENIZATION