From f4fd390f5de585fd94877158bea4e1b2d1920df3 Mon Sep 17 00:00:00 2001
From: Michael Goin <michael@neuralmagic.com>
Date: Thu, 1 Aug 2024 15:01:07 -0400
Subject: [PATCH] [Bugfix] Lower gemma's unloaded_params exception to warning
 (#7002)

---
 vllm/model_executor/models/gemma.py     | 6 +++---
 vllm/model_executor/models/gemma2.py    | 9 ++++++---
 vllm/model_executor/models/paligemma.py | 6 +++---
 3 files changed, 12 insertions(+), 9 deletions(-)

diff --git a/vllm/model_executor/models/gemma.py b/vllm/model_executor/models/gemma.py
index 7e0888b5..64aef102 100644
--- a/vllm/model_executor/models/gemma.py
+++ b/vllm/model_executor/models/gemma.py
@@ -404,6 +404,6 @@ class GemmaForCausalLM(nn.Module, SupportsLoRA):
             loaded_params.add(name)
         unloaded_params = params_dict.keys() - loaded_params
         if unloaded_params:
-            raise RuntimeError(
-                "Some weights are not initialized from checkpoints: "
-                f"{unloaded_params}")
+            logger.warning(
+                "Some weights are not initialized from checkpoints: %s",
+                unloaded_params)
diff --git a/vllm/model_executor/models/gemma2.py b/vllm/model_executor/models/gemma2.py
index 8386084c..b77c901f 100644
--- a/vllm/model_executor/models/gemma2.py
+++ b/vllm/model_executor/models/gemma2.py
@@ -23,6 +23,7 @@ from transformers import Gemma2Config
 from vllm.attention import Attention, AttentionMetadata
 from vllm.config import CacheConfig, LoRAConfig
 from vllm.distributed import get_tensor_model_parallel_world_size
+from vllm.logger import init_logger
 from vllm.model_executor.layers.activation import GeluAndMul
 from vllm.model_executor.layers.layernorm import GemmaRMSNorm
 from vllm.model_executor.layers.linear import (MergedColumnParallelLinear,
@@ -41,6 +42,8 @@ from vllm.sequence import IntermediateTensors, SamplerOutput
 
 from .interfaces import SupportsLoRA
 
+logger = init_logger(__name__)
+
 
 class Gemma2MLP(nn.Module):
 
@@ -390,6 +393,6 @@ class Gemma2ForCausalLM(nn.Module, SupportsLoRA):
 
         unloaded_params = params_dict.keys() - loaded_params
         if unloaded_params:
-            raise RuntimeError(
-                "Some weights are not initialized from checkpoints: "
-                f"{unloaded_params}")
+            logger.warning(
+                "Some weights are not initialized from checkpoints: %s",
+                unloaded_params)
diff --git a/vllm/model_executor/models/paligemma.py b/vllm/model_executor/models/paligemma.py
index 2af48b6b..fe91611c 100644
--- a/vllm/model_executor/models/paligemma.py
+++ b/vllm/model_executor/models/paligemma.py
@@ -342,6 +342,6 @@ class PaliGemmaForConditionalGeneration(nn.Module, SupportsVision):
 
         unloaded_params = params_dict.keys() - loaded_params
         if unloaded_params:
-            raise RuntimeError(
-                "Some weights are not initialized from checkpoints: "
-                f"{unloaded_params}")
+            logger.warning(
+                "Some weights are not initialized from checkpoints: %s",
+                unloaded_params)