diff --git a/MM-LLMs-survey.md b/MM-LLMs-survey.md
new file mode 100644
index 0000000..122f5b6
--- /dev/null
+++ b/MM-LLMs-survey.md
@@ -0,0 +1,47 @@
+# MM-LLMs: Recent Advances in MultiModal Large Language Models
+
+## Overview
+[MM-LLMs: Recent Advances in MultiModal Large Language Models](https://arxiv.org/abs/2401.13601) (January 2024) represents a crucial milestone in understanding the current state and future directions of multimodal AI systems.
+
+## Key Significance
+This survey introduces a novel unified framework for analyzing MM-LLMs, particularly valuable for A2A (AI-to-AI) applications due to its systematic breakdown of integration approaches between perception and language components. The paper's significance stems from its comprehensive analysis of architectural patterns, training methodologies, and real-world deployment considerations.
+
+## Technical Implementation
+
+### Core Architecture Patterns
+```python
+class MMArchitecture:
+    def __init__(self, architecture_type="early_fusion"):
+        self.architecture_type = architecture_type
+        self.visual_encoder = VisualProcessor()
+        self.text_encoder = TextProcessor()
+        self.fusion_module = ModalityFusion()
+        
+    def process_multimodal_input(self, image, text):
+        visual_features = self.visual_encoder(image)
+        text_features = self.text_encoder(text)
+        
+        if self.architecture_type == "early_fusion":
+            return self.fusion_module.early_fusion(
+                visual_features, 
+                text_features
+            )
+        elif self.architecture_type == "late_fusion":
+            return self.fusion_module.late_fusion(
+                visual_features, 
+                text_features
+            )
+Key Takeaways for A2A Development
+Architectural Insights: The survey identifies two primary integration paradigms - perception-and-language and perception-language integration, crucial for designing efficient A2A systems.
+Training Strategies: Details emerging techniques for efficient training of MM-LLMs, including knowledge distillation and parameter-efficient fine-tuning.
+Evaluation Framework: Provides comprehensive metrics and benchmarks for assessing MM-LLM performance in A2A contexts.
+Implementation Considerations
+Focus on modular architecture design for flexible integration
+Consider computational efficiency in cross-modal fusion
+Implement robust evaluation metrics as described in Section 4
+Related Resources
+Visual-Language Models
+Multimodal Benchmarks
+Training Strategies
+Tags
+#multimodal #llm #survey #vision-language #a2a-systems