Add detailed model descriptions for Mistral 7B and XGLM 7.5B in app and settings

app.py

@@ -35,6 +35,7 @@ from src.analytics.chat_evaluator import ChatEvaluator
 if not HF_TOKEN:
     raise ValueError("HUGGINGFACE_TOKEN not found in environment variables")
 
+# Enhanced model details for UI
 # Enhanced model details for UI
 MODEL_DETAILS = {
     "llama-7b": {
@@ -78,8 +79,96 @@ MODEL_DETAILS = {
             "Structured legal text generation"
         ],
         "documentation": "https://huggingface.co/HuggingFaceH4/zephyr-7b-beta"
+    },
+    "mistral-7b": {
+        "full_name": "Mistral 7B Instruct v0.2",
+        "capabilities": [
+            "Strong multilingual support",
+            "Superior instruction following ability",
+            "Fast inference speed",
+            "Excellent reasoning capabilities",
+            "Free for commercial use"
+        ],
+        "limitations": [
+            "May have limited knowledge of specialized legal terminology",
+            "Less exposure to legal domain than specialized models",
+            "Knowledge cutoff before latest legal developments"
+        ],
+        "use_cases": [
+            "Multilingual legal assistance",
+            "Cross-border legal questions",
+            "Clear explanations of complex legal topics",
+            "Serving international clients in their native language"
+        ],
+        "documentation": "https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.2"
+    },
+    "xglm-7.5b": {
+        "full_name": "Meta XGLM 7.5B",
+        "capabilities": [
+            "Specialized for multilingual generation",
+            "Support for 30+ languages",
+            "Strong cross-lingual transfer abilities",
+            "Consistent performance across diverse languages"
+        ],
+        "limitations": [
+            "Less instruction-tuned than dedicated chat models",
+            "May require more specific prompting",
+            "Not specifically optimized for legal domain",
+            "Slightly larger model requiring more GPU memory"
+        ],
+        "use_cases": [
+            "International legal assistance in native languages",
+            "Complex multilingual documentation",
+            "Serving clients from diverse linguistic backgrounds",
+            "Translation and summarization of legal concepts across languages"
+        ],
+        "documentation": "https://huggingface.co/facebook/xglm-7.5B"
     }
 }
+# MODEL_DETAILS = {
+#     "llama-7b": {
+#         "full_name": "Meta Llama 2 7B Chat",
+#         "capabilities": [
+#             "Multilingual support ",
+#             "Good performance on legal texts",
+#             "Free model with open license",
+#             "Can run on computers with 16GB+ RAM"
+#         ],
+#         "limitations": [
+#             "Limited knowledge of specific legal terminology",
+#             "May provide incorrect answers to complex legal questions",
+#             "Knowledge is limited to training data"
+#         ],
+#         "use_cases": [
+#             "Legal document analysis",
+#             "Answering general legal questions",
+#             "Searching through legal knowledge base",
+#             "Assistance in document drafting"
+#         ],
+#         "documentation": "https://huggingface.co/meta-llama/Llama-2-7b-chat-hf"
+#     },
+#     "zephyr-7b": {
+#         "full_name": "HuggingFaceH4 Zephyr 7B Beta",
+#         "capabilities": [
+#             "High performance on instruction-following tasks",
+#             "Good response accuracy",
+#             "Advanced reasoning capabilities",
+#             "Excellent text generation quality"
+#         ],
+#         "limitations": [
+#             "May require paid API for usage",
+#             "Limited support for languages other than English",
+#             "Less optimization for legal topics compared to specialized models"
+#         ],
+#         "use_cases": [
+#             "Complex legal reasoning",
+#             "Case analysis",
+#             "Legal research",
+#             "Structured legal text generation"
+#         ],
+#         "documentation": "https://huggingface.co/HuggingFaceH4/zephyr-7b-beta"
+#     }
+# }
 
 # Path for user preferences file
 USER_PREFERENCES_PATH = os.path.join(os.path.dirname(__file__), "user_preferences.json")

config/settings.py

@@ -64,6 +64,50 @@ MODELS = {
                 "target_modules": ["q_proj", "v_proj", "k_proj", "o_proj"]
             }
         }
+    },
+    "mistral-7b": {
+        "id": "mistralai/Mistral-7B-Instruct-v0.2",
+        "name": "Mistral 7B Instruct",
+        "description": "Mistral's 7B instruction-tuned model with better multilingual support",
+        "type": "base",
+        "parameters": {
+            "max_length": 2048,
+            "temperature": 0.7,
+            "top_p": 0.9,
+            "repetition_penalty": 1.1,
+        },
+        "training": {
+            "base_model_path": "mistralai/Mistral-7B-Instruct-v0.2",
+            "fine_tuned_path": os.path.join(TRAINING_OUTPUT_DIR, "mistral-7b-instruct-tuned"),
+            "lora_config": {
+                "r": 16,
+                "lora_alpha": 32,
+                "lora_dropout": 0.05,
+                "target_modules": ["q_proj", "v_proj", "k_proj", "o_proj"]
+            }
+        }
+    },
+    "xglm-7.5b": {
+        "id": "facebook/xglm-7.5B",
+        "name": "XGLM 7.5B",
+        "description": "Meta's multilingual model designed for cross-lingual generation",
+        "type": "base",
+        "parameters": {
+            "max_length": 2048,
+            "temperature": 0.7,
+            "top_p": 0.9,
+            "repetition_penalty": 1.1,
+        },
+        "training": {
+            "base_model_path": "facebook/xglm-7.5B",
+            "fine_tuned_path": os.path.join(TRAINING_OUTPUT_DIR, "xglm-7.5b-tuned"),
+            "lora_config": {
+                "r": 16,
+                "lora_alpha": 32,
+                "lora_dropout": 0.05,
+                "target_modules": ["q_proj", "v_proj", "k_proj", "o_proj"]
+            }
+        }
     }
 }
 
@@ -75,4 +119,4 @@ ACTIVE_MODEL = MODELS[DEFAULT_MODEL]
 EMBEDDING_MODEL = "intfloat/multilingual-e5-large"
 
 # Request settings
-USER_AGENT = "Status-Law-Assistant/1.0"
+USER_AGENT = "Status-Law-Assistant/1.0"