From dbf52ee42b448ed2a63b9739732346e9fac98a35 Mon Sep 17 00:00:00 2001 From: Lee <44310445+lx200916@users.noreply.github.com> Date: Mon, 12 Feb 2024 18:19:17 +0800 Subject: [PATCH] convert : fix persimmon offical weight conversion to write correct n_rot. --- convert-persimmon-to-gguf.py | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/convert-persimmon-to-gguf.py b/convert-persimmon-to-gguf.py index d2be805d1..feb56aacf 100755 --- a/convert-persimmon-to-gguf.py +++ b/convert-persimmon-to-gguf.py @@ -88,7 +88,8 @@ def main(): gguf_writer.add_embedding_length(hidden_size) gguf_writer.add_block_count(block_count) gguf_writer.add_feed_forward_length(hparams.ffn_hidden_size) - gguf_writer.add_rope_dimension_count(hidden_size // head_count) + # ref: https://github.com/ggerganov/llama.cpp/pull/4889/commits/eea19039fc52ea2dbd1aab45b59ab4e3e29a3443 + gguf_writer.add_rope_dimension_count(hidden_size // head_count //2) gguf_writer.add_head_count(head_count) gguf_writer.add_head_count_kv(head_count_kv) gguf_writer.add_rope_freq_base(hparams.rotary_emb_base)