diff options
| author | Lee <44310445+lx200916@users.noreply.github.com> | 2024-02-13 01:29:57 +0800 |
|---|---|---|
| committer | GitHub <noreply@github.com> | 2024-02-12 19:29:57 +0200 |
| commit | dbd8828eb03b9aa8d0af7e4c533d3c2f5b38aba6 (patch) | |
| tree | 904196c336bece3e9ca276af449da89a6ab576e1 | |
| parent | 43fe07c1a4f3a58612e1d9543f7c6b556710f5d0 (diff) | |
py : fix persimmon `n_rot` conversion (#5460)
* convert : fix persimmon offical weight conversion to write correct n_rot.
* Update convert-persimmon-to-gguf.py
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
| -rwxr-xr-x | convert-persimmon-to-gguf.py | 3 |
1 files changed, 2 insertions, 1 deletions
diff --git a/convert-persimmon-to-gguf.py b/convert-persimmon-to-gguf.py index d2be805d..def21053 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) |