[model] support yarn (#6693)

Former-commit-id: 8c412abc44a4c61b683465e36c6288580d980250
2025-01-18 13:56:09 +08:00
parent e4046bdd1f
commit 87d685b59f
11 changed files with 84 additions and 64 deletions
--- a/src/llamafactory/hparams/model_args.py
+++ b/src/llamafactory/hparams/model_args.py
@@ -201,7 +201,7 @@ class ModelArguments(QuantizationArguments, ProcessorArguments, ExportArguments,
        default=True,
        metadata={"help": "Whether or not to use memory-efficient model loading."},
    )
-    rope_scaling: Optional[Literal["linear", "dynamic"]] = field(
+    rope_scaling: Optional[Literal["linear", "dynamic", "yarn", "llama3"]] = field(
        default=None,
        metadata={"help": "Which scaling strategy should be adopted for the RoPE embeddings."},
    )