[MODEL] support minicpm-o 2.6

gptqmodel/models/_const.py

@@ -155,6 +155,7 @@ def get_best_device(backend: BACKEND = BACKEND.AUTO) -> torch.device:
     "cohere2",
     "minicpm",
     "minicpm3"
+    "minicpmo"
     "qwen2_moe",
     "qwen2_vl",
     "dbrx_converted",

gptqmodel/models/auto.py

@@ -66,6 +66,7 @@
 from .definitions.longllama import LongLlamaGPTQ  # noqa: E402
 from .definitions.minicpm import MiniCPMGPTQ  # noqa: E402
 from .definitions.minicpm3 import MiniCPM3GPTQ  # noqa: E402
+from .definitions.minicpmo import MiniCPMOGPTQ  # noqa: E402
 from .definitions.mistral import MistralGPTQ  # noqa: E402
 from .definitions.mixtral import MixtralGPTQ  # noqa: E402
 from .definitions.mllama import MLlamaGPTQ  # noqa: E402
@@ -126,7 +127,8 @@
     "phi3": Phi3GPTQ,
     "mpt": MPTGPTQ,
     "minicpm": MiniCPMGPTQ,
-    "minicpm3":MiniCPM3GPTQ,
+    "minicpm3": MiniCPM3GPTQ,
+    "minicpmo": MiniCPMOGPTQ,
     "qwen2_moe": Qwen2MoeGPTQ,
     "qwen2_vl": Qwen2VLGPTQ,
     "dbrx": DbrxGPTQ,

gptqmodel/models/definitions/__init__.py

@@ -39,6 +39,7 @@
 from .llama import LlamaGPTQ
 from .longllama import LongLlamaGPTQ
 from .minicpm3 import MiniCPM3GPTQ
+from .minicpmo import MiniCPMOGPTQ
 from .mistral import MistralGPTQ
 from .mixtral import MixtralGPTQ
 from .mllama import MLlamaGPTQ

gptqmodel/models/definitions/minicpmo.py

@@ -0,0 +1,29 @@
+# Copyright 2025 ModelCloud
+# Contact: [email protected], x.com/qubitium
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from ..base import BaseGPTQModel
+
+
+class MiniCPMOGPTQ(BaseGPTQModel):
+    layer_type = "Qwen2DecoderLayer"
+    layers_node = "llm.model.layers"
+    base_modules = ["llm.model.embed_tokens", "llm.model.norm", "vpm", "apm", "tts", "audio_projection_layer",
+                    "resampler"]
+    layer_modules = [
+        ["self_attn.k_proj", "self_attn.v_proj", "self_attn.q_proj"],
+        ["self_attn.o_proj"],
+        ["mlp.up_proj", "mlp.gate_proj"],
+        ["mlp.down_proj"],
+    ]