From: jaeminSon <redacted>
Date: Wed, 17 May 2023 15:49:37 +0000 (+0900)
Subject: examples : fix a hyperparameter value in gpt-neox (#161) (#162)
X-Git-Tag: upstream/0.0.1642~1466
X-Git-Url: https://git.djapps.eu/?a=commitdiff_plain;h=74705055853f7922e9622bdd0a1ebde2b8f57431;p=pkg%2Fggml%2Fsources%2Fggml

examples : fix a hyperparameter value in gpt-neox (#161) (#162)
---

diff --git a/examples/gpt-neox/convert-h5-to-ggml.py b/examples/gpt-neox/convert-h5-to-ggml.py
index 0019810e..9d21226f 100644
--- a/examples/gpt-neox/convert-h5-to-ggml.py
+++ b/examples/gpt-neox/convert-h5-to-ggml.py
@@ -58,7 +58,7 @@ fout.write(struct.pack("i", hparams["hidden_size"]))
 fout.write(struct.pack("i", hparams["num_attention_heads"]))
 fout.write(struct.pack("i", hparams["num_hidden_layers"]))
 fout.write(struct.pack("i", int(hparams["rotary_pct"]*(hparams["hidden_size"]//hparams["num_attention_heads"]))))
-fout.write(struct.pack("i", hparams["use_parallel_residual"]))
+fout.write(struct.pack("i", hparams["use_parallel_residual"] if "use_parallel_residual" in hparams else True))
 fout.write(struct.pack("i", ftype))
 
 # TODO: temporary hack to not deal with implementing the tokenizer