A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 8, 05:00 UTC
Revert "[Misc] fix qwen3.5 config" (#34610)
Roger Wang committed
Feb 16, 2026 at 01:06 UTC
b5475d0534421df9eb93a67f046462cdaed43d1d
2 files changed
+4
-4
vllm/transformers_utils/configs/qwen3_5.py
+2
-2
@@ -68,10 +68,10 @@ class Qwen3_5TextConfig(PretrainedConfig):
68
eos_token_id=None,
69
**kwargs,
70
):
71
- kwargs["ignore_keys_at_rope_validation"] = {
71
+ kwargs["ignore_keys_at_rope_validation"] = [
72
"mrope_section",
73
"mrope_interleaved",
74
- }
74
+ ]
75
self.vocab_size = vocab_size
76
self.max_position_embeddings = max_position_embeddings
77
self.hidden_size = hidden_size
vllm/transformers_utils/configs/qwen3_5_moe.py
+2
-2
@@ -75,10 +75,10 @@ class Qwen3_5MoeTextConfig(PretrainedConfig):
75
eos_token_id=None,
76
**kwargs,
77
):
78
- kwargs["ignore_keys_at_rope_validation"] = {
78
+ kwargs["ignore_keys_at_rope_validation"] = [
79
"mrope_section",
80
"mrope_interleaved",
81
- }
81
+ ]
82
self.vocab_size = vocab_size
83
self.max_position_embeddings = max_position_embeddings
84
self.hidden_size = hidden_size