inferencesh 0.2.31__py3-none-any.whl → 0.2.33__py3-none-any.whl

inferencesh/models/llm.py

@@ -229,7 +229,6 @@ def build_messages(
             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}] if input_data.system_prompt is not None and input_data.system_prompt != "" else []
 
     def merge_messages(messages: List[ContextMessage]) -> ContextMessage:
@@ -238,7 +237,17 @@ def build_messages(
         image = images[0] if images else None # TODO: handle multiple images
         return ContextMessage(role=messages[0].role, text=text, image=image)
 
-    user_msg = ContextMessage(role=ContextMessageRole.USER, text=input_data.text, image=input_data.image)
+    user_input_text = ""
+    if hasattr(input_data, "text"):
+        user_input_text = transform_user_message(input_data.text) if transform_user_message else input_data.text
+        
+    user_input_image = None
+    multipart = any(m.image for m in input_data.context)
+    if hasattr(input_data, "image"):
+        user_input_image = input_data.image
+        multipart = multipart or input_data.image is not None
+
+    user_msg = ContextMessage(role=ContextMessageRole.USER, text=user_input_text, image=user_input_image)
 
     input_data.context.append(user_msg)
 
@@ -585,7 +594,7 @@ def stream_generate(
     output_cls: type[BaseLLMOutput] = LLMOutput,
 ) -> Generator[BaseLLMOutput, None, None]:
     """Stream generate from LLaMA.cpp model with timing and usage tracking."""
-    
+        
     # Create queues for communication between threads
     response_queue = Queue()
     error_queue = Queue()
@@ -617,8 +626,6 @@ def stream_generate(
             completion = model.create_chat_completion(**completion_kwargs)
             
             for chunk in completion:
-                if verbose:
-                    print(chunk)
                 response_queue.put(("chunk", chunk))
                 # Update keep-alive timestamp
                 keep_alive_queue.put(("alive", time.time()))
@@ -627,7 +634,9 @@ def stream_generate(
             response_queue.put(("done", None))
             
         except Exception as e:
-            error_queue.put(e)
+            # Preserve the full exception with traceback
+            import sys
+            error_queue.put((e, sys.exc_info()[2]))
             response_queue.put(("error", str(e)))
     
     with timing_context() as timing:
@@ -657,14 +666,22 @@ def stream_generate(
                 raise RuntimeError(f"Model failed to initialize within {init_timeout} seconds")
             
             while True:
-                # Check for errors
+                # Check for errors - now with proper exception chaining
                 if not error_queue.empty():
-                    raise error_queue.get()
+                    exc, tb = error_queue.get()
+                    if isinstance(exc, Exception):
+                        raise exc.with_traceback(tb)
+                    else:
+                        raise RuntimeError(f"Unknown error in worker thread: {exc}")
                 
                 # Check keep-alive
-                while not keep_alive_queue.empty():
-                    _, timestamp = keep_alive_queue.get_nowait()
-                    last_activity = timestamp
+                try:
+                    while not keep_alive_queue.empty():
+                        _, timestamp = keep_alive_queue.get_nowait()
+                        last_activity = timestamp
+                except Empty:
+                    # Ignore empty queue - this is expected
+                    pass
                 
                 # Check for timeout
                 if time.time() - last_activity > chunk_timeout:
@@ -677,12 +694,17 @@ def stream_generate(
                     continue
                 
                 if msg_type == "error":
+                    # If we get an error message but no exception in error_queue,
+                    # create a new error
                     raise RuntimeError(f"Generation error: {data}")
                 elif msg_type == "done":
                     break
                 
                 chunk = data
                 
+                if verbose:
+                    print(chunk)
+                
                 # Mark first token time
                 if not timing.first_token_time:
                     timing.mark_first_token()
@@ -700,12 +722,17 @@ def stream_generate(
                     break
             
             # Wait for generation thread to finish
-            generation_thread.join(timeout=5.0)  # Increased timeout to 5 seconds
             if generation_thread.is_alive():
-                # Thread didn't finish - this shouldn't happen normally
-                # but we handle it gracefully
-                raise RuntimeError("Generation thread failed to finish")
+                generation_thread.join(timeout=5.0)  # Increased timeout to 5 seconds
+                if generation_thread.is_alive():
+                    # Thread didn't finish - this shouldn't happen normally
+                    raise RuntimeError("Generation thread failed to finish")
                     
         except Exception as e:
-            # Ensure any error is properly propagated
-            raise e 
+            # Check if there's a thread error we should chain with
+            if not error_queue.empty():
+                thread_exc, thread_tb = error_queue.get()
+                if isinstance(thread_exc, Exception):
+                    raise e from thread_exc
+            # If no thread error, raise the original exception
+            raise 

{inferencesh-0.2.31.dist-info → inferencesh-0.2.33.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: inferencesh
-Version: 0.2.31
+Version: 0.2.33
 Summary: inference.sh Python SDK
 Author: Inference Shell Inc.
 Author-email: "Inference Shell Inc." <hello@inference.sh>

{inferencesh-0.2.31.dist-info → inferencesh-0.2.33.dist-info}/RECORD

@@ -2,13 +2,13 @@ inferencesh/__init__.py,sha256=WdADtOhfa3HDOunoE9HLFCTFlXRykYstBIH1FpyWvj8,613
 inferencesh/models/__init__.py,sha256=FDwcdtT6c4hbRitymjmN-hZMlQa8RbKSftkZZyjtUXA,536
 inferencesh/models/base.py,sha256=4gZQRi8J7y9U6PrGD9pRIehd1MJVJAqGakPQDs2AKFM,3251
 inferencesh/models/file.py,sha256=5xnpypcRahM1YcEjj64rv9g2gTimxrZb41YT4r440hU,7393
-inferencesh/models/llm.py,sha256=Yj7BGtAlBJBgvEawZQeGb4AB2WvHWY5DG8wEMIXiPoo,27047
+inferencesh/models/llm.py,sha256=KTy9XAXQS_2p4HThyofSl4Nu8F87sA4IPJqmP233VA8,28304
 inferencesh/utils/__init__.py,sha256=-xiD6uo2XzcrPAWFb_fUbaimmnW4KFKc-8IvBzaxNd4,148
 inferencesh/utils/download.py,sha256=7n5twvoNYDcFnKJyefImaj2YfzRI7vddQw4usZbj38c,1521
 inferencesh/utils/storage.py,sha256=E4J8emd4eFKdmdDgAqzz3TpaaDd3n0l8gYlMHuY8yIU,519
-inferencesh-0.2.31.dist-info/licenses/LICENSE,sha256=OsgqEWIh2el_QMj0y8O1A5Q5Dl-dxqqYbFE6fszuR4s,1086
-inferencesh-0.2.31.dist-info/METADATA,sha256=cXGDMenDbFRfRfCugNmRlTzoFBw8kC-UuVEKFLMWqZI,2757
-inferencesh-0.2.31.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
-inferencesh-0.2.31.dist-info/entry_points.txt,sha256=6IC-fyozAqW3ljsMLGCXxJ0_ui2Jb-2fLHtoH1RTnEE,45
-inferencesh-0.2.31.dist-info/top_level.txt,sha256=TSMHg3T1ThMl1HGAWmzBClwOYH1ump5neof9BfHIwaA,12
-inferencesh-0.2.31.dist-info/RECORD,,
+inferencesh-0.2.33.dist-info/licenses/LICENSE,sha256=OsgqEWIh2el_QMj0y8O1A5Q5Dl-dxqqYbFE6fszuR4s,1086
+inferencesh-0.2.33.dist-info/METADATA,sha256=Thb63zVHWjix6fCnfOEn4FFwnWnZ_XXCUWGrwpnkeTk,2757
+inferencesh-0.2.33.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
+inferencesh-0.2.33.dist-info/entry_points.txt,sha256=6IC-fyozAqW3ljsMLGCXxJ0_ui2Jb-2fLHtoH1RTnEE,45
+inferencesh-0.2.33.dist-info/top_level.txt,sha256=TSMHg3T1ThMl1HGAWmzBClwOYH1ump5neof9BfHIwaA,12
+inferencesh-0.2.33.dist-info/RECORD,,