lm-deluge 0.0.59__tar.gz → 0.0.60__tar.gz

{lm_deluge-0.0.59/src/lm_deluge.egg-info → lm_deluge-0.0.60}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: lm_deluge
-Version: 0.0.59
+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-0.0.59 → lm_deluge-0.0.60}/pyproject.toml

@@ -3,7 +3,7 @@ requires = ["setuptools", "wheel"]
 
 [project]
 name = "lm_deluge"
-version = "0.0.59"
+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-0.0.59 → 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-0.0.59 → 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-0.0.59 → 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-0.0.59 → 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
@@ -75,9 +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:
@@ -105,8 +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)
 
@@ -283,9 +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"}}
@@ -322,8 +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)
 
@@ -526,8 +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-0.0.59 → lm_deluge-0.0.60}/src/lm_deluge/client.py

@@ -117,13 +117,120 @@ class _LLMClient(BaseModel):
 
     # NEW! Builder methods
     def with_model(self, model: str):
-        self.model_names = [model]
+        self._update_models([model])
         return self
 
     def with_models(self, models: list[str]):
-        self.model_names = models
+        self._update_models(models)
         return self
 
+    def _update_models(self, models: list[str]) -> None:
+        normalized, per_model_efforts = self._normalize_model_names(models)
+        if self.reasoning_effort is None:
+            unique_efforts = {eff for eff in per_model_efforts if eff is not None}
+            if len(normalized) == 1 and per_model_efforts[0] is not None:
+                self.reasoning_effort = per_model_efforts[0]
+            elif (
+                len(unique_efforts) == 1
+                and len(unique_efforts) != 0
+                and None not in per_model_efforts
+            ):
+                self.reasoning_effort = next(iter(unique_efforts))  # type: ignore
+        self.model_names = normalized
+        self._align_sampling_params(per_model_efforts)
+        self._reset_model_weights()
+
+    def _normalize_model_names(
+        self, models: list[str]
+    ) -> tuple[list[str], list[Literal["low", "medium", "high"] | None]]:
+        reasoning_effort_suffixes: dict[str, Literal["low", "medium", "high"]] = {
+            "-low": "low",
+            "-medium": "medium",
+            "-high": "high",
+        }
+        normalized: list[str] = []
+        efforts: list[Literal["low", "medium", "high"] | None] = []
+
+        for name in models:
+            base_name = name
+            effort: Literal["low", "medium", "high"] | None = None
+            for suffix, candidate in reasoning_effort_suffixes.items():
+                if name.endswith(suffix) and len(name) > len(suffix):
+                    base_name = name[: -len(suffix)]
+                    effort = candidate
+                    break
+            normalized.append(base_name)
+            efforts.append(effort)
+
+        return normalized, efforts
+
+    def _align_sampling_params(
+        self, per_model_efforts: list[Literal["low", "medium", "high"] | None]
+    ) -> None:
+        if len(per_model_efforts) < len(self.model_names):
+            per_model_efforts = per_model_efforts + [None] * (
+                len(self.model_names) - len(per_model_efforts)
+            )
+
+        if not self.model_names:
+            self.sampling_params = []
+            return
+
+        if not self.sampling_params:
+            self.sampling_params = []
+
+        if len(self.sampling_params) == 0:
+            for _ in self.model_names:
+                self.sampling_params.append(
+                    SamplingParams(
+                        temperature=self.temperature,
+                        top_p=self.top_p,
+                        json_mode=self.json_mode,
+                        max_new_tokens=self.max_new_tokens,
+                        reasoning_effort=self.reasoning_effort,
+                        logprobs=self.logprobs,
+                        top_logprobs=self.top_logprobs,
+                    )
+                )
+        elif len(self.sampling_params) == 1 and len(self.model_names) > 1:
+            base_param = self.sampling_params[0]
+            self.sampling_params = [
+                base_param.model_copy(deep=True) for _ in self.model_names
+            ]
+        elif len(self.sampling_params) != len(self.model_names):
+            base_param = self.sampling_params[0]
+            self.sampling_params = [
+                base_param.model_copy(deep=True) for _ in self.model_names
+            ]
+
+        if self.reasoning_effort is not None:
+            for sp in self.sampling_params:
+                sp.reasoning_effort = self.reasoning_effort
+        else:
+            for sp, effort in zip(self.sampling_params, per_model_efforts):
+                if effort is not None:
+                    sp.reasoning_effort = effort
+
+    def _reset_model_weights(self) -> None:
+        if not self.model_names:
+            self.model_weights = []
+            return
+
+        if isinstance(self.model_weights, list):
+            if len(self.model_weights) == len(self.model_names) and any(
+                self.model_weights
+            ):
+                total = sum(self.model_weights)
+                if total == 0:
+                    self.model_weights = [
+                        1 / len(self.model_names) for _ in self.model_names
+                    ]
+                else:
+                    self.model_weights = [w / total for w in self.model_weights]
+                return
+        # Fallback to uniform distribution
+        self.model_weights = [1 / len(self.model_names) for _ in self.model_names]
+
     def with_limits(
         self,
         max_requests_per_minute: int | None = None,
@@ -150,8 +257,29 @@ class _LLMClient(BaseModel):
     @model_validator(mode="before")
     @classmethod
     def fix_lists(cls, data) -> "_LLMClient":
+        # Parse reasoning effort from model name suffixes (e.g., "gpt-5-high")
+        # Only applies when a single model string is provided
         if isinstance(data.get("model_names"), str):
+            model_name = data["model_names"]
+            reasoning_effort_suffixes = {
+                "-low": "low",
+                "-medium": "medium",
+                "-high": "high",
+            }
+
+            for suffix, effort in reasoning_effort_suffixes.items():
+                if model_name.endswith(suffix):
+                    # Extract base model name by removing suffix
+                    base_model = model_name[: -len(suffix)]
+                    data["model_names"] = base_model
+
+                    # Set reasoning_effort if not already explicitly set
+                    if data.get("reasoning_effort") is None:
+                        data["reasoning_effort"] = effort
+                    break
+
             data["model_names"] = [data["model_names"]]
+
         if not isinstance(data.get("sampling_params", []), list):
             data["sampling_params"] = [data["sampling_params"]]
         if "sampling_params" not in data or len(data.get("sampling_params", [])) == 0:

{lm_deluge-0.0.59 → lm_deluge-0.0.60}/src/lm_deluge/models/openai.py

@@ -10,6 +10,20 @@ OPENAI_MODELS = {
     #                ░███
     #                █████
     #               ░░░░░
+    "gpt-5-codex": {
+        "id": "gpt-5-codex",
+        "name": "gpt-5-codex",
+        "api_base": "https://api.openai.com/v1",
+        "api_key_env_var": "OPENAI_API_KEY",
+        "supports_json": False,
+        "supports_logprobs": True,
+        "supports_responses": True,
+        "api_spec": "openai",
+        "input_cost": 1.25,
+        "cached_input_cost": 0.125,
+        "output_cost": 10.0,
+        "reasoning_model": True,
+    },
     "gpt-5": {
         "id": "gpt-5",
         "name": "gpt-5",
@@ -79,6 +93,20 @@ OPENAI_MODELS = {
         "output_cost": 12.0,
         "reasoning_model": False,
     },
+    "codex-mini-latest": {
+        "id": "codex-mini-latest",
+        "name": "codex-mini-latest",
+        "api_base": "https://api.openai.com/v1",
+        "api_key_env_var": "OPENAI_API_KEY",
+        "supports_json": True,
+        "supports_logprobs": False,
+        "supports_responses": True,
+        "api_spec": "openai",
+        "input_cost": 1.5,
+        "cached_input_cost": 0.375,
+        "output_cost": 6.0,
+        "reasoning_model": True,
+    },
     "o3": {
         "id": "o3",
         "name": "o3-2025-04-16",

{lm_deluge-0.0.59 → lm_deluge-0.0.60}/src/lm_deluge/prompt.py

@@ -9,6 +9,7 @@ import xxhash
 
 from lm_deluge.file import File
 from lm_deluge.image import Image, MediaType
+from lm_deluge.warnings import deprecated
 
 CachePattern = Literal[
     "tools_only",
@@ -415,12 +416,17 @@ class Message:
 
         return cls(role, parts)
 
-    def add_text(self, content: str) -> "Message":
+    def with_text(self, content: str) -> "Message":
         """Append a text block and return self for chaining."""
         self.parts.append(Text(content))
         return self
 
-    def add_image(
+    @deprecated("with_text")
+    def add_text(self, content: str) -> "Message":
+        """Append a text block and return self for chaining."""
+        return self.with_text(content)
+
+    def with_image(
         self,
         data: bytes | str | Path | io.BytesIO | Image,
         *,
@@ -446,7 +452,27 @@ class Message:
         self.parts.append(img)
         return self
 
-    def add_file(
+    @deprecated("with_image")
+    def add_image(
+        self,
+        data: bytes | str | Path | io.BytesIO | Image,
+        *,
+        media_type: MediaType | None = None,
+        detail: Literal["low", "high", "auto"] = "auto",
+        max_size: int | None = None,
+    ) -> "Message":
+        """
+        Append an image block and return self for chaining.
+
+        If max_size is provided, the image will be resized so that its longer
+        dimension equals max_size, but only if the longer dimension is currently
+        larger than max_size.
+        """
+        return self.with_image(
+            data=data, media_type=media_type, detail=detail, max_size=max_size
+        )
+
+    def with_file(
         self,
         data: bytes | str | Path | io.BytesIO,
         *,
@@ -460,11 +486,29 @@ class Message:
         self.parts.append(file)
         return self
 
-    def add_tool_call(self, id: str, name: str, arguments: dict) -> "Message":
+    @deprecated("with_file")
+    def add_file(
+        self,
+        data: bytes | str | Path | io.BytesIO,
+        *,
+        media_type: str | None = None,
+        filename: str | None = None,
+    ) -> "Message":
+        """
+        Append a file block and return self for chaining.
+        """
+        return self.with_file(data, media_type=media_type, filename=filename)
+
+    def with_tool_call(self, id: str, name: str, arguments: dict) -> "Message":
         """Append a tool call block and return self for chaining."""
         self.parts.append(ToolCall(id=id, name=name, arguments=arguments))
         return self
 
+    @deprecated("with_tool_call")
+    def add_tool_call(self, id: str, name: str, arguments: dict) -> "Message":
+        """Append a tool call block and return self for chaining."""
+        return self.with_tool_call(id, name, arguments)
+
     def with_tool_result(
         self, tool_call_id: str, result: str | list[ToolResultPart]
     ) -> "Message":
@@ -472,11 +516,23 @@ class Message:
         self.parts.append(ToolResult(tool_call_id=tool_call_id, result=result))
         return self
 
-    def add_thinking(self, content: str) -> "Message":
+    @deprecated("with_tool_result")
+    def add_tool_result(
+        self, tool_call_id: str, result: str | list[ToolResultPart]
+    ) -> "Message":
+        """Append a tool result block and return self for chaining."""
+        return self.with_tool_result(tool_call_id, result)
+
+    def with_thinking(self, content: str) -> "Message":
         """Append a thinking block and return self for chaining."""
         self.parts.append(Thinking(content=content))
         return self
 
+    @deprecated("with_thinking")
+    def add_thinking(self, content: str) -> "Message":
+        """Append a thinking block and return self for chaining."""
+        return self.with_thinking(content)
+
     # -------- convenient constructors --------
     @classmethod
     def user(
@@ -488,25 +544,25 @@ class Message:
     ) -> "Message":
         res = cls("user", [])
         if text is not None:
-            res.add_text(text)
+            res.with_text(text)
         if image is not None:
-            res.add_image(image)
+            res.with_image(image)
         if file is not None:
-            res.add_file(file)
+            res.with_file(file)
         return res
 
     @classmethod
     def system(cls, text: str | None = None) -> "Message":
         res = cls("system", [])
         if text is not None:
-            res.add_text(text)
+            res.with_text(text)
         return res
 
     @classmethod
     def ai(cls, text: str | None = None) -> "Message":
         res = cls("assistant", [])
         if text is not None:
-            res.add_text(text)
+            res.with_text(text)
         return res
 
     # ──── provider-specific constructors ───
@@ -698,9 +754,9 @@ class Conversation:
     ) -> "Conversation":
         msg = Message.user(text)
         if image is not None:
-            msg.add_image(image)
+            msg.with_image(image)
         if file is not None:
-            msg.add_file(file)
+            msg.with_file(file)
         return cls([msg])
 
     @classmethod
@@ -1211,11 +1267,11 @@ class Conversation:
                 for i, tool_result in enumerate(m.tool_results):
                     images = tool_result.get_images()
                     if len(images) > 0:
-                        user_msg.add_text(
+                        user_msg.with_text(
                             f"[Images for Tool Call {tool_result.tool_call_id}]"
                         )
                         for img in images:
-                            user_msg.add_image(img)
+                            user_msg.with_image(img)
 
             else:
                 result.append(m.oa_chat())

lm_deluge-0.0.60/src/lm_deluge/warnings.py

@@ -0,0 +1,46 @@
+import functools
+import os
+import warnings
+
+WARNINGS: dict[str, str] = {
+    "WARN_JSON_MODE_UNSUPPORTED": "JSON mode requested for {model_name} but response_format parameter not supported.",
+    "WARN_REASONING_UNSUPPORTED": "Ignoring reasoning_effort param for non-reasoning model: {model_name}.",
+    "WARN_CACHING_UNSUPPORTED": "Cache parameter '{cache_param}' is not supported, ignoring for {model_name}.",
+    "WARN_LOGPROBS_UNSUPPORTED": "Ignoring logprobs param for non-logprobs model: {model_name}",
+}
+
+
+def maybe_warn(warning: str, **kwargs):
+    if os.getenv(warning):
+        pass
+    else:
+        warnings.warn(WARNINGS[warning].format(**kwargs))
+        os.environ[warning] = "1"
+
+
+def deprecated(replacement: str):
+    """Decorator to mark methods as deprecated and suggest replacement.
+
+    Only shows the warning once per method to avoid spam.
+
+    Args:
+        replacement: The name of the replacement method to suggest
+    """
+
+    def decorator(func):
+        warning_key = f"DEPRECATED_{func.__module__}_{func.__qualname__}"
+
+        @functools.wraps(func)
+        def wrapper(*args, **kwargs):
+            if not os.getenv(warning_key):
+                warnings.warn(
+                    f"{func.__name__} is deprecated, use {replacement} instead",
+                    DeprecationWarning,
+                    stacklevel=2,
+                )
+                os.environ[warning_key] = "1"
+            return func(*args, **kwargs)
+
+        return wrapper
+
+    return decorator

{lm_deluge-0.0.59 → lm_deluge-0.0.60/src/lm_deluge.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: lm_deluge
-Version: 0.0.59
+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-0.0.59 → lm_deluge-0.0.60}/src/lm_deluge.egg-info/SOURCES.txt

@@ -17,6 +17,7 @@ src/lm_deluge/rerank.py
 src/lm_deluge/tool.py
 src/lm_deluge/tracker.py
 src/lm_deluge/usage.py
+src/lm_deluge/warnings.py
 src/lm_deluge.egg-info/PKG-INFO
 src/lm_deluge.egg-info/SOURCES.txt
 src/lm_deluge.egg-info/dependency_links.txt