inferencesh 0.2.13__tar.gz → 0.2.15__tar.gz

{inferencesh-0.2.13/src/inferencesh.egg-info → inferencesh-0.2.15}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: inferencesh
-Version: 0.2.13
+Version: 0.2.15
 Summary: inference.sh Python SDK
 Author: Inference Shell Inc.
 Author-email: "Inference Shell Inc." <hello@inference.sh>

{inferencesh-0.2.13 → inferencesh-0.2.15}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "inferencesh"
-version = "0.2.13"
+version = "0.2.15"
 description = "inference.sh Python SDK"
 authors = [
     {name = "Inference Shell Inc.", email = "hello@inference.sh"},

{inferencesh-0.2.13 → inferencesh-0.2.15}/src/inferencesh/models/llm.py

@@ -5,6 +5,7 @@ from queue import Queue
 from threading import Thread
 import time
 from contextlib import contextmanager
+import base64
 
 from .base import BaseAppInput, BaseAppOutput
 from .file import File
@@ -136,55 +137,51 @@ def timing_context():
     finally:
         pass
 
+def image_to_base64_data_uri(file_path):
+    with open(file_path, "rb") as img_file:
+        base64_data = base64.b64encode(img_file.read()).decode('utf-8')
+        return f"data:image/png;base64,{base64_data}"
 
 def build_messages(
     input_data: LLMInput,
     transform_user_message: Optional[Callable[[str], str]] = None
 ) -> List[Dict[str, Any]]:
     """Build messages for LLaMA.cpp chat completion.
-    
-    Args:
-        input_data: The input data
-        transform_user_message: Optional function to transform user message text before building messages
-    """
-    messages = [
-        {
-            "role": "system",
-            "content": [{"type": "text", "text": input_data.system_prompt}],
-        }
-    ]
 
-    # Add context messages
-    for msg in input_data.context:
-        message_content = []
-        text = msg.text
-        if transform_user_message and msg.role == ContextMessageRole.USER:
-            text = transform_user_message(text)
+    If any message includes image content, builds OpenAI-style multipart format.
+    Otherwise, uses plain string-only format.
+    """
+    def render_message(msg: ContextMessage, allow_multipart: bool) -> str | List[dict]:
+        parts = []
+        text = transform_user_message(msg.text) if transform_user_message and msg.role == ContextMessageRole.USER else msg.text
         if text:
-            message_content.append({"type": "text", "text": text})
-        if hasattr(msg, 'image') and msg.image:
+            parts.append({"type": "text", "text": text})
+        if msg.image:
             if msg.image.path:
-                message_content.append({"type": "image_url", "image_url": {"url": msg.image.path}})
+                image_data_uri = image_to_base64_data_uri(msg.image.path)
+                parts.append({"type": "image_url", "image_url": {"url": image_data_uri}})
             elif msg.image.uri:
-                message_content.append({"type": "image_url", "image_url": {"url": msg.image.uri}})
+                parts.append({"type": "image_url", "image_url": {"url": msg.image.uri}})
+        if allow_multipart:
+            return parts
+        if len(parts) == 1 and parts[0]["type"] == "text":
+            return parts[0]["text"]
+        raise ValueError("Image content requires multipart support")
+
+    multipart = any(m.image for m in input_data.context) or input_data.image is not None
+    messages = [{"role": "system", "content": input_data.system_prompt}]
+
+    for msg in input_data.context:
         messages.append({
             "role": msg.role,
-            "content": message_content
+            "content": render_message(msg, allow_multipart=multipart)
         })
 
-    # Add user message
-    user_content = []
-    text = input_data.text
-    if transform_user_message:
-        text = transform_user_message(text)
-    if text:
-        user_content.append({"type": "text", "text": text})
-    if hasattr(input_data, 'image') and input_data.image:
-        if input_data.image.path:
-            user_content.append({"type": "image_url", "image_url": {"url": input_data.image.path}})
-        elif input_data.image.uri:
-            user_content.append({"type": "image_url", "image_url": {"url": input_data.image.uri}})
-    messages.append({"role": "user", "content": user_content})
+    user_msg = ContextMessage(role=ContextMessageRole.USER, text=input_data.text, image=input_data.image)
+    messages.append({
+        "role": "user",
+        "content": render_message(user_msg, allow_multipart=multipart)
+    })
 
     return messages
 

{inferencesh-0.2.13 → inferencesh-0.2.15/src/inferencesh.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: inferencesh
-Version: 0.2.13
+Version: 0.2.15
 Summary: inference.sh Python SDK
 Author: Inference Shell Inc.
 Author-email: "Inference Shell Inc." <hello@inference.sh>