Merge pull request #11 from veden/patch-1

Fix incorrect ratio calculation for draft model
This commit is contained in:
turboderp
2023-11-20 04:23:34 +01:00
committed by GitHub

View File

@@ -101,8 +101,6 @@ class ModelContainer:
self.draft_config.model_dir = str(draft_model_path.resolve())
self.draft_config.prepare()
self.draft_config.max_seq_len = self.config.max_seq_len
if "draft_rope_alpha" in kwargs:
self.draft_config.scale_alpha_value = kwargs.get("draft_rope_alpha") or 1
else:
@@ -110,6 +108,8 @@ class ModelContainer:
alpha = -0.13436 + 0.80541 * ratio + 0.28833 * ratio ** 2
self.draft_config.scale_alpha_value = alpha
self.draft_config.max_seq_len = self.config.max_seq_len
if "chunk_size" in kwargs:
self.draft_config.max_input_len = kwargs["chunk_size"]
self.draft_config.max_attn_size = kwargs["chunk_size"] ** 2