原始代码

_qwen_xinference_demo/opro/ollama_client.py

@@ -0,0 +1,52 @@
+import requests
+import re
+import config
+
+OLLAMA_URL = config.OLLAMA_GENERATE_URL
+TAGS_URL = config.OLLAMA_TAGS_URL
+MODEL_NAME = config.DEFAULT_CHAT_MODEL
+
+def call_qwen(prompt: str, temperature: float = 0.8, max_tokens: int = 512, model_name: str | None = None) -> str:
+    def _payload(m: str):
+        return {
+            "model": m,
+            "prompt": prompt,
+            "stream": False,
+            "options": {
+                "temperature": temperature,
+                "num_predict": max_tokens
+            }
+        }
+    primary = model_name or MODEL_NAME
+    try:
+        resp = requests.post(OLLAMA_URL, json=_payload(primary), timeout=60)
+        resp.raise_for_status()
+        data = resp.json()
+        return data.get("response", "") or data.get("text", "")
+    except requests.HTTPError as e:
+        # Try fallback to default when user-selected model fails
+        if model_name and model_name != MODEL_NAME:
+            try:
+                resp = requests.post(OLLAMA_URL, json=_payload(MODEL_NAME), timeout=60)
+                resp.raise_for_status()
+                data = resp.json()
+                return data.get("response", "") or data.get("text", "")
+            except Exception:
+                pass
+        raise
+
+def list_models() -> list[str]:
+    try:
+        r = requests.get(TAGS_URL, timeout=10)
+        r.raise_for_status()
+        data = r.json() or {}
+        items = data.get("models") or []
+        names = []
+        for m in items:
+            name = m.get("name") or m.get("model")
+            if name:
+                names.append(name)
+        names = [n for n in names if not re.search(r"embedding|rerank|reranker|bge", n, re.I)]
+        return names
+    except Exception:
+        return [MODEL_NAME]