Implement prompt caching, speculative decoding config, parallel slots, and dynamic pricing metrics

backend/services/token_stats.py

@@ -27,9 +27,11 @@ def get_stats() -> dict:
                 data["prompt_tokens"] = 0
             if "completion_tokens" not in data:
                 data["completion_tokens"] = 0
+            if "models" not in data:
+                data["models"] = {}
             return data
     except Exception:
-        return {"prompt_tokens": 0, "completion_tokens": 0}
+        return {"prompt_tokens": 0, "completion_tokens": 0, "models": {}}
 
 def save_stats(stats: dict):
     try:
@@ -39,8 +41,16 @@ def save_stats(stats: dict):
     except Exception:
         pass
 
-def increment_tokens(prompt: int, completion: int):
+def increment_tokens(prompt: int, completion: int, model: str = None):
     stats = get_stats()
     stats["prompt_tokens"] += prompt
     stats["completion_tokens"] += completion
+    if model:
+        model = model.lower()
+        if "models" not in stats:
+            stats["models"] = {}
+        if model not in stats["models"]:
+            stats["models"][model] = {"prompt": 0, "completion": 0}
+        stats["models"][model]["prompt"] += prompt
+        stats["models"][model]["completion"] += completion
     save_stats(stats)