Failed to run on 8xB200 - AssertionError: Tried to load weights of size torch.Size([6144, 30720])to a parameter of size torch.Size([6144, 18432])

#1
by shakhizat - opened

Hello,

Below is the issue encountered when running on 8×B200 GPUs.

cc @mgoin

Error logs:

Loading safetensors checkpoint shards: 89% Completed | 42/47 [01:15<00:07, 1.50s/it]
(Worker_TP2 pid=559535) INFO 07-25 10:10:23 [weight_utils.py:574] No model.safetensors.index.json found in remote.
(Worker_TP2 pid=559535) INFO 07-25 10:10:23 [weight_utils.py:811] Prefetching checkpoint files into page cache started (in background, num_threads=8, block_size=16777216 bytes)
(Worker_TP2 pid=559535) INFO 07-25 10:10:23 [weight_utils.py:806] Prefetching checkpoint files into page cache finished in 0.00s
(Worker_TP5 pid=559538) INFO 07-25 10:10:24 [weight_utils.py:574] No model.safetensors.index.json found in remote.
(Worker_TP5 pid=559538) INFO 07-25 10:10:24 [weight_utils.py:811] Prefetching checkpoint files into page cache started (in background, num_threads=8, block_size=16777216 bytes)
(Worker_TP5 pid=559538) INFO 07-25 10:10:24 [weight_utils.py:806] Prefetching checkpoint files into page cache finished in 0.00s
Loading safetensors checkpoint shards: 91% Completed | 43/47 [01:16<00:06, 1.51s/it]
(Worker_TP6 pid=559539) INFO 07-25 10:10:24 [weight_utils.py:574] No model.safetensors.index.json found in remote.
(Worker_TP6 pid=559539) INFO 07-25 10:10:24 [weight_utils.py:811] Prefetching checkpoint files into page cache started (in background, num_threads=8, block_size=16777216 bytes)
(Worker_TP6 pid=559539) INFO 07-25 10:10:24 [weight_utils.py:806] Prefetching checkpoint files into page cache finished in 0.00s
(Worker_TP2 pid=559535) ERROR 07-25 10:10:24 [multiproc_executor.py:898] WorkerProc failed to start.
(Worker_TP2 pid=559535) ERROR 07-25 10:10:24 [multiproc_executor.py:898] Traceback (most recent call last):
(Worker_TP2 pid=559535) ERROR 07-25 10:10:24 [multiproc_executor.py:898] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/v1/executor/multiproc_executor.py", line 865, in worker_main
(Worker_TP2 pid=559535) ERROR 07-25 10:10:24 [multiproc_executor.py:898] worker = WorkerProc(*args, **kwargs)
(Worker_TP2 pid=559535) ERROR 07-25 10:10:24 [multiproc_executor.py:898] ^^^^^^^^^^^^^^^^^^^^^^^^^^^
(Worker_TP2 pid=559535) ERROR 07-25 10:10:24 [multiproc_executor.py:898] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/tracing/otel.py", line 178, in sync_wrapper
(Worker_TP2 pid=559535) ERROR 07-25 10:10:24 [multiproc_executor.py:898] return func(*args, **kwargs)
(Worker_TP2 pid=559535) ERROR 07-25 10:10:24 [multiproc_executor.py:898] ^^^^^^^^^^^^^^^^^^^^^
(Worker_TP2 pid=559535) ERROR 07-25 10:10:24 [multiproc_executor.py:898] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/v1/executor/multiproc_executor.py", line 634, in __init__
(Worker_TP2 pid=559535) ERROR 07-25 10:10:24 [multiproc_executor.py:898] self.worker.load_model()
(Worker_TP2 pid=559535) ERROR 07-25 10:10:24 [multiproc_executor.py:898] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/v1/worker/gpu_worker.py", line 413, in load_model
(Worker_TP2 pid=559535) ERROR 07-25 10:10:24 [multiproc_executor.py:898] self.model_runner.load_model(load_dummy_weights=load_dummy_weights)
(Worker_TP2 pid=559535) ERROR 07-25 10:10:24 [multiproc_executor.py:898] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/v1/worker/gpu/model_runner.py", line 295, in load_model
(Worker_TP2 pid=559535) ERROR 07-25 10:10:24 [multiproc_executor.py:898] self.speculator.load_model(self.model)
(Worker_TP2 pid=559535) ERROR 07-25 10:10:24 [multiproc_executor.py:898] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/v1/worker/gpu/spec_decode/speculator.py", line 160, in load_model
(Worker_TP2 pid=559535) ERROR 07-25 10:10:24 [multiproc_executor.py:898] self.model = self.load_draft_model(target_model, target_attn_layer_names)
(Worker_TP2 pid=559535) ERROR 07-25 10:10:24 [multiproc_executor.py:898] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(Worker_TP2 pid=559535) ERROR 07-25 10:10:24 [multiproc_executor.py:898] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/v1/worker/gpu/spec_decode/dspark/speculator.py", line 82, in load_draft_model
(Worker_TP2 pid=559535) ERROR 07-25 10:10:24 [multiproc_executor.py:898] model = load_dspark_model(target_model, self.vllm_config)
(Worker_TP2 pid=559535) ERROR 07-25 10:10:24 [multiproc_executor.py:898] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(Worker_TP2 pid=559535) ERROR 07-25 10:10:24 [multiproc_executor.py:898] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/v1/worker/gpu/spec_decode/dspark/utils.py", line 31, in load_dspark_model
(Worker_TP2 pid=559535) ERROR 07-25 10:10:24 [multiproc_executor.py:898] draft_model = get_model(
(Worker_TP2 pid=559535) ERROR 07-25 10:10:24 [multiproc_executor.py:898] ^^^^^^^^^^
(Worker_TP2 pid=559535) ERROR 07-25 10:10:24 [multiproc_executor.py:898] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/model_executor/model_loader/__init__.py", line 140, in get_model
(Worker_TP2 pid=559535) ERROR 07-25 10:10:24 [multiproc_executor.py:898] return loader.load_model(
(Worker_TP2 pid=559535) ERROR 07-25 10:10:24 [multiproc_executor.py:898] ^^^^^^^^^^^^^^^^^^
(Worker_TP2 pid=559535) ERROR 07-25 10:10:24 [multiproc_executor.py:898] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/tracing/otel.py", line 178, in sync_wrapper
(Worker_TP2 pid=559535) ERROR 07-25 10:10:24 [multiproc_executor.py:898] return func(*args, **kwargs)
(Worker_TP2 pid=559535) ERROR 07-25 10:10:24 [multiproc_executor.py:898] ^^^^^^^^^^^^^^^^^^^^^
(Worker_TP2 pid=559535) ERROR 07-25 10:10:24 [multiproc_executor.py:898] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/model_executor/model_loader/base_loader.py", line 64, in load_model
(Worker_TP2 pid=559535) ERROR 07-25 10:10:24 [multiproc_executor.py:898] self.load_weights(model, model_config)
(Worker_TP2 pid=559535) ERROR 07-25 10:10:24 [multiproc_executor.py:898] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/tracing/otel.py", line 178, in sync_wrapper
(Worker_TP2 pid=559535) ERROR 07-25 10:10:24 [multiproc_executor.py:898] return func(*args, **kwargs)
(Worker_TP2 pid=559535) ERROR 07-25 10:10:24 [multiproc_executor.py:898] ^^^^^^^^^^^^^^^^^^^^^
(Worker_TP2 pid=559535) ERROR 07-25 10:10:24 [multiproc_executor.py:898] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/model_executor/model_loader/default_loader.py", line 427, in load_weights
(Worker_TP2 pid=559535) ERROR 07-25 10:10:24 [multiproc_executor.py:898] loaded_weights = model.load_weights(self.get_all_weights(model_config, model))
(Worker_TP2 pid=559535) ERROR 07-25 10:10:24 [multiproc_executor.py:898] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(Worker_TP2 pid=559535) ERROR 07-25 10:10:24 [multiproc_executor.py:898] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/model_executor/models/qwen3_dspark.py", line 184, in load_weights
(Worker_TP2 pid=559535) ERROR 07-25 10:10:24 [multiproc_executor.py:898] loader.load_weights(model_weights.items())
(Worker_TP2 pid=559535) ERROR 07-25 10:10:24 [multiproc_executor.py:898] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/model_executor/model_loader/reload/torchao_decorator.py", line 50, in patched_model_load_weights
(Worker_TP2 pid=559535) ERROR 07-25 10:10:24 [multiproc_executor.py:898] return original_load_weights(self, weights, *args, **kwargs)
(Worker_TP2 pid=559535) ERROR 07-25 10:10:24 [multiproc_executor.py:898] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(Worker_TP2 pid=559535) ERROR 07-25 10:10:24 [multiproc_executor.py:898] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/model_executor/models/utils.py", line 421, in load_weights
(Worker_TP2 pid=559535) ERROR 07-25 10:10:24 [multiproc_executor.py:898] autoloaded_weights = set(self._load_module("", self.module, weights))
(Worker_TP2 pid=559535) ERROR 07-25 10:10:24 [multiproc_executor.py:898] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(Worker_TP2 pid=559535) ERROR 07-25 10:10:24 [multiproc_executor.py:898] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/model_executor/models/utils.py", line 355, in _load_module
(Worker_TP2 pid=559535) ERROR 07-25 10:10:24 [multiproc_executor.py:898] yield from self._load_module(
(Worker_TP2 pid=559535) ERROR 07-25 10:10:24 [multiproc_executor.py:898] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/model_executor/models/utils.py", line 328, in _load_module
(Worker_TP2 pid=559535) ERROR 07-25 10:10:24 [multiproc_executor.py:898] loaded_params = module_load_weights(weights)
(Worker_TP2 pid=559535) ERROR 07-25 10:10:24 [multiproc_executor.py:898] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(Worker_TP2 pid=559535) ERROR 07-25 10:10:24 [multiproc_executor.py:898] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/model_executor/models/qwen3_dflash.py", line 657, in load_weights
(Worker_TP2 pid=559535) ERROR 07-25 10:10:24 [multiproc_executor.py:898] weight_loader(param, loaded_weight)
(Worker_TP2 pid=559535) ERROR 07-25 10:10:24 [multiproc_executor.py:898] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/model_executor/layers/linear.py", line 369, in weight_loader
(Worker_TP2 pid=559535) ERROR 07-25 10:10:24 [multiproc_executor.py:898] assert param.size() == loaded_weight.size(), (
(Worker_TP2 pid=559535) ERROR 07-25 10:10:24 [multiproc_executor.py:898] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(Worker_TP2 pid=559535) ERROR 07-25 10:10:24 [multiproc_executor.py:898] AssertionError: Tried to load weights of size torch.Size([6144, 30720])to a parameter of size torch.Size([6144, 18432])
(EngineCore pid=559457) INFO 07-25 10:10:24 [multiproc_executor.py:426] [shutdown] Executor: waiting for worker exit count=8
(Worker_TP5 pid=559538) ERROR 07-25 10:10:25 [multiproc_executor.py:898] WorkerProc failed to start.
(Worker_TP5 pid=559538) ERROR 07-25 10:10:25 [multiproc_executor.py:898] Traceback (most recent call last):
(Worker_TP5 pid=559538) ERROR 07-25 10:10:25 [multiproc_executor.py:898] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/v1/executor/multiproc_executor.py", line 865, in worker_main
(Worker_TP5 pid=559538) ERROR 07-25 10:10:25 [multiproc_executor.py:898] worker = WorkerProc(*args, **kwargs)
(Worker_TP5 pid=559538) ERROR 07-25 10:10:25 [multiproc_executor.py:898] ^^^^^^^^^^^^^^^^^^^^^^^^^^^
(Worker_TP5 pid=559538) ERROR 07-25 10:10:25 [multiproc_executor.py:898] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/tracing/otel.py", line 178, in sync_wrapper
(Worker_TP5 pid=559538) ERROR 07-25 10:10:25 [multiproc_executor.py:898] return func(*args, **kwargs)
(Worker_TP5 pid=559538) ERROR 07-25 10:10:25 [multiproc_executor.py:898] ^^^^^^^^^^^^^^^^^^^^^
(Worker_TP5 pid=559538) ERROR 07-25 10:10:25 [multiproc_executor.py:898] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/v1/executor/multiproc_executor.py", line 634, in __init__
(Worker_TP5 pid=559538) ERROR 07-25 10:10:25 [multiproc_executor.py:898] self.worker.load_model()
(Worker_TP5 pid=559538) ERROR 07-25 10:10:25 [multiproc_executor.py:898] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/v1/worker/gpu_worker.py", line 413, in load_model
(Worker_TP5 pid=559538) ERROR 07-25 10:10:25 [multiproc_executor.py:898] self.model_runner.load_model(load_dummy_weights=load_dummy_weights)
(Worker_TP5 pid=559538) ERROR 07-25 10:10:25 [multiproc_executor.py:898] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/v1/worker/gpu/model_runner.py", line 295, in load_model
(Worker_TP5 pid=559538) ERROR 07-25 10:10:25 [multiproc_executor.py:898] self.speculator.load_model(self.model)
(Worker_TP5 pid=559538) ERROR 07-25 10:10:25 [multiproc_executor.py:898] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/v1/worker/gpu/spec_decode/speculator.py", line 160, in load_model
(Worker_TP5 pid=559538) ERROR 07-25 10:10:25 [multiproc_executor.py:898] self.model = self.load_draft_model(target_model, target_attn_layer_names)
(Worker_TP5 pid=559538) ERROR 07-25 10:10:25 [multiproc_executor.py:898] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(Worker_TP5 pid=559538) ERROR 07-25 10:10:25 [multiproc_executor.py:898] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/v1/worker/gpu/spec_decode/dspark/speculator.py", line 82, in load_draft_model
(Worker_TP5 pid=559538) ERROR 07-25 10:10:25 [multiproc_executor.py:898] model = load_dspark_model(target_model, self.vllm_config)
(Worker_TP5 pid=559538) ERROR 07-25 10:10:25 [multiproc_executor.py:898] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(Worker_TP5 pid=559538) ERROR 07-25 10:10:25 [multiproc_executor.py:898] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/v1/worker/gpu/spec_decode/dspark/utils.py", line 31, in load_dspark_model
(Worker_TP5 pid=559538) ERROR 07-25 10:10:25 [multiproc_executor.py:898] draft_model = get_model(
(Worker_TP5 pid=559538) ERROR 07-25 10:10:25 [multiproc_executor.py:898] ^^^^^^^^^^
(Worker_TP5 pid=559538) ERROR 07-25 10:10:25 [multiproc_executor.py:898] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/model_executor/model_loader/__init__.py", line 140, in get_model
(Worker_TP5 pid=559538) ERROR 07-25 10:10:25 [multiproc_executor.py:898] return loader.load_model(
(Worker_TP5 pid=559538) ERROR 07-25 10:10:25 [multiproc_executor.py:898] ^^^^^^^^^^^^^^^^^^
(Worker_TP5 pid=559538) ERROR 07-25 10:10:25 [multiproc_executor.py:898] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/tracing/otel.py", line 178, in sync_wrapper
(Worker_TP5 pid=559538) ERROR 07-25 10:10:25 [multiproc_executor.py:898] return func(*args, **kwargs)
(Worker_TP5 pid=559538) ERROR 07-25 10:10:25 [multiproc_executor.py:898] ^^^^^^^^^^^^^^^^^^^^^
(Worker_TP5 pid=559538) ERROR 07-25 10:10:25 [multiproc_executor.py:898] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/model_executor/model_loader/base_loader.py", line 64, in load_model
(Worker_TP5 pid=559538) ERROR 07-25 10:10:25 [multiproc_executor.py:898] self.load_weights(model, model_config)
(Worker_TP5 pid=559538) ERROR 07-25 10:10:25 [multiproc_executor.py:898] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/tracing/otel.py", line 178, in sync_wrapper
(Worker_TP5 pid=559538) ERROR 07-25 10:10:25 [multiproc_executor.py:898] return func(*args, **kwargs)
(Worker_TP5 pid=559538) ERROR 07-25 10:10:25 [multiproc_executor.py:898] ^^^^^^^^^^^^^^^^^^^^^
(Worker_TP5 pid=559538) ERROR 07-25 10:10:25 [multiproc_executor.py:898] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/model_executor/model_loader/default_loader.py", line 427, in load_weights
(Worker_TP5 pid=559538) ERROR 07-25 10:10:25 [multiproc_executor.py:898] loaded_weights = model.load_weights(self.get_all_weights(model_config, model))
(Worker_TP5 pid=559538) ERROR 07-25 10:10:25 [multiproc_executor.py:898] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(Worker_TP5 pid=559538) ERROR 07-25 10:10:25 [multiproc_executor.py:898] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/model_executor/models/qwen3_dspark.py", line 184, in load_weights
(Worker_TP5 pid=559538) ERROR 07-25 10:10:25 [multiproc_executor.py:898] loader.load_weights(model_weights.items())
(Worker_TP5 pid=559538) ERROR 07-25 10:10:25 [multiproc_executor.py:898] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/model_executor/model_loader/reload/torchao_decorator.py", line 50, in patched_model_load_weights
(Worker_TP5 pid=559538) ERROR 07-25 10:10:25 [multiproc_executor.py:898] return original_load_weights(self, weights, *args, **kwargs)
(Worker_TP5 pid=559538) ERROR 07-25 10:10:25 [multiproc_executor.py:898] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(Worker_TP5 pid=559538) ERROR 07-25 10:10:25 [multiproc_executor.py:898] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/model_executor/models/utils.py", line 421, in load_weights
(Worker_TP5 pid=559538) ERROR 07-25 10:10:25 [multiproc_executor.py:898] autoloaded_weights = set(self._load_module("", self.module, weights))
(Worker_TP5 pid=559538) ERROR 07-25 10:10:25 [multiproc_executor.py:898] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(Worker_TP5 pid=559538) ERROR 07-25 10:10:25 [multiproc_executor.py:898] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/model_executor/models/utils.py", line 355, in _load_module
(Worker_TP5 pid=559538) ERROR 07-25 10:10:25 [multiproc_executor.py:898] yield from self._load_module(
(Worker_TP5 pid=559538) ERROR 07-25 10:10:25 [multiproc_executor.py:898] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/model_executor/models/utils.py", line 328, in _load_module
(Worker_TP5 pid=559538) ERROR 07-25 10:10:25 [multiproc_executor.py:898] loaded_params = module_load_weights(weights)
(Worker_TP5 pid=559538) ERROR 07-25 10:10:25 [multiproc_executor.py:898] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(Worker_TP5 pid=559538) ERROR 07-25 10:10:25 [multiproc_executor.py:898] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/model_executor/models/qwen3_dflash.py", line 657, in load_weights
(Worker_TP5 pid=559538) ERROR 07-25 10:10:25 [multiproc_executor.py:898] weight_loader(param, loaded_weight)
(Worker_TP5 pid=559538) ERROR 07-25 10:10:25 [multiproc_executor.py:898] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/model_executor/layers/linear.py", line 369, in weight_loader
(Worker_TP5 pid=559538) ERROR 07-25 10:10:25 [multiproc_executor.py:898] assert param.size() == loaded_weight.size(), (
(Worker_TP5 pid=559538) ERROR 07-25 10:10:25 [multiproc_executor.py:898] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(Worker_TP5 pid=559538) ERROR 07-25 10:10:25 [multiproc_executor.py:898] AssertionError: Tried to load weights of size torch.Size([6144, 30720])to a parameter of size torch.Size([6144, 18432])
(Worker_TP6 pid=559539) ERROR 07-25 10:10:25 [multiproc_executor.py:898] WorkerProc failed to start.
(Worker_TP6 pid=559539) ERROR 07-25 10:10:25 [multiproc_executor.py:898] Traceback (most recent call last):
(Worker_TP6 pid=559539) ERROR 07-25 10:10:25 [multiproc_executor.py:898] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/v1/executor/multiproc_executor.py", line 865, in worker_main
(Worker_TP6 pid=559539) ERROR 07-25 10:10:25 [multiproc_executor.py:898] worker = WorkerProc(*args, **kwargs)
(Worker_TP6 pid=559539) ERROR 07-25 10:10:25 [multiproc_executor.py:898] ^^^^^^^^^^^^^^^^^^^^^^^^^^^
(Worker_TP6 pid=559539) ERROR 07-25 10:10:25 [multiproc_executor.py:898] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/tracing/otel.py", line 178, in sync_wrapper
(Worker_TP6 pid=559539) ERROR 07-25 10:10:25 [multiproc_executor.py:898] return func(*args, **kwargs)
(Worker_TP6 pid=559539) ERROR 07-25 10:10:25 [multiproc_executor.py:898] ^^^^^^^^^^^^^^^^^^^^^
(Worker_TP6 pid=559539) ERROR 07-25 10:10:25 [multiproc_executor.py:898] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/v1/executor/multiproc_executor.py", line 634, in __init__
(Worker_TP6 pid=559539) ERROR 07-25 10:10:25 [multiproc_executor.py:898] self.worker.load_model()
(Worker_TP6 pid=559539) ERROR 07-25 10:10:25 [multiproc_executor.py:898] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/v1/worker/gpu_worker.py", line 413, in load_model
(Worker_TP6 pid=559539) ERROR 07-25 10:10:25 [multiproc_executor.py:898] self.model_runner.load_model(load_dummy_weights=load_dummy_weights)
(Worker_TP6 pid=559539) ERROR 07-25 10:10:25 [multiproc_executor.py:898] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/v1/worker/gpu/model_runner.py", line 295, in load_model
(Worker_TP6 pid=559539) ERROR 07-25 10:10:25 [multiproc_executor.py:898] self.speculator.load_model(self.model)
(Worker_TP6 pid=559539) ERROR 07-25 10:10:25 [multiproc_executor.py:898] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/v1/worker/gpu/spec_decode/speculator.py", line 160, in load_model
(Worker_TP6 pid=559539) ERROR 07-25 10:10:25 [multiproc_executor.py:898] self.model = self.load_draft_model(target_model, target_attn_layer_names)
(Worker_TP6 pid=559539) ERROR 07-25 10:10:25 [multiproc_executor.py:898] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(Worker_TP6 pid=559539) ERROR 07-25 10:10:25 [multiproc_executor.py:898] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/v1/worker/gpu/spec_decode/dspark/speculator.py", line 82, in load_draft_model
(Worker_TP6 pid=559539) ERROR 07-25 10:10:25 [multiproc_executor.py:898] model = load_dspark_model(target_model, self.vllm_config)
(Worker_TP6 pid=559539) ERROR 07-25 10:10:25 [multiproc_executor.py:898] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(Worker_TP6 pid=559539) ERROR 07-25 10:10:25 [multiproc_executor.py:898] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/v1/worker/gpu/spec_decode/dspark/utils.py", line 31, in load_dspark_model
(Worker_TP6 pid=559539) ERROR 07-25 10:10:25 [multiproc_executor.py:898] draft_model = get_model(
(Worker_TP6 pid=559539) ERROR 07-25 10:10:25 [multiproc_executor.py:898] ^^^^^^^^^^
(Worker_TP6 pid=559539) ERROR 07-25 10:10:25 [multiproc_executor.py:898] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/model_executor/model_loader/__init__.py", line 140, in get_model
(Worker_TP6 pid=559539) ERROR 07-25 10:10:25 [multiproc_executor.py:898] return loader.load_model(
(Worker_TP6 pid=559539) ERROR 07-25 10:10:25 [multiproc_executor.py:898] ^^^^^^^^^^^^^^^^^^
(Worker_TP6 pid=559539) ERROR 07-25 10:10:25 [multiproc_executor.py:898] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/tracing/otel.py", line 178, in sync_wrapper
(Worker_TP6 pid=559539) ERROR 07-25 10:10:25 [multiproc_executor.py:898] return func(*args, **kwargs)
(Worker_TP6 pid=559539) ERROR 07-25 10:10:25 [multiproc_executor.py:898] ^^^^^^^^^^^^^^^^^^^^^
(Worker_TP6 pid=559539) ERROR 07-25 10:10:25 [multiproc_executor.py:898] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/model_executor/model_loader/base_loader.py", line 64, in load_model
(Worker_TP6 pid=559539) ERROR 07-25 10:10:25 [multiproc_executor.py:898] self.load_weights(model, model_config)
(Worker_TP6 pid=559539) ERROR 07-25 10:10:25 [multiproc_executor.py:898] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/tracing/otel.py", line 178, in sync_wrapper
(Worker_TP6 pid=559539) ERROR 07-25 10:10:25 [multiproc_executor.py:898] return func(*args, **kwargs)
(Worker_TP6 pid=559539) ERROR 07-25 10:10:25 [multiproc_executor.py:898] ^^^^^^^^^^^^^^^^^^^^^
(Worker_TP6 pid=559539) ERROR 07-25 10:10:25 [multiproc_executor.py:898] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/model_executor/model_loader/default_loader.py", line 427, in load_weights
(Worker_TP6 pid=559539) ERROR 07-25 10:10:25 [multiproc_executor.py:898] loaded_weights = model.load_weights(self.get_all_weights(model_config, model))
(Worker_TP6 pid=559539) ERROR 07-25 10:10:25 [multiproc_executor.py:898] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(Worker_TP6 pid=559539) ERROR 07-25 10:10:25 [multiproc_executor.py:898] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/model_executor/models/qwen3_dspark.py", line 184, in load_weights
(Worker_TP6 pid=559539) ERROR 07-25 10:10:25 [multiproc_executor.py:898] loader.load_weights(model_weights.items())
(Worker_TP6 pid=559539) ERROR 07-25 10:10:25 [multiproc_executor.py:898] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/model_executor/model_loader/reload/torchao_decorator.py", line 50, in patched_model_load_weights
(Worker_TP6 pid=559539) ERROR 07-25 10:10:25 [multiproc_executor.py:898] return original_load_weights(self, weights, *args, **kwargs)
(Worker_TP6 pid=559539) ERROR 07-25 10:10:25 [multiproc_executor.py:898] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(Worker_TP6 pid=559539) ERROR 07-25 10:10:25 [multiproc_executor.py:898] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/model_executor/models/utils.py", line 421, in load_weights
(Worker_TP6 pid=559539) ERROR 07-25 10:10:25 [multiproc_executor.py:898] autoloaded_weights = set(self._load_module("", self.module, weights))
(Worker_TP6 pid=559539) ERROR 07-25 10:10:25 [multiproc_executor.py:898] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(Worker_TP6 pid=559539) ERROR 07-25 10:10:25 [multiproc_executor.py:898] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/model_executor/models/utils.py", line 355, in _load_module
(Worker_TP6 pid=559539) ERROR 07-25 10:10:25 [multiproc_executor.py:898] yield from self._load_module(
(Worker_TP6 pid=559539) ERROR 07-25 10:10:25 [multiproc_executor.py:898] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/model_executor/models/utils.py", line 328, in _load_module
(Worker_TP6 pid=559539) ERROR 07-25 10:10:25 [multiproc_executor.py:898] loaded_params = module_load_weights(weights)
(Worker_TP6 pid=559539) ERROR 07-25 10:10:25 [multiproc_executor.py:898] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(Worker_TP6 pid=559539) ERROR 07-25 10:10:25 [multiproc_executor.py:898] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/model_executor/models/qwen3_dflash.py", line 657, in load_weights
(Worker_TP6 pid=559539) ERROR 07-25 10:10:25 [multiproc_executor.py:898] weight_loader(param, loaded_weight)
(Worker_TP6 pid=559539) ERROR 07-25 10:10:25 [multiproc_executor.py:898] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/model_executor/layers/linear.py", line 369, in weight_loader
(Worker_TP6 pid=559539) ERROR 07-25 10:10:25 [multiproc_executor.py:898] assert param.size() == loaded_weight.size(), (
(Worker_TP6 pid=559539) ERROR 07-25 10:10:25 [multiproc_executor.py:898] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(Worker_TP6 pid=559539) ERROR 07-25 10:10:25 [multiproc_executor.py:898] AssertionError: Tried to load weights of size torch.Size([6144, 30720])to a parameter of size torch.Size([6144, 18432])
Loading safetensors checkpoint shards: 94% Completed | 44/47 [01:18<00:04, 1.53s/it]
Loading safetensors checkpoint shards: 96% Completed | 45/47 [01:18<00:02, 1.26s/it]
Loading safetensors checkpoint shards: 100% Completed | 47/47 [01:18<00:00, 1.68s/it]
(Worker_TP0 pid=559533)
(Worker_TP0 pid=559533) INFO 07-25 10:10:26 [default_loader.py:430] Loading weights took 78.91 seconds
(Worker_TP0 pid=559533) WARNING 07-25 10:10:26 [kv_cache.py:134] Checkpoint does not provide a q scaling factor. Setting it to k_scale. This only matters for FP8 Attention backends (flash-attn or flashinfer).
(Worker_TP0 pid=559533) WARNING 07-25 10:10:26 [kv_cache.py:148] Using KV cache scaling factor 1.0 for fp8_e4m3. If this is unintended, verify that k/v_scale scaling factors are properly set in the checkpoint.
(Worker_TP0 pid=559533) INFO 07-25 10:10:26 [nvfp4.py:543] Using MoEPrepareAndFinalizeNoDPEPMonolithic
(EngineCore pid=559457) WARNING 07-25 10:10:29 [multiproc_executor.py:438] [shutdown] Executor: workers still running after grace period; sending SIGTERM count=5
(Worker_TP3 pid=559536) ERROR 07-25 10:10:30 [multiproc_executor.py:898] WorkerProc failed to start.
(Worker_TP3 pid=559536) ERROR 07-25 10:10:30 [multiproc_executor.py:898] Traceback (most recent call last):
(Worker_TP3 pid=559536) ERROR 07-25 10:10:30 [multiproc_executor.py:898] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/v1/executor/multiproc_executor.py", line 865, in worker_main
(Worker_TP3 pid=559536) ERROR 07-25 10:10:30 [multiproc_executor.py:898] worker = WorkerProc(*args, **kwargs)
(Worker_TP3 pid=559536) ERROR 07-25 10:10:30 [multiproc_executor.py:898] ^^^^^^^^^^^^^^^^^^^^^^^^^^^
(Worker_TP3 pid=559536) ERROR 07-25 10:10:30 [multiproc_executor.py:898] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/tracing/otel.py", line 178, in sync_wrapper
(Worker_TP3 pid=559536) ERROR 07-25 10:10:30 [multiproc_executor.py:898] return func(*args, **kwargs)
(Worker_TP3 pid=559536) ERROR 07-25 10:10:30 [multiproc_executor.py:898] ^^^^^^^^^^^^^^^^^^^^^
(Worker_TP3 pid=559536) ERROR 07-25 10:10:30 [multiproc_executor.py:898] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/v1/executor/multiproc_executor.py", line 634, in __init__
(Worker_TP3 pid=559536) ERROR 07-25 10:10:30 [multiproc_executor.py:898] self.worker.load_model()
(Worker_TP3 pid=559536) ERROR 07-25 10:10:30 [multiproc_executor.py:898] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/v1/worker/gpu_worker.py", line 413, in load_model
(Worker_TP3 pid=559536) ERROR 07-25 10:10:30 [multiproc_executor.py:898] self.model_runner.load_model(load_dummy_weights=load_dummy_weights)
(Worker_TP3 pid=559536) ERROR 07-25 10:10:30 [multiproc_executor.py:898] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/v1/worker/gpu/model_runner.py", line 283, in load_model
(Worker_TP3 pid=559536) ERROR 07-25 10:10:30 [multiproc_executor.py:898] self.model = model_loader.load_model(
(Worker_TP3 pid=559536) ERROR 07-25 10:10:30 [multiproc_executor.py:898] ^^^^^^^^^^^^^^^^^^^^^^^^
(Worker_TP3 pid=559536) ERROR 07-25 10:10:30 [multiproc_executor.py:898] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/tracing/otel.py", line 178, in sync_wrapper
(Worker_TP3 pid=559536) ERROR 07-25 10:10:30 [multiproc_executor.py:898] return func(*args, **kwargs)
(Worker_TP3 pid=559536) ERROR 07-25 10:10:30 [multiproc_executor.py:898] ^^^^^^^^^^^^^^^^^^^^^
(Worker_TP3 pid=559536) ERROR 07-25 10:10:30 [multiproc_executor.py:898] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/model_executor/model_loader/base_loader.py", line 64, in load_model
(Worker_TP3 pid=559536) ERROR 07-25 10:10:30 [multiproc_executor.py:898] self.load_weights(model, model_config)
(Worker_TP3 pid=559536) ERROR 07-25 10:10:30 [multiproc_executor.py:898] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/tracing/otel.py", line 178, in sync_wrapper
(Worker_TP3 pid=559536) ERROR 07-25 10:10:30 [multiproc_executor.py:898] return func(*args, **kwargs)
(Worker_TP3 pid=559536) ERROR 07-25 10:10:30 [multiproc_executor.py:898] ^^^^^^^^^^^^^^^^^^^^^
(Worker_TP3 pid=559536) ERROR 07-25 10:10:30 [multiproc_executor.py:898] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/model_executor/model_loader/default_loader.py", line 427, in load_weights
(Worker_TP3 pid=559536) ERROR 07-25 10:10:30 [multiproc_executor.py:898] loaded_weights = model.load_weights(self.get_all_weights(model_config, model))
(Worker_TP3 pid=559536) ERROR 07-25 10:10:30 [multiproc_executor.py:898] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(Worker_TP3 pid=559536) ERROR 07-25 10:10:30 [multiproc_executor.py:898] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/model_executor/models/deepseek_v2.py", line 1903, in load_weights
(Worker_TP3 pid=559536) ERROR 07-25 10:10:30 [multiproc_executor.py:898] return loader.load_weights(weights)
(Worker_TP3 pid=559536) ERROR 07-25 10:10:30 [multiproc_executor.py:898] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(Worker_TP3 pid=559536) ERROR 07-25 10:10:30 [multiproc_executor.py:898] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/model_executor/model_loader/reload/torchao_decorator.py", line 50, in patched_model_load_weights
(Worker_TP3 pid=559536) ERROR 07-25 10:10:30 [multiproc_executor.py:898] return original_load_weights(self, weights, *args, **kwargs)
(Worker_TP3 pid=559536) ERROR 07-25 10:10:30 [multiproc_executor.py:898] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(Worker_TP3 pid=559536) ERROR 07-25 10:10:30 [multiproc_executor.py:898] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/model_executor/models/utils.py", line 421, in load_weights
(Worker_TP3 pid=559536) ERROR 07-25 10:10:30 [multiproc_executor.py:898] autoloaded_weights = set(self._load_module("", self.module, weights))
(Worker_TP3 pid=559536) ERROR 07-25 10:10:30 [multiproc_executor.py:898] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(Worker_TP3 pid=559536) ERROR 07-25 10:10:30 [multiproc_executor.py:898] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/model_executor/models/utils.py", line 355, in _load_module
(Worker_TP3 pid=559536) ERROR 07-25 10:10:30 [multiproc_executor.py:898] yield from self._load_module(
(Worker_TP3 pid=559536) ERROR 07-25 10:10:30 [multiproc_executor.py:898] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/model_executor/models/utils.py", line 328, in _load_module
(Worker_TP3 pid=559536) ERROR 07-25 10:10:30 [multiproc_executor.py:898] loaded_params = module_load_weights(weights)
(Worker_TP3 pid=559536) ERROR 07-25 10:10:30 [multiproc_executor.py:898] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(Worker_TP3 pid=559536) ERROR 07-25 10:10:30 [multiproc_executor.py:898] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/model_executor/models/deepseek_v2.py", line 1555, in load_weights
(Worker_TP3 pid=559536) ERROR 07-25 10:10:30 [multiproc_executor.py:898] for name, loaded_weight in weights:
(Worker_TP3 pid=559536) ERROR 07-25 10:10:30 [multiproc_executor.py:898] ^^^^^^^
(Worker_TP3 pid=559536) ERROR 07-25 10:10:30 [multiproc_executor.py:898] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/model_executor/models/utils.py", line 230, in
(Worker_TP3 pid=559536) ERROR 07-25 10:10:30 [multiproc_executor.py:898] for parts, weights_data in group
(Worker_TP3 pid=559536) ERROR 07-25 10:10:30 [multiproc_executor.py:898] ^^^^^
(Worker_TP3 pid=559536) ERROR 07-25 10:10:30 [multiproc_executor.py:898] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/model_executor/models/utils.py", line 220, in
(Worker_TP3 pid=559536) ERROR 07-25 10:10:30 [multiproc_executor.py:898] for weight_name, weight_data in weights
(Worker_TP3 pid=559536) ERROR 07-25 10:10:30 [multiproc_executor.py:898] ^^^^^^^
(Worker_TP3 pid=559536) ERROR 07-25 10:10:30 [multiproc_executor.py:898] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/model_executor/models/utils.py", line 418, in
(Worker_TP3 pid=559536) ERROR 07-25 10:10:30 [multiproc_executor.py:898] (name, weight) for name, weight in weights if not self._can_skip(name)
(Worker_TP3 pid=559536) ERROR 07-25 10:10:30 [multiproc_executor.py:898] ^^^^^^^
(Worker_TP3 pid=559536) ERROR 07-25 10:10:30 [multiproc_executor.py:898] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/model_executor/models/utils.py", line 137, in apply
(Worker_TP3 pid=559536) ERROR 07-25 10:10:30 [multiproc_executor.py:898] for name, data in weights:
(Worker_TP3 pid=559536) ERROR 07-25 10:10:30 [multiproc_executor.py:898] ^^^^^^^
(Worker_TP3 pid=559536) ERROR 07-25 10:10:30 [multiproc_executor.py:898] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/model_executor/model_loader/default_loader.py", line 333, in get_all_weights
(Worker_TP3 pid=559536) ERROR 07-25 10:10:30 [multiproc_executor.py:898] yield from self._get_weights_iterator(primary_weights)
(Worker_TP3 pid=559536) ERROR 07-25 10:10:30 [multiproc_executor.py:898] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/model_executor/model_loader/default_loader.py", line 319, in
(Worker_TP3 pid=559536) ERROR 07-25 10:10:30 [multiproc_executor.py:898] return ((source.prefix + name, tensor) for (name, tensor) in weights_iterator)
(Worker_TP3 pid=559536) ERROR 07-25 10:10:30 [multiproc_executor.py:898] ^^^^^^^^^^^^^^^^
(Worker_TP3 pid=559536) ERROR 07-25 10:10:30 [multiproc_executor.py:898] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/model_executor/model_loader/weight_utils.py", line 953, in safetensors_weights_iterator
(Worker_TP3 pid=559536) ERROR 07-25 10:10:30 [multiproc_executor.py:898] param = f.get_tensor(name)
(Worker_TP3 pid=559536) ERROR 07-25 10:10:30 [multiproc_executor.py:898] ^^^^^^^^^^^^^^^^^^
(Worker_TP3 pid=559536) ERROR 07-25 10:10:30 [multiproc_executor.py:898] ValueError: could not determine the shape of object type 'torch.storage.UntypedStorage'
[rank0]:[W725 10:10:31.309884130 ProcessGroupNCCL.cpp:1575] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
(EngineCore pid=559457) WARNING 07-25 10:10:33 [multiproc_executor.py:448] [shutdown] Executor: workers still running after SIGTERM; sending SIGKILL count=0
(EngineCore pid=559457) ERROR 07-25 10:10:33 [core.py:1231] EngineCore failed to start.
(EngineCore pid=559457) ERROR 07-25 10:10:33 [core.py:1231] Traceback (most recent call last):
(EngineCore pid=559457) ERROR 07-25 10:10:33 [core.py:1231] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/v1/engine/core.py", line 1200, in run_engine_core
(EngineCore pid=559457) ERROR 07-25 10:10:33 [core.py:1231] engine_core = EngineCoreProc(*args, engine_index=dp_rank, **kwargs)
(EngineCore pid=559457) ERROR 07-25 10:10:33 [core.py:1231] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(EngineCore pid=559457) ERROR 07-25 10:10:33 [core.py:1231] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/tracing/otel.py", line 178, in sync_wrapper
(EngineCore pid=559457) ERROR 07-25 10:10:33 [core.py:1231] return func(*args, **kwargs)
(EngineCore pid=559457) ERROR 07-25 10:10:33 [core.py:1231] ^^^^^^^^^^^^^^^^^^^^^
(EngineCore pid=559457) ERROR 07-25 10:10:33 [core.py:1231] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/v1/engine/core.py", line 966, in init
(EngineCore pid=559457) ERROR 07-25 10:10:33 [core.py:1231] super().init(
(EngineCore pid=559457) ERROR 07-25 10:10:33 [core.py:1231] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/v1/engine/core.py", line 123, in init
(EngineCore pid=559457) ERROR 07-25 10:10:33 [core.py:1231] self.model_executor = executor_class(vllm_config)
(EngineCore pid=559457) ERROR 07-25 10:10:33 [core.py:1231] ^^^^^^^^^^^^^^^^^^^^^^^^^^^
(EngineCore pid=559457) ERROR 07-25 10:10:33 [core.py:1231] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/v1/executor/multiproc_executor.py", line 108, in init
(EngineCore pid=559457) ERROR 07-25 10:10:33 [core.py:1231] super().init(vllm_config)
(EngineCore pid=559457) ERROR 07-25 10:10:33 [core.py:1231] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/tracing/otel.py", line 178, in sync_wrapper
(EngineCore pid=559457) ERROR 07-25 10:10:33 [core.py:1231] return func(*args, **kwargs)
(EngineCore pid=559457) ERROR 07-25 10:10:33 [core.py:1231] ^^^^^^^^^^^^^^^^^^^^^
(EngineCore pid=559457) ERROR 07-25 10:10:33 [core.py:1231] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/v1/executor/abstract.py", line 109, in init
(EngineCore pid=559457) ERROR 07-25 10:10:33 [core.py:1231] self._init_executor()
(EngineCore pid=559457) ERROR 07-25 10:10:33 [core.py:1231] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/v1/executor/multiproc_executor.py", line 201, in _init_executor
(EngineCore pid=559457) ERROR 07-25 10:10:33 [core.py:1231] self.workers = WorkerProc.wait_for_ready(unready_workers)
(EngineCore pid=559457) ERROR 07-25 10:10:33 [core.py:1231] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(EngineCore pid=559457) ERROR 07-25 10:10:33 [core.py:1231] File "/home/shakhizat/.vllm/lib64/python3.12/site-packages/vllm/v1/executor/multiproc_executor.py", line 762, in wait_for_ready
(EngineCore pid=559457) ERROR 07-25 10:10:33 [core.py:1231] raise e from None
(EngineCore pid=559457) ERROR 07-25 10:10:33 [core.py:1231] Exception: WorkerProc initialization failed due to an exception in a background process. See stack trace for root cause.
(EngineCore pid=559457) Process EngineCore:

Hi @shakhizat - I traced this; it's a draft-model config translation bug in vLLM, not your hardware or the checkpoint.

Root cause: the speculator was trained with 5 auxiliary hidden-state layers (aux_hidden_state_layer_ids: [2, 20, 39, 58, 75] in config.json), so its fc weight is [6144, 30720] (5 x 6144). At load time vLLM fails to pick up those layer ids and falls back to sizing fc from the draft's num_hidden_layers: 3 - [6144, 18432]. Hence the exact assertion you hit.

Specifically, vLLM's speculators-format translator update_dspark() (vllm/transformers_utils/configs/speculators/algos.py) doesn't populate dflash_config.target_layer_ids, while update_dflash() does - which is why the RedHatAI .dflash checkpoints with the same 5-aux/3-draft-layer layout load fine and this .dspark one doesn't. This is the bug fixed in open PR vllm-project/vllm#48932 ("Fix speculators-format DSpark loading"), which reports the same weight-size assertion.

Workaround until the PR lands - add this to the draft model's config.json (local copy), keeping DSpark's i-1 layer-id semantics (vLLM adds +1 back internally):

"dflash_config": {
  "mask_token_id": 154856,
  "target_layer_ids": [1, 19, 38, 57, 74]
}

or patch your vLLM install with the algos.py hunk from the PR (a ~9-line change to update_dspark).

Two side notes on the log: the ValueError: could not determine the shape of object type 'torch.storage.UntypedStorage' on Worker_TP3 happens after the executor begins shutdown - it's teardown noise, not a second issue. And No model.safetensors.index.json found in remote is expected for this repo layout.

Would be great to get #48932 reviewed/merged - happy to help validate the fix on multi-GPU Blackwell once it's in a nightly. cc @mgoin

Red Hat AI org

@shakhizat @rigoretto Could you pull the latest main from vllm? This issue should have been resolved. Thanks! Let me know if bug still exists.

hi @shanjiaz , failed to run using v0.26.0, and haven't managed to build from the source, will try it later on.

Red Hat AI org

Yeah the new version did not include the fix https://github.com/vllm-project/vllm/pull/48524

Yeah the new version did not include the fix https://github.com/vllm-project/vllm/pull/48524


Hi @shanjiaz ,

I cherry-picked PR #48524 into my build and confirmed the dspark weight-loading issue is fixed (thanks!). However, the engine still fails to start with the same setup:

NotImplementedError: Layer model.layers.0.self_attn.indexer.k_cache: page size is not divisible by the maximum page size and cannot be padded.

Start command:

docker run -d \
  --runtime /usr/bin/nvidia-container-runtime \
  --name glm-5.2 \
  --gpus all \
  --ipc=host \
  -p 33000:33000 \
  --ulimit memlock=-1 \
  --ulimit stack=67108864 \
  --restart always \
  -v /models:/models \
  -e VLLM_ENGINE_READY_TIMEOUT_S=3600 \
  vllm-openai:glm52-dspark-pr48524 \
  --model /models/GLM-5.2-FP8 \
  --host 0.0.0.0 \
  --port 33000 \
  --served-model-name glm-5.2 \
  --trust-remote-code \
  --enable-chunked-prefill \
  --enable-prefix-caching \
  --max-model-len auto \
  --max-num-seqs 20 \
  --max-num-batched-tokens 8192 \
  --tensor-parallel-size 8 \
  --tool-call-parser glm47 \
  --enable-auto-tool-choice \
  --reasoning-parser glm45 \
  --safetensors-load-strategy=prefetch \
  --default-chat-template-kwargs '{"thinking": true, "reasoning_effort": "high"}' \
  --speculative-config '{"model": "/models/GLM-5.2-speculator.dspark","num_speculative_tokens": 7,"method": "dspark"}'

What should I do next to get this running? Should I open a separate issue, or is there a config flag / patch I’m missing?


Red Hat AI org

Ah thanks for trying! I think you might need the latest main or https://github.com/vllm-project/vllm/pull/49704.

Hello @shanjiaz , It now shows a CUDAOOM error. Shouldn't it fit on 8xB200 GPUs?

Here's the command to reproduce it.

vllm serve nvidia/GLM-5.2-NVFP4
--tensor-parallel-size 8
--enable-expert-parallel
--spec-model RedHatAI/GLM-5.2-speculator.dspark
--spec-method dspark
--spec-tokens 7
--trust-remote-code
--reasoning-parser glm45
--tool-call-parser glm47
--enable-auto-tool-choice
--kv-cache-dtype fp8_e4m3
--host 0.0.0.0 --port 8000

Red Hat AI org

It should fit, taking a look!

@shanjiaz , I attached an error log here: https://github.com/vllm-project/vllm/issues/49475, a vanilla command without speculator works fine.

Sign up or log in to comment