everything working again after cleanup

vibe_bot/llama_wrapper.py

@@ -6,9 +6,11 @@ Allows custom endpoints for each of the above supported functions.
 
 from __future__ import annotations
 
+import json
 from typing import TYPE_CHECKING, cast
 
 import openai
+import requests
 
 if TYPE_CHECKING:
     from io import BufferedReader, BytesIO
@@ -54,8 +56,12 @@ def chat_completion(
         model=model,
         messages=messages,
         max_tokens=max_tokens,
+        timeout=60.0,
     )
 
+    if not response.choices:
+        return ""
+
     content = response.choices[0].message.content
     if content:
         return content.strip()
@@ -101,8 +107,12 @@ def chat_completion_with_history(
         messages=messages,
         max_tokens=max_tokens,
         seed=-1,
+        timeout=60.0,
     )
 
+    if not response.choices:
+        return ""
+
     content = response.choices[0].message.content
     if content:
         return content.strip()
@@ -148,8 +158,12 @@ def chat_completion_instruct(
         messages=messages,
         max_tokens=max_tokens,
         seed=-1,
+        timeout=60.0,
     )
 
+    if not response.choices:
+        return ""
+
     content = response.choices[0].message.content
     if content:
         return content.strip()
@@ -158,8 +172,10 @@ def chat_completion_instruct(
 
 def image_generation(
     prompt: str,
+    *,
     openai_url: str,
     openai_api_key: str,
+    model: str = "gen",
     n: int = 1,
 ) -> str:
     """Generate an image using the given prompt.
@@ -168,19 +184,28 @@ def image_generation(
         prompt: The image generation prompt.
         openai_url: The OpenAI-compatible API URL.
         openai_api_key: The API key for authentication.
+        model: The model to use for image generation.
         n: Number of images to generate.
 
     Returns:
         The base64 encoded image data. Decode and write to a file.
 
     """
-    client = openai.OpenAI(base_url=openai_url, api_key=openai_api_key)
-    response = client.images.generate(
-        prompt=prompt,
-        n=n,
-        size="1024x1024",
-        model="gen",
+    client = openai.OpenAI(
+        base_url=openai_url,
+        api_key=openai_api_key,
+        max_retries=0,
     )
+    try:
+        response = client.images.generate(
+            prompt=prompt,
+            n=n,
+            size="1024x1024",
+            model=model,
+            timeout=120.0,
+        )
+    except openai.APIConnectionError:
+        return ""
     if response.data:
         return response.data[0].b64_json or ""
     return ""
@@ -189,8 +214,10 @@ def image_generation(
 def image_edit(
     image: BufferedReader | BytesIO | list[BufferedReader] | list[BytesIO],
     prompt: str,
+    *,
     openai_url: str,
     openai_api_key: str,
+    model: str = "edit",
     n: int = 1,
 ) -> str:
     """Edit an existing image using a prompt.
@@ -200,6 +227,7 @@ def image_edit(
         prompt: The edit instruction.
         openai_url: The OpenAI-compatible API URL.
         openai_api_key: The API key for authentication.
+        model: The model to use for image editing.
         n: Number of edited images to generate.
 
     Returns:
@@ -212,7 +240,7 @@ def image_edit(
         prompt=prompt,
         n=n,
         size="1024x1024",
-        model="edit",
+        model=model,
     )
     if response.data:
         return response.data[0].b64_json or ""
@@ -228,6 +256,9 @@ def embedding(
 ) -> list[float]:
     """Generate an embedding vector for the given text.
 
+    Uses a raw HTTP request to avoid the OpenAI SDK injecting
+    unsupported parameters like encoding_format.
+
     Args:
         text: The text to embed.
         openai_url: The OpenAI-compatible API URL.
@@ -238,17 +269,26 @@ def embedding(
         The embedding vector as a list of floats, or an empty list on failure.
 
     """
-    client = openai.OpenAI(base_url=openai_url, api_key=openai_api_key)
-    response = client.embeddings.create(
-        input=[text],
-        model=model,
-        encoding_format="float",
-    )
-    if response:
-        data = response.data
-        raw_data = data[0].embedding
-        # The result could be an array of floats or a single float.
-        if not isinstance(raw_data, float):
-            return list(raw_data)
-        return [raw_data]
-    return []
+    url = f"{openai_url.rstrip('/')}/embeddings"
+    headers = {
+        "Authorization": f"Bearer {openai_api_key}",
+        "Content-Type": "application/json",
+    }
+    payload = {"model": model, "input": [text]}
+
+    try:
+        resp = requests.post(url, headers=headers, json=payload, timeout=30)
+        resp.raise_for_status()
+    except requests.RequestException:
+        return []
+
+    data = resp.json()
+    if not data.get("data"):
+        return []
+
+    raw = data["data"][0].get("embedding")
+    if isinstance(raw, str):
+        raw = json.loads(raw)
+    if not isinstance(raw, list):
+        raw = list(raw)
+    return raw