self.gate dtype update for GLM-4.5 (vllm-project#22203)

Cherry-pick: vllm-project@6fa41e0

Signed-off-by: zRzRzRzRzRzRzR <2448370773@qq.com>

Author: zRzRzRzRzRzRzR

docs/models/supported_models.md

@@ -524,7 +524,7 @@ Specified using `--task generate`.
 | `GLM4VForCausalLM`<sup>^</sup>               | GLM-4V                                                                   | T + I                                                                 | `zai-org/glm-4v-9b`, `zai-org/cogagent-9b-20241220` etc.                                                                                                    | ✅︎                     | ✅︎                          | ✅︎                    |
 | `Glm4vForConditionalGeneration`              | GLM-4.1V-Thinking                                                        | T + I<sup>E+</sup> + V<sup>E+</sup>                                   | `zai-org/GLM-4.1V-9B-Thinkg`,  etc.                                                                                                                       | ✅︎                     | ✅︎                          | ✅︎                    |
 | `Glm4MoeForCausalLM` | GLM-4.5 | T + I<sup>E+</sup> + V<sup>E+</sup> | `zai-org/GLM-4.5`, etc. | ✅︎ | ✅︎ | ✅︎ |
-| `Glm4v_moeForConditionalGeneration` | GLM-4.5V | T + I<sup>E+</sup> + V<sup>E+</sup> | `zai-org/GLM-4.5V-Air`, etc. | ✅︎ | ✅︎ | ✅︎ |
+| `Glm4v_moeForConditionalGeneration` | GLM-4.5V | T + I<sup>E+</sup> + V<sup>E+</sup> | `zai-org/GLM-4.5V`, etc. | ✅︎ | ✅︎ | ✅︎ |
 | `GraniteSpeechForConditionalGeneration`      | Granite Speech                                                           | T + A                                                                 | `ibm-granite/granite-speech-3.3-8b`                                                                                                                     | ✅︎                     | ✅︎                          | ✅︎                    |
 | `H2OVLChatModel`                             | H2OVL                                                                    | T + I<sup>E+</sup>                                                    | `h2oai/h2ovl-mississippi-800m`, `h2oai/h2ovl-mississippi-2b`, etc.                                                                                      |                       | ✅︎                          | ✅︎\*                     |
 | `Idefics3ForConditionalGeneration`           | Idefics3                                                                 | T + I                                                                 | `HuggingFaceM4/Idefics3-8B-Llama3` etc.                                                                                                                 | ✅︎                     |                           |  ✅︎                     |

tests/models/registry.py

@@ -325,7 +325,7 @@ def check_available_online(
                                         trust_remote_code=True,
                                         hf_overrides={"architectures": ["GLM4VForCausalLM"]}),  # noqa: E501
     "Glm4vForConditionalGeneration": _HfExamplesInfo("zai-org/GLM-4.1V-9B-Thinking"),  # noqa: E501
-    "Glm4v_moeForConditionalGeneration": _HfExamplesInfo("zai-org/GLM-4.5V-Air",
+    "Glm4v_moeForConditionalGeneration": _HfExamplesInfo("zai-org/GLM-4.5V",
                                           is_available_online=False),   # noqa: E501
     "H2OVLChatModel": _HfExamplesInfo("h2oai/h2ovl-mississippi-800m",
                                       extras={"2b": "h2oai/h2ovl-mississippi-2b"},  # noqa: E501

vllm/model_executor/models/glm4_moe.py

@@ -123,6 +123,7 @@ def __init__(
                                      config.n_routed_experts,
                                      bias=False,
                                      quant_config=None,
+                                     params_dtype=torch.float32,
                                      prefix=f"{prefix}.gate")
 
         self.gate.e_score_correction_bias = nn.Parameter(
@@ -180,7 +181,7 @@ def forward(self, hidden_states: torch.Tensor) -> torch.Tensor:
 
         if self.n_shared_experts is not None:
             shared_output = self.shared_experts(hidden_states)
-        router_logits, _ = self.gate(hidden_states)
+        router_logits, _ = self.gate(hidden_states.to(dtype=torch.float32))
         final_hidden_states = self.experts(
             hidden_states=hidden_states,
             router_logits=router_logits) * self.routed_scaling_factor