Unable to Load This Model with SGLang
I noticed that SGLang version 0.5.4 supports autoround quantization. However, when I try to load this model using the following command:
python3 -m sglang.launch_server --model-path /mnt/models/Qwen3-Next-80B-A3B-Thinking-int4-mixed-AutoRound/ --tensor-parallel 4 --port 6004 --host 0.0.0.0 --max-running 10 --served-model-name Qwen3-Next --mem-fraction-static 0.95 --expert-parallel 4
I encounter the following error:
[2025-11-06 10:35:50] WARNING model_config.py:715: auto-round quantization is not fully optimized yet. The speed can be slower than non-quantized models.
[2025-11-06 10:35:50] WARNING server_args.py:1165: Attention backend not explicitly specified. Use flashinfer backend by default.
[2025-11-06 10:35:50] INFO trace.py:52: opentelemetry package is not installed, tracing disabled
[2025-11-06 10:35:52] server_args=ServerArgs(model_path='/mnt/models/Qwen3-Next-80B-A3B-Thinking-int4-mixed-AutoRound/', tokenizer_path='/mnt/models/Qwen3-Next-80B-A3B-Thinking-int4-mixed-AutoRound/', tokenizer_mode='auto', tokenizer_worker_num=1, skip_tokenizer_init=False, load_format='auto', model_loader_extra_config='{}', trust_remote_code=False, context_length=None, is_embedding=False, enable_multimodal=None, revision=None, model_impl='auto', host='0.0.0.0', port=6004, grpc_mode=False, skip_server_warmup=False, warmups=None, nccl_port=None, checkpoint_engine_wait_weights_before_ready=False, dtype='auto', quantization=None, quantization_param_path=None, kv_cache_dtype='auto', enable_fp32_lm_head=False, modelopt_quant=None, modelopt_checkpoint_restore_path=None, modelopt_checkpoint_save_path=None, modelopt_export_path=None, quantize_and_serve=False, mem_fraction_static=0.95, max_running_requests=10, max_queued_requests=None, max_total_tokens=None, chunked_prefill_size=2048, max_prefill_tokens=16384, schedule_policy='fcfs', enable_priority_scheduling=False, abort_on_priority_when_disabled=False, schedule_low_priority_values_first=False, priority_scheduling_preemption_threshold=10, schedule_conservativeness=1.0, page_size=1, hybrid_kvcache_ratio=None, swa_full_tokens_ratio=0.8, disable_hybrid_swa_memory=False, radix_eviction_policy='lru', device='cuda', tp_size=4, pp_size=1, pp_max_micro_batch_size=None, stream_interval=1, stream_output=False, random_seed=544690698, constrained_json_whitespace_pattern=None, constrained_json_disable_any_whitespace=False, watchdog_timeout=300, dist_timeout=None, download_dir=None, base_gpu_id=0, gpu_id_step=1, sleep_on_idle=False, log_level='info', log_level_http=None, log_requests=False, log_requests_level=2, crash_dump_folder=None, show_time_cost=False, enable_metrics=False, enable_metrics_for_all_schedulers=False, tokenizer_metrics_custom_labels_header='x-custom-labels', tokenizer_metrics_allowed_custom_labels=None, bucket_time_to_first_token=None, bucket_inter_token_latency=None, bucket_e2e_request_latency=None, collect_tokens_histogram=False, prompt_tokens_buckets=None, generation_tokens_buckets=None, gc_warning_threshold_secs=0.0, decode_log_interval=40, enable_request_time_stats_logging=False, kv_events_config=None, enable_trace=False, otlp_traces_endpoint='localhost:4317', api_key=None, served_model_name='Qwen3-Next', weight_version='default', chat_template=None, completion_template=None, file_storage_path='sglang_storage', enable_cache_report=False, reasoning_parser=None, tool_call_parser=None, tool_server=None, sampling_defaults='model', dp_size=1, load_balance_method='round_robin', load_watch_interval=0.1, prefill_round_robin_balance=False, dist_init_addr=None, nnodes=1, node_rank=0, json_model_override_args='{}', preferred_sampling_params=None, enable_lora=None, max_lora_rank=None, lora_target_modules=None, lora_paths=None, max_loaded_loras=None, max_loras_per_batch=8, lora_eviction_policy='lru', lora_backend='csgmv', max_lora_chunk_size=16, attention_backend='flashinfer', decode_attention_backend=None, prefill_attention_backend=None, sampling_backend='flashinfer', grammar_backend='xgrammar', mm_attention_backend=None, nsa_prefill_backend='flashmla_sparse', nsa_decode_backend='fa3', speculative_algorithm=None, speculative_draft_model_path=None, speculative_draft_model_revision=None, speculative_draft_load_format=None, speculative_num_steps=None, speculative_eagle_topk=None, speculative_num_draft_tokens=None, speculative_accept_threshold_single=1.0, speculative_accept_threshold_acc=1.0, speculative_token_map=None, speculative_attention_mode='prefill', speculative_moe_runner_backend=None, speculative_ngram_min_match_window_size=1, speculative_ngram_max_match_window_size=12, speculative_ngram_min_bfs_breadth=1, speculative_ngram_max_bfs_breadth=10, speculative_ngram_match_type='BFS', speculative_ngram_branch_length=18, speculative_ngram_capacity=10000000, ep_size=4, moe_a2a_backend='none', moe_runner_backend='auto', flashinfer_mxfp4_moe_precision='default', enable_flashinfer_allreduce_fusion=False, deepep_mode='auto', ep_num_redundant_experts=0, ep_dispatch_algorithm='static', init_expert_location='trivial', enable_eplb=False, eplb_algorithm='auto', eplb_rebalance_num_iterations=1000, eplb_rebalance_layers_per_chunk=None, eplb_min_rebalancing_utilization_threshold=1.0, expert_distribution_recorder_mode=None, expert_distribution_recorder_buffer_size=1000, enable_expert_distribution_metrics=False, deepep_config=None, moe_dense_tp_size=None, elastic_ep_backend=None, mooncake_ib_device=None, max_mamba_cache_size=None, mamba_ssm_dtype='float32', mamba_full_memory_ratio=0.9, enable_hierarchical_cache=False, hicache_ratio=2.0, hicache_size=0, hicache_write_policy='write_through', hicache_io_backend='kernel', hicache_mem_layout='layer_first', hicache_storage_backend=None, hicache_storage_prefetch_policy='best_effort', hicache_storage_backend_extra_config=None, enable_lmcache=False, kt_amx_weight_path=None, kt_amx_method='AMXINT4', kt_cpuinfer=None, kt_threadpool_count=2, kt_num_gpu_experts=None, enable_double_sparsity=False, ds_channel_config_path=None, ds_heavy_channel_num=32, ds_heavy_token_num=256, ds_heavy_channel_type='qk', ds_sparse_decode_threshold=4096, cpu_offload_gb=0, offload_group_size=-1, offload_num_in_group=1, offload_prefetch_step=1, offload_mode='cpu', multi_item_scoring_delimiter=None, disable_radix_cache=False, cuda_graph_max_bs=80, cuda_graph_bs=[1, 2, 4, 8, 12, 16, 24, 32, 40, 48, 56, 64, 72, 80], disable_cuda_graph=False, disable_cuda_graph_padding=False, enable_profile_cuda_graph=False, enable_cudagraph_gc=False, enable_nccl_nvls=False, enable_symm_mem=False, disable_flashinfer_cutlass_moe_fp4_allgather=False, enable_tokenizer_batch_encode=False, disable_tokenizer_batch_decode=False, disable_outlines_disk_cache=False, disable_custom_all_reduce=False, enable_mscclpp=False, enable_torch_symm_mem=False, disable_overlap_schedule=False, enable_mixed_chunk=False, enable_dp_attention=False, enable_dp_lm_head=False, enable_two_batch_overlap=False, enable_single_batch_overlap=False, tbo_token_distribution_threshold=0.48, enable_torch_compile=False, enable_piecewise_cuda_graph=False, torch_compile_max_bs=32, piecewise_cuda_graph_max_tokens=4096, piecewise_cuda_graph_tokens=[4, 8, 12, 16, 20, 24, 28, 32, 48, 64, 80, 96, 112, 128, 144, 160, 176, 192, 208, 224, 240, 256, 288, 320, 352, 384, 416, 448, 480, 512, 640, 768, 896, 1024, 1152, 1280, 1408, 1536, 1664, 1792, 1920, 2048, 2176, 2304, 2432, 2560, 2688, 2816, 2944, 3072, 3200, 3328, 3456, 3584, 3712, 3840, 3968, 4096], piecewise_cuda_graph_compiler='eager', torchao_config='', enable_nan_detection=False, enable_p2p_check=False, triton_attention_reduce_in_fp32=False, triton_attention_num_kv_splits=8, triton_attention_split_tile_size=None, num_continuous_decode_steps=1, delete_ckpt_after_loading=False, enable_memory_saver=False, enable_weights_cpu_backup=False, allow_auto_truncate=False, enable_custom_logit_processor=False, flashinfer_mla_disable_ragged=False, disable_shared_experts_fusion=False, disable_chunked_prefix_cache=False, disable_fast_image_processor=False, keep_mm_feature_on_device=False, enable_return_hidden_states=False, scheduler_recv_interval=1, numa_node=None, enable_deterministic_inference=False, rl_on_policy_target=None, enable_dynamic_batch_tokenizer=False, dynamic_batch_tokenizer_batch_size=32, dynamic_batch_tokenizer_batch_timeout=0.002, debug_tensor_dump_output_folder=None, debug_tensor_dump_layers=-1, debug_tensor_dump_input_file=None, debug_tensor_dump_inject=False, disaggregation_mode='null', disaggregation_transfer_backend='mooncake', disaggregation_bootstrap_port=8998, disaggregation_decode_tp=None, disaggregation_decode_dp=None, disaggregation_prefill_pp=1, disaggregation_ib_device=None, disaggregation_decode_enable_offload_kvcache=False, num_reserved_decode_tokens=512, disaggregation_decode_polling_interval=1, custom_weight_loader=[], weight_loader_disable_mmap=False, remote_instance_weight_loader_seed_instance_ip=None, remote_instance_weight_loader_seed_instance_service_port=None, remote_instance_weight_loader_send_weights_group_ports=None, enable_pdmux=False, pdmux_config_path=None, sm_group_num=8, mm_max_concurrent_calls=32, mm_per_request_timeout=10.0, decrypted_config_file=None, decrypted_draft_config_file=None)
[2025-11-06 10:35:52] auto-round quantization is not fully optimized yet. The speed can be slower than non-quantized models.
[2025-11-06 10:35:53] Using default HuggingFace chat template with detected content format: string
[2025-11-06 10:35:53] Detected the force reasoning pattern in chat template.
[2025-11-06 10:36:03] INFO trace.py:52: opentelemetry package is not installed, tracing disabled
[2025-11-06 10:36:03 TP0 EP0] auto-round quantization is not fully optimized yet. The speed can be slower than non-quantized models.
[2025-11-06 10:36:04] INFO trace.py:52: opentelemetry package is not installed, tracing disabled
[2025-11-06 10:36:04] INFO trace.py:52: opentelemetry package is not installed, tracing disabled
[2025-11-06 10:36:04 TP1 EP1] auto-round quantization is not fully optimized yet. The speed can be slower than non-quantized models.
[2025-11-06 10:36:04 TP2 EP2] auto-round quantization is not fully optimized yet. The speed can be slower than non-quantized models.
[2025-11-06 10:36:04] INFO trace.py:52: opentelemetry package is not installed, tracing disabled
[2025-11-06 10:36:04 TP3 EP3] auto-round quantization is not fully optimized yet. The speed can be slower than non-quantized models.
[2025-11-06 10:36:04 TP0 EP0] auto-round quantization is not fully optimized yet. The speed can be slower than non-quantized models.
[2025-11-06 10:36:04 TP0 EP0] Init torch distributed begin.
[2025-11-06 10:36:04 TP1 EP1] auto-round quantization is not fully optimized yet. The speed can be slower than non-quantized models.
[2025-11-06 10:36:04 TP1 EP1] Init torch distributed begin.
[2025-11-06 10:36:04 TP2 EP2] auto-round quantization is not fully optimized yet. The speed can be slower than non-quantized models.
[2025-11-06 10:36:04 TP2 EP2] Init torch distributed begin.
[2025-11-06 10:36:05] INFO trace.py:52: opentelemetry package is not installed, tracing disabled
[2025-11-06 10:36:05 TP3 EP3] auto-round quantization is not fully optimized yet. The speed can be slower than non-quantized models.
[2025-11-06 10:36:05 TP3 EP3] Init torch distributed begin.
[Gloo] Rank 3 is connected to 3 peer ranks. Expected number of connected peer ranks is : 3
[Gloo] Rank 0 is connected to 3 peer ranks. Expected number of connected peer ranks is : 3
[Gloo] Rank 2 is connected to 3 peer ranks. Expected number of connected peer ranks is : 3
[Gloo] Rank 1 is connected to 3 peer ranks. Expected number of connected peer ranks is : 3
[Gloo] Rank 0 is connected to 3 peer ranks. Expected number of connected peer ranks is : 3
[Gloo] Rank 1 is connected to 3 peer ranks. Expected number of connected peer ranks is : 3
[Gloo] Rank 3 is connected to 3 peer ranks. Expected number of connected peer ranks is : 3
[Gloo] Rank 2 is connected to 3 peer ranks. Expected number of connected peer ranks is : 3
[2025-11-06 10:36:05 TP0 EP0] sglang is using nccl==2.27.3
[2025-11-06 10:36:06 TP0 EP0] Custom allreduce is disabled because it's not supported on more than two PCIe-only GPUs. To silence this warning, specify disable_custom_all_reduce=True explicitly.
[2025-11-06 10:36:06 TP2 EP2] Custom allreduce is disabled because it's not supported on more than two PCIe-only GPUs. To silence this warning, specify disable_custom_all_reduce=True explicitly.
[2025-11-06 10:36:06 TP3 EP3] Custom allreduce is disabled because it's not supported on more than two PCIe-only GPUs. To silence this warning, specify disable_custom_all_reduce=True explicitly.
[2025-11-06 10:36:06 TP1 EP1] Custom allreduce is disabled because it's not supported on more than two PCIe-only GPUs. To silence this warning, specify disable_custom_all_reduce=True explicitly.
[Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
[Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
[Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
[Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
[Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
[Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
[Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
[Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
[Gloo] Rank 3 is connected to 3 peer ranks. Expected number of connected peer ranks is : 3
[Gloo] Rank 0 is connected to 3 peer ranks. Expected number of connected peer ranks is : 3
[Gloo] Rank 1 is connected to 3 peer ranks. Expected number of connected peer ranks is : 3
[Gloo] Rank 2 is connected to 3 peer ranks. Expected number of connected peer ranks is : 3
[2025-11-06 10:36:06 TP2 EP2] Init torch distributed ends. mem usage=0.14 GB
[2025-11-06 10:36:06 TP3 EP3] Init torch distributed ends. mem usage=0.14 GB
[2025-11-06 10:36:06 TP0 EP0] Init torch distributed ends. mem usage=0.14 GB
[2025-11-06 10:36:06 TP1 EP1] Init torch distributed ends. mem usage=0.14 GB
[2025-11-06 10:36:11 TP0 EP0] Load weight begin. avail mem=22.71 GBtorch_dtype is deprecated! Use dtype instead!
[2025-11-06 10:36:11 TP0 EP0] using attn output gate!
[2025-11-06 10:36:11 TP3 EP3] Load weight begin. avail mem=23.05 GBtorch_dtype is deprecated! Use dtype instead!
[2025-11-06 10:36:11 TP2 EP2] Load weight begin. avail mem=23.05 GB
[2025-11-06 10:36:11 TP3 EP3] using attn output gate!torch_dtype is deprecated! Use dtype instead!
[2025-11-06 10:36:11 TP2 EP2] using attn output gate!
[2025-11-06 10:36:12 TP1 EP1] Load weight begin. avail mem=23.06 GBtorch_dtype is deprecated! Use dtype instead!
[2025-11-06 10:36:12 TP1 EP1] using attn output gate!
[2025-11-06 10:36:12 TP2 EP2] Scheduler hit an exception: Traceback (most recent call last):
File "/root/sgl-0.5.4.2/.venv/lib/python3.12/site-packages/sglang/srt/managers/scheduler.py", line 2791, in run_scheduler_process
scheduler = Scheduler(
^^^^^^^^^^
File "/root/sgl-0.5.4.2/.venv/lib/python3.12/site-packages/sglang/srt/managers/scheduler.py", line 319, in init
self.tp_worker = TpModelWorker(
^^^^^^^^^^^^^^
File "/root/sgl-0.5.4.2/.venv/lib/python3.12/site-packages/sglang/srt/managers/tp_worker.py", line 237, in init
self._model_runner = ModelRunner(
^^^^^^^^^^^^
File "/root/sgl-0.5.4.2/.venv/lib/python3.12/site-packages/sglang/srt/model_executor/model_runner.py", line 322, in init
self.initialize(min_per_gpu_memory)
File "/root/sgl-0.5.4.2/.venv/lib/python3.12/site-packages/sglang/srt/model_executor/model_runner.py", line 398, in initialize
self.load_model()
File "/root/sgl-0.5.4.2/.venv/lib/python3.12/site-packages/sglang/srt/model_executor/model_runner.py", line 752, in load_model
self.model = get_model(
^^^^^^^^^^
File "/root/sgl-0.5.4.2/.venv/lib/python3.12/site-packages/sglang/srt/model_loader/init.py", line 28, in get_model
return loader.load_model(
^^^^^^^^^^^^^^^^^^
File "/root/sgl-0.5.4.2/.venv/lib/python3.12/site-packages/sglang/srt/model_loader/loader.py", line 599, in load_model
self.load_weights_and_postprocess(
File "/root/sgl-0.5.4.2/.venv/lib/python3.12/site-packages/sglang/srt/model_loader/loader.py", line 607, in load_weights_and_postprocess
model.load_weights(weights)
File "/root/sgl-0.5.4.2/.venv/lib/python3.12/site-packages/sglang/srt/models/qwen3_next.py", line 1054, in load_weights
param = params_dict[name]
~~~~~~~~~~~^^^^^^
KeyError: 'model.layers.28.linear_attn.in_proj_ba.qweight'
[2025-11-06 10:36:12] Received sigquit from a child process. It usually means the child failed.
Loading safetensors checkpoint shards: 0% Completed | 0/9 [00:00<?, ?it/s]
[2025-11-06 10:36:12 TP3 EP3] Scheduler hit an exception: Traceback (most recent call last):
File "/root/sgl-0.5.4.2/.venv/lib/python3.12/site-packages/sglang/srt/managers/scheduler.py", line 2791, in run_scheduler_process
scheduler = Scheduler(
^^^^^^^^^^
File "/root/sgl-0.5.4.2/.venv/lib/python3.12/site-packages/sglang/srt/managers/scheduler.py", line 319, in init
self.tp_worker = TpModelWorker(
^^^^^^^^^^^^^^
File "/root/sgl-0.5.4.2/.venv/lib/python3.12/site-packages/sglang/srt/managers/tp_worker.py", line 237, in init
self._model_runner = ModelRunner(
^^^^^^^^^^^^
File "/root/sgl-0.5.4.2/.venv/lib/python3.12/site-packages/sglang/srt/model_executor/model_runner.py", line 322, in init
self.initialize(min_per_gpu_memory)
File "/root/sgl-0.5.4.2/.venv/lib/python3.12/site-packages/sglang/srt/model_executor/model_runner.py", line 398, in initialize
self.load_model()
File "/root/sgl-0.5.4.2/.venv/lib/python3.12/site-packages/sglang/srt/model_executor/model_runner.py", line 752, in load_model
self.model = get_model(
^^^^^^^^^^
File "/root/sgl-0.5.4.2/.venv/lib/python3.12/site-packages/sglang/srt/model_loader/init.py", line 28, in get_model
return loader.load_model(
^^^^^^^^^^^^^^^^^^
File "/root/sgl-0.5.4.2/.venv/lib/python3.12/site-packages/sglang/srt/model_loader/loader.py", line 599, in load_model
self.load_weights_and_postprocess(
File "/root/sgl-0.5.4.2/.venv/lib/python3.12/site-packages/sglang/srt/model_loader/loader.py", line 607, in load_weights_and_postprocess
model.load_weights(weights)
File "/root/sgl-0.5.4.2/.venv/lib/python3.12/site-packages/sglang/srt/models/qwen3_next.py", line 1054, in load_weights
param = params_dict[name]
~~~~~~~~~~~^^^^^^
KeyError: 'model.layers.28.linear_attn.in_proj_ba.qweight'
[2025-11-06 10:36:12] Received sigquit from a child process. It usually means the child failed.
Using --attention-backend torch_native still got this error.
Could you please let me know if there is a way to load this model with SGLang?
Thank you very much!