add mobilellm support

gptqmodel/models/_const.py

@@ -59,6 +59,7 @@ def get_device_by_type(type_value: str):
     "grinmoe",
     "mllama",
     "granite",
+    "mobilellm",
 ]
 
 EXLLAMA_DEFAULT_MAX_INPUT_LENGTH = 2048

gptqmodel/models/auto.py

@@ -45,6 +45,7 @@
 from .definitions.starcoder2 import Starcoder2GPTQ
 from .definitions.xverse import XverseGPTQ
 from .definitions.yi import YiGPTQ
+from .definitions.mobilellm import MobileLLMGPTQ
 
 MODEL_MAP = {
     "bloom": BloomGPTQ,
@@ -89,6 +90,7 @@
     "grinmoe": GrinMOEGPTQ,
     "mllama": MLlamaGPTQ,
     "granite": GraniteGPTQ,
+    "mobilellm": MobileLLMGPTQ,
 }
 
 

gptqmodel/models/definitions/__init__.py

@@ -36,4 +36,5 @@
 from .stablelmepoch import StableLMEpochGPTQ
 from .starcoder2 import Starcoder2GPTQ
 from .xverse import XverseGPTQ
-from .yi import YiGPTQ
+from .yi import YiGPTQ
+from .mobilellm import MobileLLMGPTQ

gptqmodel/models/definitions/mobilellm.py

@@ -0,0 +1,16 @@
+from ..base import BaseGPTQModel
+
+
+class MobileLLMGPTQ(BaseGPTQModel):
+    require_trust_remote_code = True
+
+    base_modules = ["model.embed_tokens", "model.norm"]
+
+    layers_node = "model.layers"
+    layer_type = "LlamaDecoderLayer"
+    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"],
+    ]