Refactor settings.py: Update Mistral 7B Instruct model configuration for improved performance and documentation

config/settings.py

@@ -175,9 +175,9 @@ MODELS = {
         }
     },
     "mistral-7b": {
-        "id": "TheBloke/Mistral-7B-Instruct-v0.2-GGUF",
-        "name": "Mistral 7B Instruct GGUF",
-        "description": "Optimized GGUF version of Mistral-7B-Instruct for CPU inference",
+        "id": "mistralai/Mistral-7B-Instruct-v0.2",  # Меняем на стандартную версию
+        "name": "Mistral 7B Instruct",
+        "description": "High-performance instruction-tuned 7B model with strong multilingual capabilities",
         "type": "base",
         "parameters": {
             "max_length": 2048,
@@ -186,8 +186,8 @@ MODELS = {
             "repetition_penalty": 1.1,
         },
         "training": {
-            "base_model_path": "TheBloke/Mistral-7B-Instruct-v0.2-GGUF",
-            "fine_tuned_path": os.path.join(TRAINING_OUTPUT_DIR, "mistral-7b-instruct-gguf-tuned"),
+            "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,
@@ -196,27 +196,26 @@ MODELS = {
             }
         },
         "details": {
-            "full_name": "Mistral 7B Instruct v0.2 GGUF",
+            "full_name": "Mistral 7B Instruct v0.2",
             "capabilities": [
                 "Strong multilingual support",
-                "Optimized for CPU inference",
-                "Lower memory requirements",
-                "Fast inference speed",
-                "Excellent reasoning capabilities",
+                "Excellent instruction following",
+                "Advanced reasoning capabilities",
+                "High-quality text generation",
                 "Free for commercial use"
             ],
             "limitations": [
-                "May have slightly lower quality compared to full model",
-                "Less exposure to legal domain than specialized models",
-                "Knowledge cutoff before latest legal developments"
+                "Requires more RAM than GGUF version",
+                "May need GPU for optimal performance",
+                "Knowledge cutoff before latest developments"
             ],
             "use_cases": [
-                "Multilingual legal assistance",
-                "Local CPU inference",
-                "Resource-constrained environments",
-                "Cross-border legal questions"
+                "Complex legal analysis",
+                "Multilingual consultation",
+                "Document processing",
+                "General assistance"
             ],
-            "documentation": "https://huggingface.co/TheBloke/Mistral-7B-Instruct-v0.2-GGUF"
+            "documentation": "https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.2"
         }
     },
     "mixtral-8x7b": {
@@ -306,53 +305,7 @@ MODELS = {
             "documentation": "https://huggingface.co/openchat/openchat-3.5"
         }
     },
-    "gemma-7b": {
-        "id": "google/gemma-7b-it",
-        "name": "Gemma 7B Instruct",
-        "description": "Google's instruction-tuned 7B model with strong multilingual capabilities",
-        "type": "base",
-        "parameters": {
-            "max_length": 2048,
-            "temperature": 0.7,
-            "top_p": 0.9,
-            "repetition_penalty": 1.1,
-        },
-        "training": {
-            "base_model_path": "google/gemma-7b-it",
-            "fine_tuned_path": os.path.join(TRAINING_OUTPUT_DIR, "gemma-7b-instruct-tuned"),
-            "lora_config": {
-                "r": 16,
-                "lora_alpha": 32,
-                "lora_dropout": 0.05,
-                "target_modules": ["q_proj", "v_proj", "k_proj", "o_proj"]
-            }
-        },
-        "details": {
-            "full_name": "Google Gemma 7B Instruct",
-            "capabilities": [
-                "Strong performance on par with other 7B models",
-                "Excellent instruction following",
-                "Good multilingual capabilities",
-                "Efficient resource usage",
-                "Free for commercial use",
-                "Official Google release"
-            ],
-            "limitations": [
-                "Relatively new model",
-                "May need specific prompting for best results",
-                "Limited community testing compared to older models"
-            ],
-            "use_cases": [
-                "General consultation",
-                "Document analysis",
-                "Complex reasoning tasks",
-                "Educational purposes",
-                "Commercial applications"
-            ],
-            "documentation": "https://huggingface.co/google/gemma-7b-it"
-        }
-    }
-}
+ }
 
 # Update MODELS configuration
 for model in MODELS.values():