lm-deluge 0.0.15__tar.gz → 0.0.17__tar.gz

{lm_deluge-0.0.15/src/lm_deluge.egg-info → lm_deluge-0.0.17}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: lm_deluge
-Version: 0.0.15
+Version: 0.0.17
 Summary: Python utility for using LLM API models.
 Author-email: Benjamin Anderson <ben@trytaylor.ai>
 Requires-Python: >=3.10
@@ -128,6 +128,30 @@ This just works. Images can be local images on disk, URLs, bytes, base64 data UR
 
 See a full multi-turn chat example in `examples/multiturn.md`.
 
+## Files
+
+For models that support file uploads (OpenAI, Anthropic, and Gemini), you can easily include PDF files and other documents:
+
+```python
+from lm_deluge import LLMClient, Conversation
+
+# Simple file upload
+client = LLMClient.basic("gpt-4.1-mini")
+conversation = Conversation.user(
+    "Please summarize this document", 
+    file="path/to/document.pdf"
+)
+resps = client.process_prompts_sync([conversation])
+
+# You can also create File objects for more control
+from lm_deluge import File
+file = File("path/to/report.pdf", filename="Q4_Report.pdf")
+conversation = Conversation.user("Analyze this financial report")
+conversation.messages[0].parts.append(file)
+```
+
+Files can be local paths, URLs, bytes, or base64 data URLs, just like images.
+
 ## Tool Use
 
 Define tools from Python functions and use them with any model:
@@ -187,6 +211,16 @@ for tool_call in resps[0].tool_calls:
     # this is dumb sorry will make it better
     tool_to_call = [x for x in tools if x.name == tool_call.name][0]
     tool_to_call.call(**tool_call.arguments) # in async code, use .acall()
+
+# or use the built-in agent loop to handle this automatically
+import asyncio
+
+async def main():
+    conv = Conversation.user("List the files in the current directory")
+    conv, resp = await client.run_agent_loop(conv, tools=tools)
+    print(resp.content.completion)
+
+asyncio.run(main())
 ```
 
 ### Prompt Caching (Anthropic)

{lm_deluge-0.0.15 → lm_deluge-0.0.17}/README.md

@@ -101,6 +101,30 @@ This just works. Images can be local images on disk, URLs, bytes, base64 data UR
 
 See a full multi-turn chat example in `examples/multiturn.md`.
 
+## Files
+
+For models that support file uploads (OpenAI, Anthropic, and Gemini), you can easily include PDF files and other documents:
+
+```python
+from lm_deluge import LLMClient, Conversation
+
+# Simple file upload
+client = LLMClient.basic("gpt-4.1-mini")
+conversation = Conversation.user(
+    "Please summarize this document", 
+    file="path/to/document.pdf"
+)
+resps = client.process_prompts_sync([conversation])
+
+# You can also create File objects for more control
+from lm_deluge import File
+file = File("path/to/report.pdf", filename="Q4_Report.pdf")
+conversation = Conversation.user("Analyze this financial report")
+conversation.messages[0].parts.append(file)
+```
+
+Files can be local paths, URLs, bytes, or base64 data URLs, just like images.
+
 ## Tool Use
 
 Define tools from Python functions and use them with any model:
@@ -160,6 +184,16 @@ for tool_call in resps[0].tool_calls:
     # this is dumb sorry will make it better
     tool_to_call = [x for x in tools if x.name == tool_call.name][0]
     tool_to_call.call(**tool_call.arguments) # in async code, use .acall()
+
+# or use the built-in agent loop to handle this automatically
+import asyncio
+
+async def main():
+    conv = Conversation.user("List the files in the current directory")
+    conv, resp = await client.run_agent_loop(conv, tools=tools)
+    print(resp.content.completion)
+
+asyncio.run(main())
 ```
 
 ### Prompt Caching (Anthropic)

{lm_deluge-0.0.15 → lm_deluge-0.0.17}/pyproject.toml

@@ -3,7 +3,7 @@ requires = ["setuptools", "wheel"]
 
 [project]
 name = "lm_deluge"
-version = "0.0.15"
+version = "0.0.17"
 authors = [{ name = "Benjamin Anderson", email = "ben@trytaylor.ai" }]
 description = "Python utility for using LLM API models."
 readme = "README.md"

lm_deluge-0.0.17/src/lm_deluge/api_requests/__init__.py

@@ -0,0 +1 @@
+

{lm_deluge-0.0.15 → lm_deluge-0.0.17}/src/lm_deluge/api_requests/anthropic.py

@@ -1,35 +1,39 @@
-from aiohttp import ClientResponse
 import json
 import os
-from typing import Callable
+
+from aiohttp import ClientResponse
 
 from lm_deluge.prompt import (
+    CachePattern,
     Conversation,
     Message,
     Text,
-    ToolCall,
     Thinking,
-    CachePattern,
+    ToolCall,
 )
-from lm_deluge.tool import Tool
+from lm_deluge.request_context import RequestContext
+from lm_deluge.tool import MCPServer, Tool
 from lm_deluge.usage import Usage
-from .base import APIRequestBase, APIResponse
 
-from ..tracker import StatusTracker
 from ..config import SamplingParams
 from ..models import APIModel
-from ..computer_use.anthropic_tools import get_anthropic_cu_tools
+from .base import APIRequestBase, APIResponse
+
+
+def _add_beta(headers: dict, beta: str):
+    if "anthropic-beta" in headers and headers["anthropic-beta"]:
+        if beta not in headers["anthropic-beta"]:
+            headers["anthropic-beta"] += f",{beta}"
+    else:
+        headers["anthropic-beta"] = beta
 
 
 def _build_anthropic_request(
     model: APIModel,
     prompt: Conversation,
-    tools: list[Tool] | None,
+    tools: list[Tool | dict | MCPServer] | None,
     sampling_params: SamplingParams,
     cache_pattern: CachePattern | None = None,
-    computer_use: bool = False,
-    display_width: int = 1024,
-    display_height: int = 768,
 ):
     system_message, messages = prompt.to_anthropic(cache_pattern=cache_pattern)
     request_header = {
@@ -38,10 +42,6 @@ def _build_anthropic_request(
         "content-type": "application/json",
     }
 
-    # Add beta header for Computer Use
-    if computer_use:
-        request_header["anthropic-beta"] = "computer-use-2025-01-24"
-
     request_json = {
         "model": model.name,
         "messages": messages,
@@ -69,89 +69,61 @@ def _build_anthropic_request(
             print("ignoring reasoning_effort for non-reasoning model")
     if system_message is not None:
         request_json["system"] = system_message
-    if tools or computer_use:
+    if tools:
+        mcp_servers = []
         tool_definitions = []
-        if tools:
-            tool_definitions.extend([tool.dump_for("anthropic") for tool in tools])
-        # Add Computer Use tools
-        if computer_use:
-            cu_tools = get_anthropic_cu_tools(
-                model=model.id,
-                display_width=display_width,  # todo: set from ComputerUseParams
-                display_height=display_height,
-            )
-            tool_definitions.extend(cu_tools)
+        for tool in tools:
+            if isinstance(tool, Tool):
+                tool_definitions.append(tool.dump_for("anthropic"))
+            elif isinstance(tool, dict):
+                tool_definitions.append(tool)
+                # add betas if needed
+                if tool["type"] in [
+                    "computer_20241022",
+                    "text_editor_20241022",
+                    "bash_20241022",
+                ]:
+                    _add_beta(request_header, "computer-use-2024-10-22")
+                elif tool["type"] == "computer_20250124":
+                    _add_beta(request_header, "computer-use-2025-01-24")
+                elif tool["type"] == "code_execution_20250522":
+                    _add_beta(request_header, "code-execution-2025-05-22")
+            elif isinstance(tool, MCPServer):
+                _add_beta(request_header, "mcp-client-2025-04-04")
+                mcp_servers.append(tool.for_anthropic())
 
         # Add cache control to last tool if tools_only caching is specified
         if cache_pattern == "tools_only" and tool_definitions:
             tool_definitions[-1]["cache_control"] = {"type": "ephemeral"}
 
         request_json["tools"] = tool_definitions
+        if len(mcp_servers) > 0:
+            request_json["mcp_servers"] = mcp_servers
 
     return request_json, request_header
 
 
 class AnthropicRequest(APIRequestBase):
-    def __init__(
-        self,
-        task_id: int,
-        # should always be 'role', 'content' keys.
-        # internal logic should handle translating to specific API format
-        model_name: str,  # must correspond to registry
-        prompt: Conversation,
-        attempts_left: int,
-        status_tracker: StatusTracker,
-        results_arr: list,
-        request_timeout: int = 30,
-        sampling_params: SamplingParams = SamplingParams(),
-        callback: Callable | None = None,
-        # for retries
-        all_model_names: list[str] | None = None,
-        all_sampling_params: list[SamplingParams] | None = None,
-        tools: list | None = None,
-        cache: CachePattern | None = None,
-        # Computer Use support
-        computer_use: bool = False,
-        display_width: int = 1024,
-        display_height: int = 768,
-    ):
-        super().__init__(
-            task_id=task_id,
-            model_name=model_name,
-            prompt=prompt,
-            attempts_left=attempts_left,
-            status_tracker=status_tracker,
-            results_arr=results_arr,
-            request_timeout=request_timeout,
-            sampling_params=sampling_params,
-            callback=callback,
-            all_model_names=all_model_names,
-            all_sampling_params=all_sampling_params,
-            tools=tools,
-            cache=cache,
-        )
-        self.computer_use = computer_use
-        self.display_width = display_width
-        self.display_height = display_height
-        self.model = APIModel.from_registry(model_name)
+    def __init__(self, context: RequestContext):
+        super().__init__(context=context)
+
+        self.model = APIModel.from_registry(self.context.model_name)
         self.url = f"{self.model.api_base}/messages"
 
         # Lock images as bytes if caching is enabled
-        if cache is not None:
-            prompt.lock_images_as_bytes()
+        if self.context.cache is not None:
+            self.context.prompt.lock_images_as_bytes()
 
         self.request_json, self.request_header = _build_anthropic_request(
             self.model,
-            prompt,
-            tools,
-            sampling_params,
-            cache,
-            computer_use,
-            display_width,
-            display_height,
+            self.context.prompt,
+            self.context.tools,
+            self.context.sampling_params,
+            self.context.cache,
         )
 
     async def handle_response(self, http_response: ClientResponse) -> APIResponse:
+        data = None
         is_error = False
         error_message = None
         thinking = None
@@ -160,6 +132,7 @@ class AnthropicRequest(APIRequestBase):
         status_code = http_response.status
         mimetype = http_response.headers.get("Content-Type", None)
         rate_limits = {}
+        assert self.context.status_tracker
         for header in [
             "anthropic-ratelimit-requests-limit",
             "anthropic-ratelimit-requests-remaining",
@@ -215,20 +188,21 @@ class AnthropicRequest(APIRequestBase):
                 or "overloaded" in error_message.lower()
             ):
                 error_message += " (Rate limit error, triggering cooldown.)"
-                self.status_tracker.rate_limit_exceeded()
+                self.context.status_tracker.rate_limit_exceeded()
             if "context length" in error_message:
                 error_message += " (Context length exceeded, set retries to 0.)"
-                self.attempts_left = 0
+                self.context.attempts_left = 0
 
         return APIResponse(
-            id=self.task_id,
+            id=self.context.task_id,
             status_code=status_code,
             is_error=is_error,
             error_message=error_message,
-            prompt=self.prompt,
+            prompt=self.context.prompt,
             content=content,
             thinking=thinking,
-            model_internal=self.model_name,
-            sampling_params=self.sampling_params,
+            model_internal=self.context.model_name,
+            sampling_params=self.context.sampling_params,
             usage=usage,
+            raw_response=data,
         )

lm_deluge-0.0.17/src/lm_deluge/api_requests/base.py

@@ -0,0 +1,120 @@
+import asyncio
+import traceback
+from abc import ABC, abstractmethod
+
+import aiohttp
+from aiohttp import ClientResponse
+
+from ..errors import raise_if_modal_exception
+from ..request_context import RequestContext
+from .response import APIResponse
+
+
+class APIRequestBase(ABC):
+    """
+    Class for handling API requests. All model/endpoint-specific logic should be
+    handled by overriding __init__ and implementing the handle_response method.
+    For call_api to work, the __init__ must handle setting:
+        - url
+        - request_header
+        - request_json
+    """
+
+    def __init__(
+        self,
+        context: RequestContext,
+    ):
+        # If context is provided, use it; otherwise construct one from individual parameters
+        self.context = context
+
+        # Everything is now accessed through self.context - no copying!
+        self.system_prompt = None
+        self.result = []  # list of APIResponse objects from each attempt
+
+        # these should be set in the __init__ of the subclass
+        self.url = None
+        self.request_header = None
+        self.request_json = None
+        self.region = None
+
+    def increment_pbar(self):
+        if self.context.status_tracker:
+            self.context.status_tracker.increment_pbar()
+
+    def call_callback(self):
+        if self.context.callback is not None:
+            # the APIResponse in self.result includes all the information
+            self.context.callback(self.result[-1], self.context.status_tracker)
+
+    def handle_success(self, data):
+        self.call_callback()
+        if self.context.status_tracker:
+            self.context.status_tracker.task_succeeded(self.context.task_id)
+
+    async def execute_once(self) -> APIResponse:
+        """Send the HTTP request once and return the parsed APIResponse."""
+        assert self.context.status_tracker
+        try:
+            self.context.status_tracker.total_requests += 1
+            timeout = aiohttp.ClientTimeout(total=self.context.request_timeout)
+            async with aiohttp.ClientSession(timeout=timeout) as session:
+                assert self.url is not None, "URL is not set"
+                async with session.post(
+                    url=self.url,
+                    headers=self.request_header,
+                    json=self.request_json,
+                ) as http_response:
+                    response: APIResponse = await self.handle_response(http_response)
+            return response
+
+        except asyncio.TimeoutError:
+            return APIResponse(
+                id=self.context.task_id,
+                model_internal=self.context.model_name,
+                prompt=self.context.prompt,
+                sampling_params=self.context.sampling_params,
+                status_code=None,
+                is_error=True,
+                error_message="Request timed out (terminated by client).",
+                content=None,
+                usage=None,
+            )
+
+        except Exception as e:
+            raise_if_modal_exception(e)
+            tb = traceback.format_exc()
+            print(tb)
+            return APIResponse(
+                id=self.context.task_id,
+                model_internal=self.context.model_name,
+                prompt=self.context.prompt,
+                sampling_params=self.context.sampling_params,
+                status_code=None,
+                is_error=True,
+                error_message=f"Unexpected {type(e).__name__}: {str(e) or 'No message.'}",
+                content=None,
+                usage=None,
+            )
+
+    @abstractmethod
+    async def handle_response(self, http_response: ClientResponse) -> APIResponse:
+        raise NotImplementedError
+
+
+def deduplicate_responses(results: list[APIRequestBase]) -> list[APIResponse]:
+    deduplicated = {}
+    for request in results:
+        if request.context.task_id not in deduplicated:
+            deduplicated[request.context.task_id] = request.result[-1]
+        else:
+            current_response: APIResponse = deduplicated[request.context.task_id]
+            # only replace if the current request has no completion and the new one does
+            if (
+                request.result[-1].completion is not None
+                and current_response.completion is None
+            ):
+                deduplicated[request.context.task_id] = request.result[-1]
+
+    output = [deduplicated[request.context.task_id] for request in results]
+
+    return output