llmir 0.0.4__tar.gz → 0.0.6__tar.gz

{llmir-0.0.4 → llmir-0.0.6}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: llmir
-Version: 0.0.4
+Version: 0.0.6
 Summary: Core message and tool IR for LLM pipelines
 Author: Mathis Siebert
 Requires-Python: >=3.11

{llmir-0.0.4 → llmir-0.0.6}/llmir/__init__.py

@@ -1,10 +1,10 @@
-from .chunks import AIChunk, AIChunkText, AIChunkFile, AIChunkImageURL, AIChunkToolCall
+from .chunks import AIChunks, AIChunkText, AIChunkFile, AIChunkImageURL, AIChunkToolCall
 from .roles import AIRoles
 from .messages import AIMessages, AIMessage, AIMessageToolResponse
 from .tools import Tool
 
 __all__ = [
-    "AIChunk",
+    "AIChunks",
     "AIChunkText",
     "AIChunkFile",
     "AIChunkImageURL",

llmir-0.0.6/llmir/adapter/__init__.py

@@ -0,0 +1,13 @@
+from .openai import to_openai, OpenAIMessage, OpenAIMessageToolResponse, OpenAIContents, OpenAITextContent, OpenAIImageURLContent, OpenAIImageURLURL, OpenAIToolCallContent, OpenAIToolCallFunction
+
+__all__ = [
+    "to_openai",
+    "OpenAIMessage",
+    "OpenAIMessageToolResponse",
+    "OpenAIContents",
+    "OpenAITextContent",
+    "OpenAIImageURLContent",
+    "OpenAIImageURLURL",
+    "OpenAIToolCallContent",
+    "OpenAIToolCallFunction",
+]

llmir-0.0.6/llmir/adapter/openai.py

@@ -0,0 +1,136 @@
+from typing import TypedDict, Literal
+
+from ..messages import AIMessages, AIMessageToolResponse
+from ..chunks import AIChunks, AIChunkText, AIChunkImageURL, AIChunkFile, AIChunkToolCall
+import base64
+import json
+
+
+class OpenAITextContent(TypedDict):
+    type: Literal["text"]
+    text: str
+
+class OpenAIImageURLURL(TypedDict):
+    url: str
+
+class OpenAIImageURLContent(TypedDict):
+    type: Literal["image_url"]
+    image_url: OpenAIImageURLURL
+
+
+class OpenAIToolCallFunction(TypedDict):
+    name: str
+    arguments: str
+
+class OpenAIToolCallContent(TypedDict):
+    id: str
+    type: Literal["function"]
+    function: OpenAIToolCallFunction
+
+    
+
+OpenAIContents = OpenAITextContent | OpenAIImageURLContent | OpenAIToolCallContent
+
+
+class OpenAIMessage(TypedDict):
+    role: str
+    content: list[OpenAIContents]
+
+class OpenAIMessageToolResponse(TypedDict):
+    role: Literal["tool"]
+    tool_call_id: str
+    name: str
+    content: str
+
+
+OpenAIMessages = OpenAIMessage | OpenAIMessageToolResponse
+
+
+
+def to_openai(messages: list[AIMessages]) -> list[OpenAIMessages]:
+    
+
+    result: list[OpenAIMessages] = []
+    for message in messages:
+        role = message.role.value
+        if isinstance(message, AIMessageToolResponse):
+            assert(role == "tool")
+            text: str = ""
+            media_chunks: list[AIChunkFile | AIChunkImageURL] = []
+            for chunk in message.chunks:
+                if isinstance(chunk, AIChunkText):
+                    text += chunk.text
+                elif isinstance(chunk, AIChunkImageURL) or isinstance(chunk, AIChunkFile):
+                    media_chunks.append(chunk)
+                else:
+                    raise Exception(f"Invalid chunk type for Tool Response: {chunk.type}")
+            result.append(
+                OpenAIMessageToolResponse(
+                    role=role,
+                    tool_call_id=message.id,
+                    name=message.name,
+                    content=text,
+                )
+            )
+            if media_chunks:
+                result.append(
+                    OpenAIMessage(
+                        role="user", # Hacky, but what else to circumvent API limitations in a broadly compatible way?
+                        content=[
+                            chunk_to_openai(chunk) for chunk in media_chunks
+                        ]
+                    )
+                )
+        else:
+            result.append(OpenAIMessage(
+                role= role,
+                content= [
+                    chunk_to_openai(chunk) for chunk in message.chunks
+                ]
+            ))
+    return result
+
+
+def chunk_to_openai(chunk: AIChunks) -> OpenAIContents:
+
+    match chunk:
+        case AIChunkText():
+            return OpenAITextContent(
+                type="text",
+                text=chunk.text,
+            )
+        case AIChunkImageURL():
+            return OpenAIImageURLContent(
+                type="image_url",
+                image_url={
+                    "url": chunk.url,
+                }
+            )
+        case AIChunkFile():
+            if chunk.mimetype.startswith("image/"):
+                base64_data = base64.b64encode(chunk.bytes).decode('utf-8')
+                return OpenAIImageURLContent(
+                    type= "image_url",
+                    image_url= {
+                        "url": f"data:{chunk.mimetype};base64,{base64_data}",
+                    }
+                )
+            elif chunk.mimetype == "text/plain":
+                text = chunk.bytes.decode(encoding="utf-8")
+                return OpenAITextContent(
+                    type="text",
+                    text=text
+                )
+            else:
+                raise ValueError(f"Unsupported file type for OpenAI: {chunk.mimetype}")
+        case AIChunkToolCall():
+            return OpenAIToolCallContent(
+                id=chunk.id,
+                type="function",
+                function=OpenAIToolCallFunction(
+                    name=chunk.name,
+                    arguments=json.dumps(chunk.arguments)
+                )
+            )
+        case _:
+            raise ValueError(f"Unsupported chunk type: {type(chunk)}")

{llmir-0.0.4 → llmir-0.0.6}/llmir/chunks.py

@@ -23,4 +23,4 @@ class AIChunkToolCall(RichReprMixin, BaseModel):
     arguments: dict[str, object]
 
 
-AIChunk = Union[AIChunkText, AIChunkFile, AIChunkImageURL, AIChunkToolCall]
+AIChunks = Union[AIChunkText, AIChunkFile, AIChunkImageURL, AIChunkToolCall]

{llmir-0.0.4 → llmir-0.0.6}/llmir/messages.py

@@ -1,17 +1,17 @@
 from pydantic import BaseModel, Field
 from typing import Literal
-from .chunks import AIChunk
+from .chunks import AIChunks
 from .roles import AIRoles
 
 class AIMessage(BaseModel):
 
     role: Literal[AIRoles.USER, AIRoles.MODEL, AIRoles.SYSTEM]
-    chunks: list[AIChunk] = Field(default_factory=list[AIChunk])
+    chunks: list[AIChunks] = Field(default_factory=list[AIChunks])
 
 class AIMessageToolResponse(BaseModel):
 
     role: Literal[AIRoles.TOOL] = AIRoles.TOOL
-    chunks: list[AIChunk] = Field(default_factory=list[AIChunk])
+    chunks: list[AIChunks] = Field(default_factory=list[AIChunks])
     id: str
     name: str
 

{llmir-0.0.4 → llmir-0.0.6}/llmir.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: llmir
-Version: 0.0.4
+Version: 0.0.6
 Summary: Core message and tool IR for LLM pipelines
 Author: Mathis Siebert
 Requires-Python: >=3.11

{llmir-0.0.4 → llmir-0.0.6}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "llmir"
-version = "0.0.4"
+version = "0.0.6"
 description = "Core message and tool IR for LLM pipelines"
 authors = [{ name="Mathis Siebert" }]
 requires-python = ">=3.11"

llmir-0.0.4/llmir/adapter/__init__.py

@@ -1,10 +0,0 @@
-from .openai import to_openai, OpenAIMessage, OpenAIMessageToolResponse, OpenAIContent, OpenAITextContent, OpenAIImageURLContent
-
-__all__ = [
-    "to_openai",
-    "OpenAIMessage",
-    "OpenAIMessageToolResponse",
-    "OpenAIContent",
-    "OpenAITextContent",
-    "OpenAIImageURLContent",
-]

llmir-0.0.4/llmir/adapter/openai.py

@@ -1,89 +0,0 @@
-from typing import TypedDict, Literal, Union
-
-from ..messages import AIMessages, AIMessageToolResponse
-from ..chunks import AIChunk, AIChunkText, AIChunkImageURL, AIChunkFile
-import base64
-
-
-class OpenAITextContent(TypedDict):
-    type: Literal["text"]
-    text: str
-
-
-class OpenAIImageURLContent(TypedDict):
-    type: Literal["image_url"]
-    image_url: dict[str, str]
-
-OpenAIContent = Union[OpenAITextContent, OpenAIImageURLContent]
-
-
-class OpenAIMessage(TypedDict):
-    role: str
-    content: list[OpenAIContent]
-
-class OpenAIMessageToolResponse(OpenAIMessage):
-    tool_call_id: str
-    name: str
-
-
-def to_openai(messages: list[AIMessages]) -> list[OpenAIMessage]:
-    
-
-    result: list[OpenAIMessage] = []
-    for message in messages:
-        role = message.role.value
-        if isinstance(message, AIMessageToolResponse):
-            result.append(
-                OpenAIMessageToolResponse(
-                    role=role,
-                    tool_call_id=message.id,
-                    name=message.name,
-                    content=[
-                        chunk_to_openai(chunk) for chunk in message.chunks
-                    ]
-                )
-            )
-        else:
-            result.append(OpenAIMessage(
-                role= role,
-                content= [
-                    chunk_to_openai(chunk) for chunk in message.chunks
-                ]
-            ))
-    return result
-
-
-def chunk_to_openai(chunk: AIChunk) -> OpenAIContent:
-
-    match chunk:
-        case AIChunkText():
-            return OpenAITextContent(
-                type="text",
-                text=chunk.text,
-            )
-        case AIChunkImageURL():
-            return OpenAIImageURLContent(
-                type="image_url",
-                image_url={
-                    "url": chunk.url,
-                }
-            )
-        case AIChunkFile():
-            if chunk.mimetype.startswith("image/"):
-                base64_data = base64.b64encode(chunk.bytes).decode('utf-8')
-                return OpenAIImageURLContent(
-                    type= "image_url",
-                    image_url= {
-                        "url": f"data:{chunk.mimetype};base64,{base64_data}",
-                    }
-                )
-            elif chunk.mimetype == ("text/plain"):
-                text = chunk.bytes.decode(encoding="utf-8")
-                return OpenAITextContent(
-                    type="text",
-                    text=text
-                )
-            else:
-                raise ValueError(f"Unsupported file type for OpenAI: {chunk.mimetype}")
-        case _:
-            raise ValueError(f"Unsupported chunk type: {type(chunk)}")