fix(proxy): add safe_json_loads to prevent 500 errors on streaming tool call decoding

scripts/ollama-proxy.py

@@ -277,6 +277,30 @@ def _ollama_tools_to_openai(tools):
         })
     return out
 
+def safe_json_loads(val, default=None):
+    """Safely decode JSON string, handling extra data or syntax errors gracefully."""
+    if default is None:
+        default = {}
+    if not val:
+        return default
+    if isinstance(val, dict):
+        return val
+    s = str(val).strip()
+    if not s:
+        return default
+    try:
+        return json.loads(s)
+    except Exception:
+        pass
+    try:
+        decoder = json.JSONDecoder()
+        obj, _ = decoder.raw_decode(s)
+        if isinstance(obj, dict):
+            return obj
+    except Exception:
+        pass
+    return {"raw": s}
+
 def _openai_tool_calls_to_ollama(tool_calls):
     """OpenAI tool_calls -> ollama message.tool_calls."""
     if not tool_calls:
@@ -284,10 +308,7 @@ def _openai_tool_calls_to_ollama(tool_calls):
     out = []
     for tc in tool_calls:
         fn = tc.get("function", {})
-        try:
+        args = safe_json_loads(fn.get("arguments", "{}"))
-            args = json.loads(fn.get("arguments", "{}"))
-        except json.JSONDecodeError:
-            args = {}
         out.append({
             "function": {
                 "name": fn.get("name", ""),
@@ -674,15 +695,16 @@ def ollama_chat(payload):
 
         om = {"role": "assistant", "content": ""}
         if tc_parts:
-            om["tool_calls"] = [
+            tc_list = []
-                {
+            for e in tc_parts.values():
+                args = safe_json_loads(e.get("args"))
+                tc_list.append({
                     "function": {
-                        "name": e["name"],
+                        "name": e.get("name", ""),
-                        "arguments": json.loads(e["args"]) if e["args"] else {},
+                        "arguments": args,
                     }
-                }
+                })
-                for e in tc_parts.values()
+            om["tool_calls"] = tc_list
-            ]
 
         yield json.dumps({
             "model": MODEL_NAME,
@@ -953,10 +975,7 @@ def _openai_tool_calls_to_anthropic(tool_calls):
     blocks = []
     for tc in tool_calls:
         fn = tc.get("function", {})
-        try:
+        args = safe_json_loads(fn.get("arguments", "{}"))
-            args = json.loads(fn.get("arguments", "{}"))
-        except json.JSONDecodeError:
-            args = {}
         blocks.append({
             "type": "tool_use",
             "id": tc.get("id") or "toolu_" + uuid.uuid4().hex[:24],
@@ -1115,10 +1134,7 @@ def anthropic_messages(payload):
             if finish:
                 # close any open tool_use block(s) with accumulated input
                 for tid, entry in tc_parts.items():
-                    try:
+                    args = safe_json_loads(entry.get("args"))
-                        args = json.loads(entry["args"]) if entry["args"] else {}
-                    except json.JSONDecodeError:
-                        args = {}
                     yield sse("content_block_delta", {
                         "type": "content_block_delta",
                         "index": entry["block_index"],