Normalize head weights for Baichuan 2 (#1876)

vllm-project · Dec 1, 2023 · e5452dd · e5452dd
1 parent d06980d
commit e5452dd
Show file tree

Hide file tree

Showing 3 changed files with 13 additions and 2 deletions.
diff --git a/README.md b/README.md
@@ -47,7 +47,7 @@ vLLM is flexible and easy to use with:
 vLLM seamlessly supports many Hugging Face models, including the following architectures:
 
 - Aquila & Aquila2 (`BAAI/AquilaChat2-7B`, `BAAI/AquilaChat2-34B`, `BAAI/Aquila-7B`, `BAAI/AquilaChat-7B`, etc.)
-- Baichuan (`baichuan-inc/Baichuan-7B`, `baichuan-inc/Baichuan-13B-Chat`, etc.)
+- Baichuan & Baichuan2 (`baichuan-inc/Baichuan2-13B-Chat`, `baichuan-inc/Baichuan-7B`, etc.)
 - BLOOM (`bigscience/bloom`, `bigscience/bloomz`, etc.)
 - ChatGLM (`THUDM/chatglm2-6b`, `THUDM/chatglm3-6b`, etc.)
 - Falcon (`tiiuae/falcon-7b`, `tiiuae/falcon-40b`, `tiiuae/falcon-rw-7b`, etc.)

diff --git a/docs/source/models/supported_models.rst b/docs/source/models/supported_models.rst
@@ -19,7 +19,7 @@ Alongside each architecture, we include some popular models that use it.
     - :code:`BAAI/Aquila-7B`, :code:`BAAI/AquilaChat-7B`, etc.
   * - :code:`BaiChuanForCausalLM`
     - Baichuan
-    - :code:`baichuan-inc/Baichuan-7B`, :code:`baichuan-inc/Baichuan-13B-Chat`, etc.
+    - :code:`baichuan-inc/Baichuan2-13B-Chat`, :code:`baichuan-inc/Baichuan-7B`, etc.
   * - :code:`ChatGLMModel`
     - ChatGLM
     - :code:`THUDM/chatglm2-6b`, :code:`THUDM/chatglm3-6b`, etc.

diff --git a/vllm/model_executor/models/baichuan.py b/vllm/model_executor/models/baichuan.py
@@ -341,6 +341,17 @@ def load_weights(self,
                 model_name_or_path, cache_dir, load_format, revision):
             if "rotary_emb.inv_freq" in name:
                 continue
+            if name == "lm_head.weight":
+                # Unlike Baichuan, Baichuan2 normalizes the head weights. Refer to:
+                # https://huggingface.co/baichuan-inc/Baichuan2-7B-Chat/blob/84603cde5ebffb6084e476cfaeceaf0b8b91fe54/modeling_baichuan.py#L508
+                # Distinguish between Baichuan and Baichuan2 by checking the
+                # vocab size. This is suggested by
+                # https://github.com/vllm-project/vllm/pull/1022#discussion_r1325652704
+                is_baichuan2 = self.config.vocab_size == 125696
+                if is_baichuan2:
+                    loaded_weight = torch.nn.functional.normalize(
+                        loaded_weight)
+
             for (param_name, weight_name, shard_id) in stacked_params_mapping:
                 if weight_name not in name:
                     continue