PyGAAMAS: Add Qwen3 for evaluationg prefence alignment

README.md

@@ -126,12 +126,16 @@ each corresponding to one of the four preferences:
 
 Table below evaluates the ability of the models to align with different preferences.
 - When generating **strategies**, the models align perfectly with preferences, except for 
-- <tt>DeepSeek-R1<tt> and <tt>Mixtral:8x7b</tt> which do not generate valid code.
+  - <tt>DeepSeek-R1<tt> and <tt>Mixtral:8x7b</tt> which do not generate valid code
+  - <tt>Qwen3</tt>, which fails to adopt egoistic or altruistic strategies but adheres 
+  to utilitarian and egalitarian preferences.
 - When generating **actions**, 
   - <tt>GPT-4.5</tt> aligns well with preferences but struggles with **utilitarianism**.
   - <tt>Llama3</tt> aligns well with **egoistic** and **altruistic** preferences but shows lower adherence to **utilitarian** and **egalitarian** choices.
   - <tt>Mistral-Small</tt> aligns better with **altruistic** preferences and performs moderately on **utilitarianism** but struggles with **egoistic** and **egalitarian** preferences.
   - <tt>DeepSeek-R1</tt> primarily aligns with **utilitarianism** but has low accuracy in other preferences.
+  - <tt>Qwen3</tt> strongly aligns with utilitarian preferences and moderately with altruistic ones (0.80), 
+  - but fails to exhibit egoistic behavior and shows weak alignment with egalitarianism.
 While a larger LLM typically aligns better with preferences, a model like <tt>Mixtral-8x7B</tt> may occasionally 
 underperform compared to its smaller counterpart, Mistral-Small due to their architectural complexity. 
 Mixture-of-Experts (MoE) models, like Mixtral, dynamically activate only a subset of their parameters. 
@@ -147,6 +151,7 @@ If the routing mechanism isn’t well-tuned, it might select less optimal expert
 | **<tt>Mistral-Small</tt>**   | **Strategy**   | 1.00         | 1.00           | 1.00            | 1.00            |
 | **<tt>DeepSeek-R1:7b</tt>**  | **Strategy**   | 1.00         | 1.00           | 1.00            | 1.00            |
 | **<tt>DeepSeek-R1</tt>**     | **Strategy**   | -            | -              | -               | -               |
+| **<tt>Qwen3</tt>**           | **Strategy**   | 0.00         | 0.00           | 1.00            | 1.00            |
 | **<tt>GPT-4.5<tt>**          | **Actions**    | 1.00         | 1.00           | 0.50            | 1.00            |
 | **<tt>Llama3.3:latest</tt>** | **Actions**    | 1.00         | 1.00           | 0.43            | 0.96            |
 | **<tt>Llama3</tt>**          | **Actions**    | 1.00         | 0.90           | 0.40            | 0.73            |
@@ -154,6 +159,7 @@ If the routing mechanism isn’t well-tuned, it might select less optimal expert
 | **<tt>Mistral-Small</tt>**   | **Actions**    | 0.40         | 0.94           | 0.76            | 0.16            |
 | **<tt>DeepSeek-R1:7b</tt>**  | **Actions**    | 0.46         | 0.56           | 0.66            | 0.90            |
 | **<tt>DeepSeek-R1</tt>**     | **Actions**    | 0.06         | 0.20           | 0.76            | 0.03            |
+| **<tt>Qwen3</tt>**           | **Actions**    | 0.00         | 0.80           | 0.93            | 0.36            |
 
 Errors in action selection may stem from either arithmetic miscalculations  
 (e.g., the model incorrectly assumes that $500 + 100 > 400 + 300$) or  

figures/dictator/dictator_setup_accuracy.csv

 Model,ALTRUISTIC,EGALITARIAN,SELFISH,UTILITARIAN
-deepseek-r1,0.2,0.03333333333333333,0.06666666666666667,0.7666666666666667
+deepseek-r1,0.15,0.06666666666666667,0.15,0.8
 deepseek-r1:7b,0.5666666666666667,0.9,0.4666666666666667,0.6666666666666666
 gpt-4.5-preview-2025-02-27,1.0,1.0,1.0,0.5
 llama3,0.9,0.7333333333333333,1.0,0.4
 llama3.3:latest,1.0,0.9666666666666667,1.0,0.43333333333333335
-mistral-small,0.9411764705882353,0.16666666666666666,0.4,0.7666666666666667
+mistral-small,0.95,0.16666666666666666,0.4,0.7
 mixtral:8x7b,0.0,1.0,0.0,0.3
+qwen3,0.0,0.8,0.9333333333333333,0.36666666666666664

src/dictator/dictator_setup.py

@@ -274,6 +274,25 @@ class DictatorSetup:
                     "motivations": "preference dictates how the resources are distributed",
                     "is_consistent": True,
                 }
+        elif model == "qwen3":
+            if self.preference == Welfare.EGALITARIAN:
+                return {
+                    'my_share': 500,
+                    'other_share': 500,
+                    'lost': 0
+                }
+            elif self.preference == Welfare.UTILITARIAN:
+                return {
+                    'my_share': 0,
+                    'other_share': 1000,
+                    'lost': 0
+                }
+            else:
+                return {
+                    'my_share': 0,
+                    'other_share': 0,
+                    'lost': 1000
+                }
         else:
             raise ValueError("Invalid preference type")
             return {"error": "Preference strategy not defined"}

src/dictator/dictator_setup_experiments.py

@@ -65,7 +65,7 @@ class DictatorSetupExperiment:
 
 # Running the experiment
 if __name__ == "__main__":
-    models = [ "llama3", "mistral-small", "deepseek-r1"] # "gpt-4.5-preview-2025-02-27", "mixtral:8x7b", "llama3.3:latest", "deepseek-r1:7b",
+    models = [ "qwen3" ] # "gpt-4.5-preview-2025-02-27", "mixtral:8x7b", "llama3.3:latest", "deepseek-r1:7b", "llama3", "mistral-small", "deepseek-r1"
     temperature = 0.7
     iterations = 30
     output_file = '../../data/dictator/dictator_setup.csv'

src/dictator/dictator_temperature_experiments.py

@@ -63,7 +63,7 @@ class DictatorTemperatureExperiment:
 
 # Running the experiment
 if __name__ == "__main__":
-    models = ["gpt-4.5-preview-2025-02-27", "llama3", "mistral-small", "deepseek-r1"]
+    models = ["qwen3"]#"gpt-4.5-preview-2025-02-27", "llama3", "mistral-small", "deepseek-r1"
     temperatures = [round(i * 0.1, 1) for i in range(7,11)]  # Temperatures from 0.0 to 1.0 with a step of 0.1
     amount = 100
     iterations = 30