support for minicpm3 (#385)

* support for minicpm3 * Update auto.py
ModelCloud · Sep 10, 2024 · 06b5c4e · 06b5c4e
1 parent cc9dcc0
commit 06b5c4e
Show file tree

Hide file tree

Showing 4 changed files with 19 additions and 0 deletions.
diff --git a/gptqmodel/models/__init__.py b/gptqmodel/models/__init__.py
@@ -35,3 +35,4 @@
 from .starcoder2 import Starcoder2GPTQ
 from .xverse import XverseGPTQ
 from .yi import YiGPTQ
+from .minicpm3 import MiniCPM3GPTQ
diff --git a/gptqmodel/models/_const.py b/gptqmodel/models/_const.py
@@ -51,6 +51,7 @@ def get_device_by_type(type_value: str):
     "starcoder2",
     "cohere",
     "minicpm",
+    "minicpm3"
     "qwen2_moe",
     "dbrx_converted",
     "deepseek_v2",

diff --git a/gptqmodel/models/auto.py b/gptqmodel/models/auto.py
@@ -25,6 +25,7 @@
 from .llama import LlamaGPTQ
 from .longllama import LongLlamaGPTQ
 from .minicpm import MiniCPMGPTQ
+from .minicpm3 import MiniCPM3GPTQ
 from .mistral import MistralGPTQ
 from .mixtral import MixtralGPTQ
 from .moss import MOSSGPTQ
@@ -75,6 +76,7 @@
     "phi3": Phi3GPTQ,
     "mpt": MPTGPTQ,
     "minicpm": MiniCPMGPTQ,
+    "minicpm3":MiniCPM3GPTQ,
     "qwen2_moe": Qwen2MoeGPTQ,
     "dbrx": DbrxGPTQ,
     "dbrx_converted": DbrxConvertedGPTQ,

diff --git a/gptqmodel/models/minicpm3.py b/gptqmodel/models/minicpm3.py
@@ -0,0 +1,15 @@
+from ._base import BaseGPTQModel
+class MiniCPM3GPTQ(BaseGPTQModel):
+    layer_type = "MiniCPM3DecoderLayer"
+    layers_block_name = "model.layers"
+    outside_layer_modules = [
+        "model.embed_tokens",
+    ]
+    inside_layer_modules = [
+        ["self_attn.q_a_proj","self_attn.kv_a_proj_with_mqa"],
+        ["self_attn.q_b_proj","self_attn.kv_b_proj"],
+        ["self_attn.o_proj"],
+        ["mlp.gate_proj","mlp.up_proj"],
+        ["mlp.down_proj"],
+    ]
+__all__ = ["MiniCPM3GPTQForCausalLM"]