[Bugfix] Fixes for new marlin moe usage

[mgoin]

FIX #16850 (comment)

Found using failing tests in "Weight Loading Multiple GPU Test - Large Models" on buildkite: https://buildkite.com/vllm/ci/builds/19754/steps?jid=0196bd80-d187-4367-af2f-a8fada079333#/43-1306

Minimal reproductions

Issue: Passing in group_size=None for channelwise models in compressed-tensors format

vllm serve nm-testing/Mixtral-8x7B-Instruct-v0.1-W4A16-channel-quantized -tp 2 --load-format dummy
...
(VllmWorker rank=1 pid=2902434) ERROR 05-12 18:46:04 [multiproc_executor.py:487]   File "/home/mgoin/code/vllm/vllm/model_executor/layers/fused_moe/layer.py", line 498, in __init__
(VllmWorker rank=1 pid=2902434) ERROR 05-12 18:46:04 [multiproc_executor.py:487]     self.quant_method = quant_config.get_quant_method(self, prefix)
(VllmWorker rank=1 pid=2902434) ERROR 05-12 18:46:04 [multiproc_executor.py:487]                         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(VllmWorker rank=1 pid=2902434) ERROR 05-12 18:46:04 [multiproc_executor.py:487]   File "/home/mgoin/code/vllm/vllm/model_executor/layers/quantization/compressed_tensors/compressed_tensors.py", line 101, in get_quant_method
(VllmWorker rank=1 pid=2902434) ERROR 05-12 18:46:04 [multiproc_executor.py:487]     return CompressedTensorsMoEMethod.get_moe_method(self, layer)
(VllmWorker rank=1 pid=2902434) ERROR 05-12 18:46:04 [multiproc_executor.py:487]            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(VllmWorker rank=1 pid=2902434) ERROR 05-12 18:46:04 [multiproc_executor.py:487]   File "/home/mgoin/code/vllm/vllm/model_executor/layers/quantization/compressed_tensors/compressed_tensors_moe.py", line 62, in get_moe_method
(VllmWorker rank=1 pid=2902434) ERROR 05-12 18:46:04 [multiproc_executor.py:487]     if not check_moe_marlin_supports_layer(layer,
(VllmWorker rank=1 pid=2902434) ERROR 05-12 18:46:04 [multiproc_executor.py:487]            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(VllmWorker rank=1 pid=2902434) ERROR 05-12 18:46:04 [multiproc_executor.py:487]   File "/home/mgoin/code/vllm/vllm/model_executor/layers/quantization/utils/marlin_utils.py", line 179, in check_moe_marlin_supports_layer
(VllmWorker rank=1 pid=2902434) ERROR 05-12 18:46:04 [multiproc_executor.py:487]     intermediate_size_per_partition % max(64, group_size) == 0 and \
(VllmWorker rank=1 pid=2902434) ERROR 05-12 18:46:04 [multiproc_executor.py:487]                                       ^^^^^^^^^^^^^^^^^^^
(VllmWorker rank=1 pid=2902434) ERROR 05-12 18:46:04 [multiproc_executor.py:487] TypeError: '>' not supported between instances of 'NoneType' and 'int'

Issue: The check for apply_router_weight_on_input in GPTQMarlinMoEMethod did not check boolean value

vllm serve TheBloke/Mixtral-8x7B-v0.1-GPTQ -tp 2 --load-format dummy --enforce-eager
...
(VllmWorker rank=0 pid=2933636) ERROR 05-12 19:06:21 [multiproc_executor.py:522]   File "/home/mgoin/code/vllm/vllm/model_executor/layers/fused_moe/layer.py", line 875, in forward_impl
(VllmWorker rank=0 pid=2933636) ERROR 05-12 19:06:21 [multiproc_executor.py:522]     final_hidden_states = self.quant_method.apply(
(VllmWorker rank=0 pid=2933636) ERROR 05-12 19:06:21 [multiproc_executor.py:522]                           ^^^^^^^^^^^^^^^^^^^^^^^^
(VllmWorker rank=0 pid=2933636) ERROR 05-12 19:06:21 [multiproc_executor.py:522]   File "/home/mgoin/code/vllm/vllm/model_executor/layers/quantization/gptq_marlin.py", line 614, in apply
(VllmWorker rank=0 pid=2933636) ERROR 05-12 19:06:21 [multiproc_executor.py:522]     raise NotImplementedError(
(VllmWorker rank=0 pid=2933636) ERROR 05-12 19:06:21 [multiproc_executor.py:522] NotImplementedError: Apply router weight on input is not supported forfused Marlin MoE method.

[github-actions]

👋 Hi! Thank you for contributing to the vLLM project.

💬 Join our developer Slack at https://slack.vllm.ai to discuss your PR in #pr-reviews, coordinate on features in #feat- channels, or join special interest groups in #sig- channels.

Just a reminder: PRs would not trigger full CI run by default. Instead, it would only run fastcheck CI which starts running only a small and essential subset of CI tests to quickly catch errors. You can run other CI tests on top of those by going to your fastcheck build on Buildkite UI (linked in the PR checks section) and unblock them. If you do not have permission to unblock, ping simon-mo or khluu to add you in our Buildkite org.

Once the PR is approved and ready to go, your PR reviewer(s) can run CI to test the changes comprehensively before merging.

To run CI, PR reviewers can either: Add ready label to the PR or enable auto-merge.

🚀