[Model] support MiniMax-VL-01 model (#16328)

Signed-off-by: qingjun <qingjun@minimaxi.com>

vllm/transformers_utils/configs/__init__.py

@@ -15,6 +15,8 @@ from vllm.transformers_utils.configs.internvl import InternVLChatConfig
 from vllm.transformers_utils.configs.jais import JAISConfig
 from vllm.transformers_utils.configs.kimi_vl import KimiVLConfig
 from vllm.transformers_utils.configs.medusa import MedusaConfig
+from vllm.transformers_utils.configs.minimax_text_01 import MiniMaxText01Config
+from vllm.transformers_utils.configs.minimax_vl_01 import MiniMaxVL01Config
 from vllm.transformers_utils.configs.mllama import MllamaConfig
 from vllm.transformers_utils.configs.mlp_speculator import MLPSpeculatorConfig
 from vllm.transformers_utils.configs.moonvit import MoonViTConfig
@@ -39,6 +41,8 @@ __all__ = [
     "MedusaConfig",
     "EAGLEConfig",
     "ExaoneConfig",
+    "MiniMaxText01Config",
+    "MiniMaxVL01Config",
     "MllamaConfig",
     "MLPSpeculatorConfig",
     "MoonViTConfig",

vllm/transformers_utils/configs/minimax_text_01.py

@@ -0,0 +1,69 @@
+# SPDX-License-Identifier: Apache-2.0
+""" MiniMaxText01 model configuration"""
+
+from transformers.configuration_utils import PretrainedConfig
+
+
+class MiniMaxText01Config(PretrainedConfig):
+    model_type = "MiniMaxText01"
+    keys_to_ignore_at_inference = ["past_key_values"]
+
+    def __init__(
+        self,
+        vocab_size=32000,
+        hidden_size=4096,
+        intermediate_size=14336,
+        num_hidden_layers=32,
+        num_attention_heads=32,
+        num_key_value_heads=8,
+        hidden_act="silu",
+        max_position_embeddings=4096 * 32,
+        initializer_range=0.02,
+        rms_norm_eps=1e-5,
+        use_cache=True,
+        pad_token_id=None,
+        bos_token_id=None,
+        eos_token_id=None,
+        tie_word_embeddings=False,
+        rope_theta=1e6,
+        sliding_window=None,
+        attention_dropout=0.0,
+        num_experts_per_tok=2,
+        num_local_experts=8,
+        output_router_logits=False,
+        router_aux_loss_coef=0.001,
+        router_jitter_noise=0.0,
+        **kwargs,
+    ):
+        self.vocab_size = vocab_size
+        self.max_position_embeddings = max_position_embeddings
+        self.hidden_size = hidden_size
+        self.intermediate_size = intermediate_size
+        self.num_hidden_layers = num_hidden_layers
+        self.num_attention_heads = num_attention_heads
+        self.sliding_window = sliding_window
+
+        # for backward compatibility
+        if num_key_value_heads is None:
+            num_key_value_heads = num_attention_heads
+
+        self.num_key_value_heads = num_key_value_heads
+        self.hidden_act = hidden_act
+        self.initializer_range = initializer_range
+        self.rms_norm_eps = rms_norm_eps
+        self.use_cache = use_cache
+        self.rope_theta = rope_theta
+        self.attention_dropout = attention_dropout
+
+        self.num_experts_per_tok = num_experts_per_tok
+        self.num_local_experts = num_local_experts
+        self.output_router_logits = output_router_logits
+        self.router_aux_loss_coef = router_aux_loss_coef
+        self.router_jitter_noise = router_jitter_noise
+        super().__init__(
+            pad_token_id=pad_token_id,
+            bos_token_id=bos_token_id,
+            eos_token_id=eos_token_id,
+            tie_word_embeddings=tie_word_embeddings,
+            **kwargs,
+        )

vllm/transformers_utils/configs/minimax_vl_01.py

@@ -0,0 +1,70 @@
+# SPDX-License-Identifier: Apache-2.0
+"""MiniMaxVL01 model configuration"""
+
+from transformers.configuration_utils import PretrainedConfig
+from transformers.models.auto import CONFIG_MAPPING
+
+from .minimax_text_01 import MiniMaxText01Config
+
+
+class MiniMaxVL01Config(PretrainedConfig):
+    model_type = "minimax_vl_01"
+
+    def __init__(
+        self,
+        vision_config=None,
+        text_config=None,
+        ignore_index=-100,
+        image_token_index=32000,
+        projector_hidden_act="gelu",
+        vision_feature_select_strategy="default",
+        vision_feature_layer=-2,
+        image_grid_pinpoints=None,
+        tie_word_embeddings=False,
+        image_seq_length=576,
+        **kwargs,
+    ):
+        self.ignore_index = ignore_index
+        self.image_token_index = image_token_index
+        self.projector_hidden_act = projector_hidden_act
+        self.image_seq_length = image_seq_length
+
+        if vision_feature_select_strategy not in ["default", "full"]:
+            raise ValueError("vision_feature_select_strategy should " +
+                             "be one of 'default', 'full'." +
+                             f"Got: {vision_feature_select_strategy}")
+
+        self.vision_feature_select_strategy = vision_feature_select_strategy
+        self.vision_feature_layer = vision_feature_layer
+        image_grid_pinpoints = (
+            image_grid_pinpoints if image_grid_pinpoints is not None else
+            [[336, 672], [672, 336], [672, 672], [1008, 336], [336, 1008]])
+        self.image_grid_pinpoints = image_grid_pinpoints
+
+        if isinstance(vision_config, dict):
+            if "model_type" not in vision_config:
+                vision_config["model_type"] = "clip_vision_model"
+            vision_config = CONFIG_MAPPING[vision_config["model_type"]](
+                **vision_config)
+        elif vision_config is None:
+            vision_config = CONFIG_MAPPING["clip_vision_model"](
+                intermediate_size=4096,
+                hidden_size=1024,
+                patch_size=14,
+                image_size=336,
+                num_hidden_layers=24,
+                num_attention_heads=16,
+                vocab_size=32000,
+                projection_dim=768,
+            )
+
+        self.vision_config = vision_config
+
+        if text_config is not None:
+            text_config = MiniMaxText01Config(**text_config)
+        else:
+            text_config = MiniMaxText01Config()
+
+        self.text_config = text_config
+
+        super().__init__(tie_word_embeddings=tie_word_embeddings, **kwargs)