Skip to content

[MM] Add text-only mode for Qwen3-VL #26000

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 4 commits into from
Oct 1, 2025
+45 −26
Merged

[MM] Add text-only mode for Qwen3-VL #26000

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
d5540cd
update
ywang96 Oct 1, 2025
4530461
update
ywang96 Oct 1, 2025
e0d4d89
Update vllm/model_executor/models/qwen3_vl.py
ywang96 Oct 1, 2025
ad04a4e
add to moe
ywang96 Oct 1, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
39 changes: 25 additions & 14 deletions vllm/model_executor/models/qwen3_vl.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1125,14 +1125,17 @@ def __init__(self, *, vllm_config: VllmConfig, prefix: str = "model"):
self.config = config
self.multimodal_config = multimodal_config
self.use_data_parallel = multimodal_config.mm_encoder_tp_mode == "data"

self.visual = Qwen3_VisionTransformer(
config.vision_config,
norm_eps=getattr(config, "rms_norm_eps", 1e-6),
quant_config=quant_config,
prefix=maybe_prefix(prefix, "visual"),
use_data_parallel=self.use_data_parallel,
)
if not multimodal_config.get_limit_per_prompt("image") and \
not multimodal_config.get_limit_per_prompt("video"):
self.visual = None
else:
self.visual = Qwen3_VisionTransformer(
config.vision_config,
norm_eps=getattr(config, "rms_norm_eps", 1e-6),
quant_config=quant_config,
prefix=maybe_prefix(prefix, "visual"),
use_data_parallel=self.use_data_parallel,
)

self.language_model = Qwen3LLMForCausalLM(vllm_config=vllm_config,
prefix=maybe_prefix(
Expand All @@ -1148,11 +1151,15 @@ def __init__(self, *, vllm_config: VllmConfig, prefix: str = "model"):
config.vision_config.deepstack_visual_indexes
) if self.use_deepstack else 0
# register buffer for deepstack
self.deepstack_input_embeds = [
torch.zeros(vllm_config.scheduler_config.max_num_batched_tokens,
config.text_config.hidden_size)
for _ in range(self.deepstack_num_level)
] if self.use_deepstack else None
if self.use_deepstack and self.visual is not None:
self.deepstack_input_embeds = [
torch.zeros(
vllm_config.scheduler_config.max_num_batched_tokens,
config.text_config.hidden_size)
for _ in range(self.deepstack_num_level)
]
else:
self.deepstack_input_embeds = None
self.visual_dim = config.vision_config.out_hidden_size
self.multiscale_dim = self.visual_dim * self.deepstack_num_level

Expand Down Expand Up @@ -1526,7 +1533,11 @@ def compute_logits(

def load_weights(self, weights: Iterable[tuple[str,
torch.Tensor]]) -> set[str]:
loader = AutoWeightsLoader(self)

skip_prefixes = []
if self.visual is None:
skip_prefixes.extend(["visual."])
loader = AutoWeightsLoader(self, skip_prefixes=skip_prefixes)
return loader.load_weights(weights, mapper=self.hf_to_vllm_mapper)

def get_mm_mapping(self) -> MultiModelKeys:
Expand Down
32 changes: 20 additions & 12 deletions vllm/model_executor/models/qwen3_vl_moe.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -319,13 +319,17 @@ def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
self.multimodal_config = multimodal_config
self.use_data_parallel = multimodal_config.mm_encoder_tp_mode == "data"

self.visual = Qwen3_VisionTransformer(
config.vision_config,
norm_eps=getattr(config, "rms_norm_eps", 1e-6),
quant_config=quant_config,
prefix=maybe_prefix(prefix, "visual"),
use_data_parallel=self.use_data_parallel,
)
if not multimodal_config.get_limit_per_prompt("image") and \
not multimodal_config.get_limit_per_prompt("video"):
self.visual = None
else:
self.visual = Qwen3_VisionTransformer(
config.vision_config,
norm_eps=getattr(config, "rms_norm_eps", 1e-6),
quant_config=quant_config,
prefix=maybe_prefix(prefix, "visual"),
use_data_parallel=self.use_data_parallel,
)

self.language_model = Qwen3MoeLLMForCausalLM(vllm_config=vllm_config,
prefix=maybe_prefix(
Expand All @@ -341,10 +345,14 @@ def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
config.vision_config.deepstack_visual_indexes
) if self.use_deepstack else 0
# register buffer for deepstack
self.deepstack_input_embeds = [
torch.zeros(vllm_config.scheduler_config.max_num_batched_tokens,
config.text_config.hidden_size)
for _ in range(self.deepstack_num_level)
] if self.use_deepstack else None
if self.use_deepstack and self.visual is not None:
self.deepstack_input_embeds = [
torch.zeros(
vllm_config.scheduler_config.max_num_batched_tokens,
config.text_config.hidden_size)
for _ in range(self.deepstack_num_level)
]
else:
self.deepstack_input_embeds = None
self.visual_dim = config.vision_config.out_hidden_size
self.multiscale_dim = self.visual_dim * self.deepstack_num_level