Update coding_priority_pipe.py

coding_priority_pipe.py

@@ -2,33 +2,36 @@
 title: Coding Fallback Pipe
 author: nerdur
 author_url: https://nerdur-webui.hf.space
-version: 2.0
-description: Kruži kroz NVIDIA NIM besplatne coding modele s pravim fallback-om pri 429/500 grešci.
+version: 3.0
+description: |
+  Fallback redoslijed za kodiranje:
+  1. NVIDIA NIM (4 modela)
+  2. Gemini 2.5 Flash
+  3. Groq (besplatni coding modeli)
+  4. Cerebras
+  5. SambaNova
 """
 
 from pydantic import BaseModel
-from typing import Optional, List, Union, Generator, Iterator
+from typing import Optional, List, Union, Iterator
 import requests
-import json
+import os
 
 
 class Pipe:
     class Valves(BaseModel):
-        # NVIDIA NIM Free Endpoint coding modeli — prioritet od najboljeg ka rezervnom
-        model_priority: List[str] = [
-            "mistral-ai/devstral-2-123b-instruct-2512",
-            "deepseek-ai/deepseek-v3.2",
-            "qwen/qwen3-coder-480b-a35b-instruct",
-            "zhipuai/glm-4.7",
-        ]
         nvidia_api_key: str = ""
-        nvidia_base_url: str = "https://integrate.api.nvidia.com/v1"
+        google_api_key: str = ""
+        groq_api_key: str = ""
+        cerebras_api_key: str = ""
+        sambanova_api_key: str = ""
         enabled: bool = True
+        request_timeout: int = 45
 
     def __init__(self):
         self.type = "pipe"
         self.id = "coding_fallback"
-        self.name = "Smart Coding Router (NVIDIA NIM)"
+        self.name = "🔀 Smart Coding Router (Multi-Provider)"
         self.valves = self.Valves()
 
     def pipes(self):
@@ -38,65 +41,114 @@ class Pipe:
         if not self.valves.enabled:
             return "Router je isključen."
 
-        import os
-        api_key = self.valves.nvidia_api_key or os.getenv("NVIDIA_API_KEY") or os.getenv("NVIDIA_ID_API_KEY", "")
-
-        if not api_key:
-            return "❌ NVIDIA_API_KEY nije postavljen."
-
         messages = body.get("messages", [])
         stream = body.get("stream", False)
-
-        headers = {
-            "Authorization": f"Bearer {api_key}",
-            "Content-Type": "application/json",
-        }
+        max_tokens = body.get("max_tokens", 4096)
+        temperature = body.get("temperature", 0.2)
+
+        # --- Redoslijed providera ---
+        # Svaki entry: (provider_name, base_url, model_id, api_key)
+        nvidia_key = self.valves.nvidia_api_key or os.getenv("NVIDIA_ID_API_KEY") or os.getenv("NVIDIA_API_KEY", "")
+        google_key = self.valves.google_api_key or os.getenv("GOOGLE_API_KEY", "")
+        groq_key   = self.valves.groq_api_key   or os.getenv("GROQ_API_KEY", "")
+        cerebras_key = self.valves.cerebras_api_key or os.getenv("CEREBRAS_API_KEY", "")
+        sambanova_key = self.valves.sambanova_api_key or os.getenv("SAMBANOVA_API_KEY", "")
+
+        providers = []
+
+        # 1. NVIDIA NIM — 4 coding modela
+        if nvidia_key:
+            for model in [
+                "mistral-ai/devstral-2-123b-instruct-2512",
+                "deepseek-ai/deepseek-v3.2",
+                "qwen/qwen3-coder-480b-a35b-instruct",
+                "zhipuai/glm-4.7",
+            ]:
+                providers.append(("NVIDIA", "https://integrate.api.nvidia.com/v1", model, nvidia_key))
+
+        # 2. Gemini 2.5 Flash
+        if google_key:
+            providers.append(("Gemini", "https://generativelanguage.googleapis.com/v1beta/openai", "gemini-2.5-flash", google_key))
+
+        # 3. Groq — besplatni coding modeli
+        if groq_key:
+            for model in [
+                "moonshotai/kimi-k2-instruct",
+                "deepseek-r1-distill-llama-70b",
+                "llama-3.3-70b-versatile",
+            ]:
+                providers.append(("Groq", "https://api.groq.com/openai/v1", model, groq_key))
+
+        # 4. Cerebras
+        if cerebras_key:
+            providers.append(("Cerebras", "https://api.cerebras.ai/v1", "llama-3.3-70b", cerebras_key))
+
+        # 5. SambaNova
+        if sambanova_key:
+            providers.append(("SambaNova", "https://api.sambanova.ai/v1", "Meta-Llama-3.3-70B-Instruct", sambanova_key))
+
+        if not providers:
+            return "❌ Nijedan API ključ nije postavljen. Provjeri Space Secrets."
 
         last_error = None
-        for model_id in self.valves.model_priority:
+        for provider_name, base_url, model_id, api_key in providers:
             payload = {
                 "model": model_id,
                 "messages": messages,
                 "stream": stream,
-                "max_tokens": body.get("max_tokens", 4096),
-                "temperature": body.get("temperature", 0.2),
+                "max_tokens": max_tokens,
+                "temperature": temperature,
+            }
+            headers = {
+                "Authorization": f"Bearer {api_key}",
+                "Content-Type": "application/json",
             }
 
             try:
-                print(f"Coding Router: Pokušavam {model_id}...")
+                print(f"🔀 Coding Router [{provider_name}]: Pokušavam {model_id}...")
                 response = requests.post(
-                    f"{self.valves.nvidia_base_url}/chat/completions",
+                    f"{base_url}/chat/completions",
                     headers=headers,
                     json=payload,
                     stream=stream,
-                    timeout=60,
+                    timeout=self.valves.request_timeout,
                 )
 
-                # Preskači na sljedeći model ako je rate limit ili server greška
-                if response.status_code in (429, 500, 502, 503):
-                    print(f"⚠️  {model_id} vratio {response.status_code}, prelazim na sljedeći...")
-                    last_error = f"{model_id}: HTTP {response.status_code}"
+                if response.status_code in (429, 500, 502, 503, 529):
+                    print(f"⚠️  [{provider_name}] {model_id} → HTTP {response.status_code}, prelazim...")
+                    last_error = f"[{provider_name}] {model_id}: HTTP {response.status_code}"
+                    continue
+
+                if response.status_code == 401:
+                    print(f"⚠️  [{provider_name}] Neispravan API ključ, preskačem providera...")
+                    last_error = f"[{provider_name}]: 401 Unauthorized"
+                    # Preskoči sve modele ovog providera
+                    current_provider = provider_name
+                    providers_iter = [(p, b, m, k) for p, b, m, k in providers if p != current_provider]
                     continue
 
                 response.raise_for_status()
 
                 if stream:
-                    def generate():
-                        for line in response.iter_lines():
+                    def generate(resp=response, pname=provider_name, mid=model_id):
+                        print(f"✅ [{pname}] Streaming: {mid}")
+                        for line in resp.iter_lines():
                             if line:
                                 yield line.decode("utf-8") + "\n"
                     return generate()
                 else:
                     data = response.json()
-                    return data["choices"][0]["message"]["content"]
+                    content = data["choices"][0]["message"]["content"]
+                    print(f"✅ [{provider_name}] Odgovor od: {model_id}")
+                    return content
 
             except requests.exceptions.Timeout:
-                print(f"⚠️  {model_id} timeout, prelazim...")
-                last_error = f"{model_id}: timeout"
+                print(f"⚠️  [{provider_name}] {model_id} timeout ({self.valves.request_timeout}s), prelazim...")
+                last_error = f"[{provider_name}] {model_id}: timeout"
                 continue
             except Exception as e:
-                print(f"⚠️  {model_id} greška: {e}")
+                print(f"⚠️  [{provider_name}] {model_id} greška: {e}")
                 last_error = str(e)
                 continue
 
-        return f"❌ Svi modeli neuspješni. Posljednja greška: {last_error}"
+        return f"❌ Svi provideri neuspješni.\nPosljednja greška: {last_error}"