[quantization] Fix NameError: name 'WNA16_SUPPORTED_BITS' is not defined

[kevin85421]

Motivation

This PR fixes the NameError mentioned in #11383 by importing WNA16_SUPPORTED_BITS.

• Run:

  python3 -m sglang.launch_server --model-path cpatonn/Qwen3-30B-A3B-Thinking-2507-AWQ-4bit

• Before this PR:

  File ".../sglang/python/sglang/srt/layers/quantization/compressed_tensors/compressed_tensors_moe.py", line 369, in __init__
    and self.num_bits in WNA16_SUPPORTED_BITS
                         ^^^^^^^^^^^^^^^^^^^^
  NameError: name 'WNA16_SUPPORTED_BITS' is not defined
  
  [2025-10-13 16:49:19] Received sigquit from a child process. It usually means the child failed.
  [1]    1519263 killed     python3 -m sglang.launch_server --model-path
  

• After this PR => The NameError is resolved, but it failed due to other issues. Users can update the configuration to try to launch the server.

   File ".../sglang/python/sglang/srt/layers/quantization/compressed_tensors/compressed_tensors_moe.py", line 388, in create_weights
     params_dtype == torch.float16
  AssertionError: float16 is required for MoE compressed models. Set dtype=torch.float16
  
  [2025-10-13 16:51:24] Received sigquit from a child process. It usually means the child failed.
  [1]    1519863 killed     python3 -m sglang.launch_server --model-path
  

Modifications

Accuracy Tests

Benchmarking and Profiling

Checklist

• Format your code according to the Format code with pre-commit.
• Add unit tests according to the Run and add unit tests.
• Update documentation according to Write documentations.
• Provide accuracy and speed benchmark results according to Test the accuracy and Benchmark the speed.

[merrymercy]

1. please fix the failed vllm dependency test
2. please avoid the import from vllm. For AWQ, i think sglang should be able to run it without any import from vllm

[kevin85421]

Thanks for the reivew!

please fix the failed vllm dependency test

The error does not seem to relate to this PR.

command=python3 -m sglang.launch_server --model-path hugging-quants/Mixtral-8x7B-Instruct-v0.1-AWQ-INT4 --trust-remote-code --device cuda --host 127.0.0.1 --port 13000
======================================================================
ERROR: setUpClass (__main__.TestAWQ)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/public_sglang_ci/runner-l1f-gpu-23/_work/sglang/sglang/test/srt/quant/test_awq.py", line 20, in setUpClass
    cls.process = popen_launch_server(
  File "/public_sglang_ci/runner-l1f-gpu-23/_work/sglang/sglang/python/sglang/test/test_utils.py", line 641, in popen_launch_server
    raise TimeoutError("Server failed to start within the timeout period.")
TimeoutError: Server failed to start within the timeout period.

----------------------------------------------------------------------
Ran 0 tests in 600.574s

please avoid the import from vllm. For AWQ, i think sglang should be able to run it without any import from vllm

I will take a look.

[Hongbosherlock]

1. please fix the failed vllm dependency test
2. please avoid the import from vllm. For AWQ, i think sglang should be able to run it without any import from vllm

@merrymercy I have already worked on it. It's ready to be merged now.#10750