Use Unsloth for QwQ inference

setup.py

@@ -10,12 +10,13 @@ setup(
         ],
     },
     install_requires=[
-        'torch>=2.0.0',
         'accelerate>=0.26.0',
         'aiohttp>=3.8.0',
-        'bitsandbytes>=0.41.0',
+        'bitsandbytes>=0.45',
         'dotenv-python>=0.0.1',
         'huggingface_hub>=0.16.0',
+        'ipykernel>=6.0.0', 
+        'ipywidgets>=8.0.0',
         'lxml>=4.9.0',
         'mistral-common>=1.0.0',
         'mistralai>=0.0.7',
@@ -23,7 +24,11 @@ setup(
         'psutil>=5.9.0',
         'python-dotenv>=1.0.0',
         'tiktoken>=0.4.0',
+        'torch>=2.0.0',
         'transformers>=4.30.0',
+        'trl>=0.7.8',
+        'unsloth>=2025.3',
+        'vllm==0.8.2',
         'xml_schema_validator @ file:///root/sia/lib/xml_schema_validator',
     ],
     python_requires='>=3.10',