Upload 11 files

coding_priority_pipe.py

@@ -2,49 +2,101 @@
 title: Coding Fallback Pipe
 author: nerdur
 author_url: https://nerdur-webui.hf.space
-version: 1.0
-description: Automatski kruži kroz besplatne provajdere za kodiranje (Gemini -> Groq -> Cerebras -> Sambanova).
+version: 2.0
+description: Kruži kroz NVIDIA NIM besplatne coding modele s pravim fallback-om pri 429/500 grešci.
 """
 
 from pydantic import BaseModel
-from typing import Optional, List, Union, Generator
+from typing import Optional, List, Union, Generator, Iterator
 import requests
+import json
 
 
 class Pipe:
     class Valves(BaseModel):
-        # Prioritet modela za kodiranje
+        # NVIDIA NIM Free Endpoint coding modeli — prioritet od najboljeg ka rezervnom
         model_priority: List[str] = [
-            "google_generative_ai.gemini-2.0-flash",
-            "groq.llama-3.3-70b-versatile",
-            "cerebras.llama3.1-70b",
-            "sambanova.meta-llama-3.1-405b-instruct"
+            "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"
         enabled: bool = True
 
     def __init__(self):
         self.type = "pipe"
         self.id = "coding_fallback"
-        self.name = "Smart Coding Router"
+        self.name = "Smart Coding Router (NVIDIA NIM)"
         self.valves = self.Valves()
 
-    def pipe(self, body: dict, __user__: Optional[dict] = None) -> Union[str, Generator, dict]:
+    def pipes(self):
+        return [{"id": self.id, "name": self.name}]
+
+    def pipe(self, body: dict, __user__: Optional[dict] = None) -> Union[str, Iterator, dict]:
         if not self.valves.enabled:
-            return "Router is disabled."
-
-        # Ova funkcija u Open WebUI Pipe sistemu prosljeđuje zahtjev
-        # Prvi model iz liste koji je dostupan u sistemu će biti izabran
-        # Za pravu fallback logiku (ako API vrati 429), potrebna je kompleksnija 
-        # implementacija unutar samog 'pipe' poziva, ali za početak 
-        # ovo omogućava modelu da "zna" za ostale provajdere.
-        
-        # NAPOMENA: Za pravi fallback pri grešci, modelID se mora dinamički mijenjati
-        # u slučaju Exception-a tokom streaminga.
-        
-        selected_model = body.get("model")
-        if selected_model == f"{self.id}.{self.name}":
-             # Ako je izabran sam router, stavi prvi model iz prioriteta
-             body["model"] = self.valves.model_priority[0]
-        
-        print(f"Coding Router: Preusmjeravam na {body['model']}")
-        return body
+            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",
+        }
+
+        last_error = None
+        for model_id in self.valves.model_priority:
+            payload = {
+                "model": model_id,
+                "messages": messages,
+                "stream": stream,
+                "max_tokens": body.get("max_tokens", 4096),
+                "temperature": body.get("temperature", 0.2),
+            }
+
+            try:
+                print(f"Coding Router: Pokušavam {model_id}...")
+                response = requests.post(
+                    f"{self.valves.nvidia_base_url}/chat/completions",
+                    headers=headers,
+                    json=payload,
+                    stream=stream,
+                    timeout=60,
+                )
+
+                # 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}"
+                    continue
+
+                response.raise_for_status()
+
+                if stream:
+                    def generate():
+                        for line in response.iter_lines():
+                            if line:
+                                yield line.decode("utf-8") + "\n"
+                    return generate()
+                else:
+                    data = response.json()
+                    return data["choices"][0]["message"]["content"]
+
+            except requests.exceptions.Timeout:
+                print(f"⚠️  {model_id} timeout, prelazim...")
+                last_error = f"{model_id}: timeout"
+                continue
+            except Exception as e:
+                print(f"⚠️  {model_id} greška: {e}")
+                last_error = str(e)
+                continue
+
+        return f"❌ Svi modeli neuspješni. Posljednja greška: {last_error}"

setup_endpoints.sh

@@ -39,48 +39,59 @@ fi
 
 if [ -z "$TOKEN" ]; then echo "❌ Nije moguće dobiti token."; exit 0; fi
 
-# Pokreni instalaciju funkcija
+# Pokreni instalaciju funkcija u pozadini
 python3 /app/install_functions.py "$BASE_URL" "$TOKEN" &
 
 echo "✅ Ulogovan, filtriram provajdere s praznim ključevima..."
 
-# Python skripta za dinamičko generisanje JSON-a (sprečava 500 Error: KeyError 'id')
 python3 - <<PYEOF
 import requests, os, json
 
 base_url = "$BASE_URL"
 token = "$TOKEN"
 
+# NVIDIA: podržava i NVIDIA_API_KEY i NVIDIA_ID_API_KEY
+nvidia_key = os.getenv("NVIDIA_ID_API_KEY") or os.getenv("NVIDIA_API_KEY")
+
+# Cloudflare: sklopi URL samo ako postoji account ID
+cf_account = os.getenv("CLOUDFLARE_ACCOUNT_ID") or ""
+cf_url = f"https://api.cloudflare.com/client/v4/accounts/{cf_account}/ai/v1" if cf_account else None
+
 providers = [
-    ("https://api.openai.com/v1", os.getenv("OPENAI_API_KEY")),
-    ("https://api.anthropic.com/v1", os.getenv("ANTHROPIC_API_KEY")),
+    ("https://api.openai.com/v1",                          os.getenv("OPENAI_API_KEY")),
+    ("https://api.anthropic.com/v1",                       os.getenv("ANTHROPIC_API_KEY")),
     ("https://generativelanguage.googleapis.com/v1beta/openai", os.getenv("GOOGLE_API_KEY")),
-    ("https://models.inference.ai.azure.com", os.getenv("GITHUB_TOKEN")),
-    ("https://api.groq.com/openai/v1", os.getenv("GROQ_API_KEY")),
-    ("https://api.mistral.ai/v1", os.getenv("MISTRAL_API_KEY")),
-    ("https://api.deepseek.com/v1", os.getenv("DEEPSEEK_API_KEY")),
-    ("https://api.together.xyz/v1", os.getenv("TOGETHER_API_KEY")),
-    ("https://integrate.api.nvidia.com/v1", os.getenv("NVIDIA_ID_API_KEY") or os.getenv("NVIDIA_API_KEY")),
-    ("https://api.sambanova.ai/v1", os.getenv("SAMBANOVA_API_KEY")),
-    ("https://openrouter.ai/api/v1", os.getenv("OPENROUTER_API_KEY")),
-    ("https://api.cloudflare.com/client/v4/accounts/" + (os.getenv("CLOUDFLARE_ACCOUNT_ID") or "") + "/ai/v1", os.getenv("CLOUDFLARE_API_TOKEN")),
-    ("https://api.cerebras.ai/v1", os.getenv("CEREBRAS_API_KEY")),
-    ("https://api.fireworks.ai/inference/v1", os.getenv("FIREWORKS_API_KEY")),
-    ("https://api.x.ai/v1", os.getenv("XAI_API_KEY")),
-    ("https://router.huggingface.co/v1", os.getenv("HF_INFERENCE_TOKEN")),
-    ("https://api.cohere.ai/compatibility/v1", os.getenv("COHERE_API_KEY")),
-    ("https://gen.pollinations.ai/v1", "noapikey")
+    ("https://models.inference.ai.azure.com",              os.getenv("GITHUB_TOKEN")),
+    ("https://api.groq.com/openai/v1",                     os.getenv("GROQ_API_KEY")),
+    ("https://api.mistral.ai/v1",                          os.getenv("MISTRAL_API_KEY")),
+    ("https://api.deepseek.com/v1",                        os.getenv("DEEPSEEK_API_KEY")),
+    ("https://api.together.xyz/v1",                        os.getenv("TOGETHER_API_KEY")),
+    ("https://integrate.api.nvidia.com/v1",                nvidia_key),
+    ("https://api.sambanova.ai/v1",                        os.getenv("SAMBANOVA_API_KEY")),
+    ("https://openrouter.ai/api/v1",                       os.getenv("OPENROUTER_API_KEY")),
+    ("https://api.cerebras.ai/v1",                         os.getenv("CEREBRAS_API_KEY")),
+    ("https://api.fireworks.ai/inference/v1",              os.getenv("FIREWORKS_API_KEY")),
+    ("https://api.x.ai/v1",                                os.getenv("XAI_API_KEY")),
+    ("https://router.huggingface.co/v1",                   os.getenv("HF_INFERENCE_TOKEN")),
+    ("https://api.cohere.ai/compatibility/v1",             os.getenv("COHERE_API_KEY")),
+    ("https://gen.pollinations.ai/v1",                     "noapikey"),
 ]
 
+# Dodaj Cloudflare samo ako postoji account ID
+if cf_url:
+    providers.append((cf_url, os.getenv("CLOUDFLARE_API_TOKEN")))
+
 valid_urls = []
 valid_keys = []
 
 for url, key in providers:
-    # Preskači ako je URL nepotpun ili ključ prazan
-    if "/accounts//ai" in url: continue
-    if key and key.strip() and key not in ['""', "''"]:
-        valid_urls.append(url)
-        valid_keys.append(key)
+    if not url:
+        continue
+    if not key or not key.strip() or key in ['""', "''"]:
+        continue
+    valid_urls.append(url)
+    valid_keys.append(key)
+    print(f"  ✔ {url.split('/')[2]}")
 
 payload = {
     "ENABLE_OPENAI_API": True,