lm-deluge 0.0.58__tar.gz → 0.0.60__tar.gz

{lm_deluge-0.0.58/src/lm_deluge.egg-info → lm_deluge-0.0.60}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: lm_deluge
-Version: 0.0.58
+Version: 0.0.60
 Summary: Python utility for using LLM API models.
 Author-email: Benjamin Anderson <ben@trytaylor.ai>
 Requires-Python: >=3.10

{lm_deluge-0.0.58 → lm_deluge-0.0.60}/pyproject.toml

@@ -3,7 +3,7 @@ requires = ["setuptools", "wheel"]
 
 [project]
 name = "lm_deluge"
-version = "0.0.58"
+version = "0.0.60"
 authors = [{ name = "Benjamin Anderson", email = "ben@trytaylor.ai" }]
 description = "Python utility for using LLM API models."
 readme = "README.md"

{lm_deluge-0.0.58 → lm_deluge-0.0.60}/src/lm_deluge/api_requests/base.py

@@ -1,4 +1,5 @@
 import asyncio
+import time
 import traceback
 from abc import ABC, abstractmethod
 
@@ -6,6 +7,7 @@ import aiohttp
 from aiohttp import ClientResponse
 
 from ..errors import raise_if_modal_exception
+from ..models.openai import OPENAI_MODELS
 from ..request_context import RequestContext
 from .response import APIResponse
 
@@ -82,15 +84,95 @@ class APIRequestBase(ABC):
         if self.context.status_tracker:
             self.context.status_tracker.task_succeeded(self.context.task_id)
 
+    async def _execute_once_background_mode(self) -> APIResponse:
+        """
+        ONLY for OpenAI responses API. Implement the
+        start -> poll -> result style of request.
+        """
+        assert self.context.status_tracker, "no status tracker"
+        start_time = time.time()
+        async with aiohttp.ClientSession() as session:
+            last_status: str | None = None
+
+            try:
+                self.context.status_tracker.total_requests += 1
+                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:
+                    # make sure we created the Response object
+                    http_response.raise_for_status()
+                    data = await http_response.json()
+                    response_id = data["id"]
+                    last_status = data["status"]
+
+                while True:
+                    if time.time() - start_time > self.context.request_timeout:
+                        # cancel the response
+                        async with session.post(
+                            url=f"{self.url}/{response_id}/cancel",
+                            headers=self.request_header,
+                        ) as http_response:
+                            http_response.raise_for_status()
+
+                        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,
+                        )
+                    # poll for the response
+                    await asyncio.sleep(5.0)
+                    async with session.get(
+                        url=f"{self.url}/{response_id}",
+                        headers=self.request_header,
+                    ) as http_response:
+                        http_response.raise_for_status()
+                        data = await http_response.json()
+
+                        if data["status"] != last_status:
+                            print(
+                                f"Background req {response_id} status updated to: {data['status']}"
+                            )
+                            last_status = data["status"]
+                        if last_status not in ["queued", "in_progress"]:
+                            return await self.handle_response(http_response)
+
+            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,
+                )
+
     async def execute_once(self) -> APIResponse:
         """Send the HTTP request once and return the parsed APIResponse."""
         await self.build_request()
         assert self.context.status_tracker
-        # try:
-        #     dumped = json.dumps(self.request_json)
-        # except Exception:
-        #     print("couldn't serialize request json")
-        #     print(self.request_json)
+
+        if (
+            self.context.background
+            and self.context.use_responses_api
+            and self.context.model_name in OPENAI_MODELS
+        ):
+            return await self._execute_once_background_mode()
+
         try:
             self.context.status_tracker.total_requests += 1
             timeout = aiohttp.ClientTimeout(total=self.context.request_timeout)

{lm_deluge-0.0.58 → lm_deluge-0.0.60}/src/lm_deluge/api_requests/bedrock.py

@@ -1,10 +1,11 @@
 import asyncio
 import json
 import os
-import warnings
 
 from aiohttp import ClientResponse
 
+from lm_deluge.warnings import maybe_warn
+
 try:
     from requests_aws4auth import AWS4Auth
 except ImportError:
@@ -187,9 +188,7 @@ async def _build_openai_bedrock_request(
     # Note: GPT-OSS on Bedrock doesn't support response_format parameter
     # Even though the model supports JSON, we can't use the response_format parameter
     if sampling_params.json_mode and model.supports_json:
-        warnings.warn(
-            f"JSON mode requested for {model.name} but response_format parameter not supported on Bedrock"
-        )
+        maybe_warn("WARN_JSON_MODE_UNSUPPORTED", model_name=model.name)
 
     if tools:
         request_tools = []

{lm_deluge-0.0.58 → lm_deluge-0.0.60}/src/lm_deluge/api_requests/gemini.py

@@ -1,11 +1,12 @@
 import json
 import os
-import warnings
 from typing import Any
+
 from aiohttp import ClientResponse
 
 from lm_deluge.request_context import RequestContext
 from lm_deluge.tool import Tool
+from lm_deluge.warnings import maybe_warn
 
 from ..config import SamplingParams
 from ..models import APIModel
@@ -54,9 +55,7 @@ async def _build_gemini_request(
 
     else:
         if sampling_params.reasoning_effort:
-            warnings.warn(
-                f"Ignoring reasoning_effort param for non-reasoning model: {model.name}"
-            )
+            maybe_warn("WARN_REASONING_UNSUPPORTED", model_name=model.name)
 
     # Add tools if provided
     if tools:
@@ -76,8 +75,10 @@ class GeminiRequest(APIRequestBase):
 
         # Warn if cache is specified for Gemini model
         if self.context.cache is not None:
-            warnings.warn(
-                f"Cache parameter '{self.context.cache}' is not supported for Gemini models, ignoring for {self.context.model_name}"
+            maybe_warn(
+                "WARN_CACHING_UNSUPPORTED",
+                model_name=self.context.model_name,
+                cache_param=self.context.cache,
             )
 
         self.model = APIModel.from_registry(self.context.model_name)

{lm_deluge-0.0.58 → lm_deluge-0.0.60}/src/lm_deluge/api_requests/mistral.py

@@ -1,9 +1,10 @@
 import json
 import os
-import warnings
 
 from aiohttp import ClientResponse
 
+from lm_deluge.warnings import maybe_warn
+
 from ..models import APIModel
 from ..prompt import Message
 from ..request_context import RequestContext
@@ -17,8 +18,10 @@ class MistralRequest(APIRequestBase):
 
         # Warn if cache is specified for non-Anthropic model
         if self.context.cache is not None:
-            warnings.warn(
-                f"Cache parameter '{self.context.cache}' is only supported for Anthropic models, ignoring for {self.context.model_name}"
+            maybe_warn(
+                "WARN_CACHING_UNSUPPORTED",
+                model_name=self.context.model_name,
+                cache_param=self.context.cache,
             )
         self.model = APIModel.from_registry(self.context.model_name)
 
@@ -38,13 +41,9 @@ class MistralRequest(APIRequestBase):
             "max_tokens": self.context.sampling_params.max_new_tokens,
         }
         if self.context.sampling_params.reasoning_effort:
-            warnings.warn(
-                f"Ignoring reasoning_effort param for non-reasoning model: {self.context.model_name}"
-            )
+            maybe_warn("WARN_REASONING_UNSUPPORTED", model_name=self.context.model_name)
         if self.context.sampling_params.logprobs:
-            warnings.warn(
-                f"Ignoring logprobs param for non-logprobs model: {self.context.model_name}"
-            )
+            maybe_warn("WARN_LOGPROBS_UNSUPPORTED", model_name=self.context.model_name)
         if self.context.sampling_params.json_mode and self.model.supports_json:
             self.request_json["response_format"] = {"type": "json_object"}
 

{lm_deluge-0.0.58 → lm_deluge-0.0.60}/src/lm_deluge/api_requests/openai.py

@@ -1,7 +1,6 @@
 import json
 import os
 import traceback as tb
-import warnings
 from types import SimpleNamespace
 
 import aiohttp
@@ -9,6 +8,7 @@ from aiohttp import ClientResponse
 
 from lm_deluge.request_context import RequestContext
 from lm_deluge.tool import MCPServer, Tool
+from lm_deluge.warnings import maybe_warn
 
 from ..config import SamplingParams
 from ..models import APIModel
@@ -30,6 +30,26 @@ async def _build_oa_chat_request(
         "temperature": sampling_params.temperature,
         "top_p": sampling_params.top_p,
     }
+    if context.service_tier:
+        assert context.service_tier in [
+            "auto",
+            "default",
+            "flex",
+            "priority",
+        ], f"Invalid service tier: {context.service_tier}"
+        # flex is only supported for o3, o4-mini, gpt-5 models
+        if context.service_tier == "flex":
+            model_supports_flex = any(x in model.id for x in ["o3", "o4-mini", "gpt-5"])
+            if not model_supports_flex:
+                print(
+                    f"WARNING: service_tier='flex' only supported for o3, o4-mini, gpt-5. "
+                    f"Using 'auto' instead for model {model.id}."
+                )
+                request_json["service_tier"] = "auto"
+            else:
+                request_json["service_tier"] = context.service_tier
+        else:
+            request_json["service_tier"] = context.service_tier
     # set max_tokens or max_completion_tokens dep. on provider
     if "cohere" in model.api_base:
         request_json["max_tokens"] = sampling_params.max_new_tokens
@@ -55,9 +75,8 @@ async def _build_oa_chat_request(
         request_json["reasoning_effort"] = effort
     else:
         if sampling_params.reasoning_effort:
-            warnings.warn(
-                f"Ignoring reasoning_effort param for non-reasoning model: {model.name}"
-            )
+            maybe_warn("WARN_REASONING_UNSUPPORTED", model_name=context.model_name)
+
     if sampling_params.logprobs:
         request_json["logprobs"] = True
         if sampling_params.top_logprobs is not None:
@@ -85,8 +104,10 @@ class OpenAIRequest(APIRequestBase):
 
         # Warn if cache is specified for non-Anthropic model
         if self.context.cache is not None:
-            warnings.warn(
-                f"Cache parameter '{self.context.cache}' is only supported for Anthropic models, ignoring for {self.context.model_name}"
+            maybe_warn(
+                "WARN_CACHING_UNSUPPORTED",
+                model_name=self.context.model_name,
+                cache_param=self.context.cache,
             )
         self.model = APIModel.from_registry(self.context.model_name)
 
@@ -213,9 +234,6 @@ class OpenAIRequest(APIRequestBase):
 async def _build_oa_responses_request(
     model: APIModel,
     context: RequestContext,
-    # prompt: Conversation,
-    # tools: list[Tool] | None,
-    # sampling_params: SamplingParams,
 ):
     prompt = context.prompt
     sampling_params = context.sampling_params
@@ -226,7 +244,28 @@ async def _build_oa_responses_request(
         "input": openai_responses_format["input"],
         "temperature": sampling_params.temperature,
         "top_p": sampling_params.top_p,
+        "background": context.background or False,
     }
+    if context.service_tier:
+        assert context.service_tier in [
+            "auto",
+            "default",
+            "flex",
+            "priority",
+        ], f"Invalid service tier: {context.service_tier}"
+        # flex is only supported for o3, o4-mini, gpt-5 models
+        if context.service_tier == "flex":
+            model_supports_flex = any(x in model.id for x in ["o3", "o4-mini", "gpt-5"])
+            if not model_supports_flex:
+                print(
+                    f"WARNING: service_tier='flex' only supported for o3, o4-mini, gpt-5. "
+                    f"Model {model.id} doesn't support flex. Using 'auto' instead."
+                )
+                request_json["service_tier"] = "auto"
+            else:
+                request_json["service_tier"] = context.service_tier
+        else:
+            request_json["service_tier"] = context.service_tier
     if sampling_params.max_new_tokens:
         request_json["max_output_tokens"] = sampling_params.max_new_tokens
 
@@ -245,9 +284,7 @@ async def _build_oa_responses_request(
         }
     else:
         if sampling_params.reasoning_effort:
-            warnings.warn(
-                f"Ignoring reasoning_effort for non-reasoning model: {model.id}"
-            )
+            maybe_warn("WARN_REASONING_UNSUPPORTED", model_name=context.model_name)
 
     if sampling_params.json_mode and model.supports_json:
         request_json["text"] = {"format": {"type": "json_object"}}
@@ -284,8 +321,10 @@ class OpenAIResponsesRequest(APIRequestBase):
         super().__init__(context)
         # Warn if cache is specified for non-Anthropic model
         if self.context.cache is not None:
-            warnings.warn(
-                f"Cache parameter '{self.context.cache}' is only supported for Anthropic models, ignoring for {self.context.model_name}"
+            maybe_warn(
+                "WARN_CACHING_UNSUPPORTED",
+                model_name=self.context.model_name,
+                cache_param=self.context.cache,
             )
         self.model = APIModel.from_registry(self.context.model_name)
 
@@ -488,8 +527,10 @@ async def stream_chat(
     extra_headers: dict[str, str] | None = None,
 ):
     if cache is not None:
-        warnings.warn(
-            f"Cache parameter '{cache}' is only supported for Anthropic models, ignoring for {model_name}"
+        maybe_warn(
+            "WARN_CACHING_UNSUPPORTED",
+            model_name=model_name,
+            cache_param=cache,
         )
 
     model = APIModel.from_registry(model_name)

{lm_deluge-0.0.58 → lm_deluge-0.0.60}/src/lm_deluge/batches.py

@@ -3,7 +3,7 @@ import json
 import os
 import tempfile
 import time
-from typing import Literal, Sequence
+from typing import Literal, Sequence, cast
 
 import aiohttp
 from rich.console import Console
@@ -16,7 +16,12 @@ from lm_deluge.api_requests.anthropic import _build_anthropic_request
 from lm_deluge.api_requests.openai import _build_oa_chat_request
 from lm_deluge.config import SamplingParams
 from lm_deluge.models import APIModel, registry
-from lm_deluge.prompt import CachePattern, Conversation, prompts_to_conversations
+from lm_deluge.prompt import (
+    CachePattern,
+    Conversation,
+    Prompt,
+    prompts_to_conversations,
+)
 from lm_deluge.request_context import RequestContext
 
 
@@ -166,14 +171,18 @@ async def _submit_anthropic_batch(file_path: str, headers: dict, model: str):
 async def create_batch_files_oa(
     model: str,
     sampling_params: SamplingParams,
-    prompts: Sequence[str | list[dict] | Conversation],
+    prompts: Prompt | Sequence[Prompt],
     batch_size: int = 50_000,
     destination: str | None = None,  # if none provided, temp files
 ):
     MAX_BATCH_SIZE_BYTES = 200 * 1024 * 1024  # 200MB
     MAX_BATCH_SIZE_ITEMS = batch_size
 
-    prompts = prompts_to_conversations(prompts)
+    if not isinstance(prompts, list):
+        prompts = cast(Sequence[Prompt], [prompts])
+
+    prompts = prompts_to_conversations(cast(Sequence[Prompt], prompts))
+    assert isinstance(prompts, Sequence)
     if any(p is None for p in prompts):
         raise ValueError("All prompts must be valid.")
 
@@ -251,14 +260,18 @@ async def create_batch_files_oa(
 async def submit_batches_oa(
     model: str,
     sampling_params: SamplingParams,
-    prompts: Sequence[str | list[dict] | Conversation],
+    prompts: Prompt | Sequence[Prompt],
     batch_size: int = 50_000,
 ):
     """Write OpenAI batch requests to a file and submit."""
     MAX_BATCH_SIZE_BYTES = 200 * 1024 * 1024  # 200MB
     MAX_BATCH_SIZE_ITEMS = batch_size
 
-    prompts = prompts_to_conversations(prompts)
+    if not isinstance(prompts, list):
+        prompts = prompts = cast(Sequence[Prompt], [prompts])
+
+    prompts = prompts_to_conversations(cast(Sequence[Prompt], prompts))
+    assert isinstance(prompts, Sequence)
     if any(p is None for p in prompts):
         raise ValueError("All prompts must be valid.")
 
@@ -342,7 +355,7 @@ async def submit_batches_oa(
 async def submit_batches_anthropic(
     model: str,
     sampling_params: SamplingParams,
-    prompts: Sequence[str | list[dict] | Conversation],
+    prompts: Prompt | Sequence[Prompt],
     *,
     cache: CachePattern | None = None,
     batch_size=100_000,
@@ -362,13 +375,16 @@ async def submit_batches_anthropic(
     MAX_BATCH_SIZE_ITEMS = batch_size
 
     # Convert prompts to Conversations
-    prompts = prompts_to_conversations(prompts)
+    if not isinstance(prompts, list):
+        prompts = prompts = cast(Sequence[Prompt], [prompts])
+
+    prompts = prompts_to_conversations(cast(Sequence[Prompt], prompts))
 
     request_headers = None
     batch_tasks = []
     current_batch = []
     current_batch_size = 0
-
+    assert isinstance(prompts, Sequence)
     for idx, prompt in enumerate(prompts):
         assert isinstance(prompt, Conversation)
         context = RequestContext(