nikola/llama_cpp
1
0
Fork 0
Code Issues Pull Requests Actions 17 Packages Projects Releases Wiki Activity

convert : more consistent handling of rope_parameters (#24833)

Browse Source
This commit is contained in:
Sigbjørn Skjæret
2026-06-20 13:42:36 +03:00
committed by GitHub
parent f449e05537
commit f4043fec01
15 changed files with 40 additions and 39 deletions
Download Patch File Download Diff File Expand all files Collapse all files
conversion/bailingmoe.py
+1 -1
View File
@@ -126,7 +126,7 @@ class BailingMoeV2Model(TextModel):
if (rope_dim := hparams.get("head_dim")) is None:
rope_dim = hparams["hidden_size"] // hparams["num_attention_heads"]
self.gguf_writer.add_rope_dimension_count(int(rope_dim * self.hparams.get("partial_rotary_factor", 0.5)))
self.gguf_writer.add_rope_dimension_count(int(rope_dim * self.rope_parameters.get("partial_rotary_factor", 0.5)))
self.gguf_writer.add_leading_dense_block_count(hparams["first_k_dense_replace"])
self.gguf_writer.add_vocab_size(hparams["vocab_size"])
self.gguf_writer.add_expert_feed_forward_length(hparams["moe_intermediate_size"])