Fix bitrot (#326)

Co-authored-by: Daniel Holanda <[email protected]>
Signed-off-by: Jeremy Fowers <[email protected]>

Author: jeremyfowers

setup.py

@@ -56,7 +56,8 @@
         "zstandard",
         "matplotlib",
         "tabulate",
-        # huggingface-hub==0.31.0 introduces a new transfer protocol that was causing us issues 
+        "onnxconverter-common",
+        # huggingface-hub==0.31.0 introduces a new transfer protocol that was causing us issues
         "huggingface-hub==0.30.2",
         # Conditional dependencies for ONNXRuntime backends
         "onnxruntime >=1.10.1,<1.22.0;platform_system=='Linux' and extra != 'llm-oga-cuda'",
@@ -65,8 +66,8 @@
     ],
     extras_require={
         "llm": [
-            "torch>=2.0.0",
-            "transformers",
+            "torch>=2.6.0",
+            "transformers<=4.51.3",
             "accelerate",
             "py-cpuinfo",
             "sentencepiece",
@@ -76,23 +77,20 @@
             "human-eval-windows==1.0.4",
             "fastapi",
             "uvicorn[standard]",
-            "openai>=1.66.0",
+            "openai>=1.81.0",
             "lm-eval[api]",
         ],
         "llm-oga-cpu": [
             "onnxruntime-genai==0.6.0",
-            "torch>=2.0.0,<2.4",
             "turnkeyml[llm]",
         ],
         "llm-oga-igpu": [
             "onnxruntime-genai-directml==0.6.0",
-            "torch>=2.0.0,<2.4",
             "transformers<4.45.0",
             "turnkeyml[llm]",
         ],
         "llm-oga-cuda": [
             "onnxruntime-genai-cuda==0.6.0",
-            "torch>=2.0.0,<2.4",
             "transformers<4.45.0",
             "turnkeyml[llm]",
         ],

src/lemonade/tools/server/serve.py

@@ -762,6 +762,7 @@ async def generate():
                 created_event = ResponseCreatedEvent(
                     response=response,
                     type="response.created",
+                    sequence_number=0,
                 )
                 yield f"data: {created_event.model_dump_json()}\n\n".encode("utf-8")
 
@@ -776,6 +777,7 @@ async def generate():
                         item_id="0 ",
                         output_index=0,
                         type="response.output_text.delta",
+                        sequence_number=0,
                     )
                     full_response += token
 
@@ -810,6 +812,7 @@ async def generate():
                 completed_event = ResponseCompletedEvent(
                     response=response,
                     type="response.completed",
+                    sequence_number=0,
                 )
                 yield f"data: {completed_event.model_dump_json()}\n\n".encode("utf-8")
 

src/turnkeyml/version.py

@@ -1 +1 @@
-__version__ = "6.2.4"
+__version__ = "6.2.5"