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 13, 20:45 UTC
Remove deprecated `--rope-scaling` and `--rope-theta` (#28006)
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
Harry Mellor committed
Nov 4, 2025 at 10:01 UTC
2f1cc8cef17de3b7a48d9351c2b71f1ba7c75d82
2 files changed
-33
vllm/config/model.py
-27
@@ -168,12 +168,6 @@ class ModelConfig:
168
"""The specific revision to use for the model code on the Hugging Face Hub.
169
It can be a branch name, a tag name, or a commit id. If unspecified, will
170
use the default version."""
171
- rope_scaling: dict[str, Any] = field(default_factory=dict)
172
- """RoPE scaling configuration. For example,
173
- `{"rope_type":"dynamic","factor":2.0}`."""
174
- rope_theta: float | None = None
175
- """RoPE theta. Use with `rope_scaling`. In some cases, changing the RoPE
176
- theta improves the performance of the scaled model."""
171
tokenizer_revision: str | None = None
172
"""The specific revision to use for the tokenizer on the Hugging Face Hub.
173
It can be a branch name, a tag name, or a commit id. If unspecified, will
@@ -338,8 +332,6 @@ class ModelConfig:
332
factors.append(self.generation_config)
333
factors.append(self.model_impl)
334
factors.append(self.override_generation_config)
341
- factors.append(self.rope_scaling)
342
- factors.append(self.rope_theta)
335
factors.append(self.video_pruning_rate)
336
factors.append(self.enable_prompt_embeds)
337
@@ -481,25 +473,6 @@ class ModelConfig:
473
hf_overrides_kw[key] = value
474
hf_overrides_fn = None
475
484
- if self.rope_scaling:
485
- hf_override: dict[str, Any] = {"rope_scaling": self.rope_scaling}
486
- hf_overrides_kw.update(hf_override)
487
- hf_overrides_str = json.dumps(hf_overrides_kw)
488
- msg = (
489
- "`--rope-scaling` will be removed in a future release. "
490
- f"'Please instead use `--hf-overrides '{hf_overrides_str}'`"
491
- )
492
- warnings.warn(DeprecationWarning(msg), stacklevel=2)
493
- if self.rope_theta is not None:
494
- hf_override = {"rope_theta": self.rope_theta}
495
- hf_overrides_kw.update(hf_override)
496
- hf_overrides_str = json.dumps(hf_overrides_kw)
497
- msg = (
498
- "`--rope-theta` will be removed in a future release. "
499
- f"'Please instead use `--hf-overrides '{hf_overrides_str}'`"
500
- )
501
- warnings.warn(DeprecationWarning(msg), stacklevel=2)
502
-
476
self.maybe_pull_model_tokenizer_for_runai(self.model, self.tokenizer)
477
478
if (
vllm/engine/arg_utils.py
-6
@@ -438,8 +438,6 @@ class EngineArgs:
438
aggregate_engine_logging: bool = False
439
revision: str | None = ModelConfig.revision
440
code_revision: str | None = ModelConfig.code_revision
441
- rope_scaling: dict[str, Any] = get_field(ModelConfig, "rope_scaling")
442
- rope_theta: float | None = ModelConfig.rope_theta
441
hf_token: bool | str | None = ModelConfig.hf_token
442
hf_overrides: HfOverrides = get_field(ModelConfig, "hf_overrides")
443
tokenizer_revision: str | None = ModelConfig.tokenizer_revision
@@ -617,8 +615,6 @@ class EngineArgs:
615
)
616
model_group.add_argument("--revision", **model_kwargs["revision"])
617
model_group.add_argument("--code-revision", **model_kwargs["code_revision"])
620
- model_group.add_argument("--rope-scaling", **model_kwargs["rope_scaling"])
621
- model_group.add_argument("--rope-theta", **model_kwargs["rope_theta"])
618
model_group.add_argument(
619
"--tokenizer-revision", **model_kwargs["tokenizer_revision"]
620
)
@@ -1184,8 +1180,6 @@ class EngineArgs:
1180
seed=self.seed,
1181
revision=self.revision,
1182
code_revision=self.code_revision,
1187
- rope_scaling=self.rope_scaling,
1188
- rope_theta=self.rope_theta,
1183
hf_token=self.hf_token,
1184
hf_overrides=self.hf_overrides,
1185
tokenizer_revision=self.tokenizer_revision,