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 9, 06:15 UTC
[Bugfix] Fix incompatibility between #33372 and #32863 (#33475)
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
Cyrus Leung committed
Jan 31, 2026 at 21:21 UTC
4cb59dea6a9b54cd3cae87ac69fb0a836615f196
4 files changed
+1
-6
vllm/renderers/deepseek_v32.py
-2
@@ -71,7 +71,6 @@ class DeepseekV32Renderer(RendererLike):
71
content_format="string",
72
)
73
74
- kwargs["return_dict"] = False
74
prompt_raw = tokenizer.apply_chat_template(
75
conversation=conversation,
76
messages=messages,
@@ -98,7 +97,6 @@ class DeepseekV32Renderer(RendererLike):
97
content_format="string",
98
)
99
101
- kwargs["return_dict"] = False
100
prompt_raw = tokenizer.apply_chat_template(
101
conversation=conversation,
102
messages=messages,
vllm/renderers/grok2.py
-2
@@ -71,7 +71,6 @@ class Grok2Renderer(RendererLike):
71
content_format="string",
72
)
73
74
- kwargs["return_dict"] = False
74
prompt_raw = tokenizer.apply_chat_template(
75
conversation=conversation,
76
messages=messages,
@@ -98,7 +97,6 @@ class Grok2Renderer(RendererLike):
97
content_format="string",
98
)
99
101
- kwargs["return_dict"] = False
100
prompt_raw = tokenizer.apply_chat_template(
101
conversation=conversation,
102
messages=messages,
vllm/renderers/hf.py
-1
@@ -466,7 +466,6 @@ def safe_apply_chat_template(
466
chat_template=chat_template,
467
chat_template_kwargs=kwargs,
468
)
469
- resolved_kwargs["return_dict"] = False
469
470
try:
471
return tokenizer.apply_chat_template(
vllm/renderers/params.py
+1
-1
@@ -66,7 +66,7 @@ class ChatParams:
66
"""The arguments to pass to `tokenizer.apply_chat_template`."""
67
return merge_kwargs(
68
self.chat_template_kwargs,
69
- dict(chat_template=self.chat_template),
69
+ dict(chat_template=self.chat_template, return_dict=False),
70
)
71
72