lm-deluge 0.0.79__py3-none-any.whl → 0.0.80__py3-none-any.whl

lm_deluge/api_requests/anthropic.py

@@ -16,6 +16,7 @@ from lm_deluge.util.schema import (
     prepare_output_schema,
     transform_schema_for_anthropic,
 )
+from lm_deluge.warnings import maybe_warn
 
 from ..models import APIModel
 from .base import APIRequestBase, APIResponse
@@ -62,20 +63,45 @@ def _build_anthropic_request(
         "max_tokens": sampling_params.max_new_tokens,
     }
 
+    if model.id == "claude-4.5-opus" and sampling_params.global_effort:
+        request_json["effort"] = sampling_params.global_effort
+        _add_beta(base_headers, "effort-2025-11-24")
+
     # handle thinking
-    if model.reasoning_model and sampling_params.reasoning_effort:
-        # translate reasoning effort of low, medium, high to budget tokens
-        budget = {"minimal": 256, "low": 1024, "medium": 4096, "high": 16384}.get(
-            sampling_params.reasoning_effort
-        )
-        request_json["thinking"] = {
-            "type": "enabled",
-            "budget_tokens": budget,
-        }
-        if "top_p" in request_json:
-            request_json["top_p"] = max(request_json["top_p"], 0.95)
-        request_json["temperature"] = 1.0
-        request_json["max_tokens"] += budget
+    if model.reasoning_model:
+        if (
+            sampling_params.thinking_budget is not None
+            and sampling_params.reasoning_effort is not None
+        ):
+            maybe_warn("WARN_THINKING_BUDGET_AND_REASONING_EFFORT")
+
+        if sampling_params.thinking_budget is not None:
+            budget = sampling_params.thinking_budget
+        elif sampling_params.reasoning_effort is not None:
+            # translate reasoning effort of low, medium, high to budget tokens
+            budget = {
+                "none": 0,
+                "minimal": 256,
+                "low": 1024,
+                "medium": 4096,
+                "high": 16384,
+            }.get(sampling_params.reasoning_effort)
+            assert isinstance(budget, int)
+        else:
+            budget = 0
+
+        if budget > 0:
+            request_json["thinking"] = {
+                "type": "enabled",
+                "budget_tokens": budget,
+            }
+            if "top_p" in request_json:
+                request_json["top_p"] = max(request_json["top_p"], 0.95)
+            request_json["temperature"] = 1.0
+            request_json["max_tokens"] += budget
+        else:
+            request_json["thinking"] = {"type": "disabled"}
+
     else:
         request_json["thinking"] = {"type": "disabled"}
         if sampling_params.reasoning_effort:
@@ -83,10 +109,11 @@ def _build_anthropic_request(
     if system_message is not None:
         request_json["system"] = system_message
 
-    # handle temp + top_p for opus 4.1/sonnet 4.5
+    # handle temp + top_p for opus 4.1/sonnet 4.5.
+    # TODO: make clearer / more user-friendly so there can be NotGiven
+    # and user can control which one they want to use
     if "4-1" in model.name or "4-5" in model.name:
-        if "temperature" in request_json and "top_p" in request_json:
-            request_json.pop("top_p")
+        request_json.pop("top_p")
 
     # Handle structured outputs (output_format)
     if context.output_schema:

lm_deluge/api_requests/gemini.py

@@ -1,6 +1,5 @@
 import json
 import os
-from typing import Any
 
 from aiohttp import ClientResponse
 
@@ -52,47 +51,61 @@ async def _build_gemini_request(
         request_json["systemInstruction"] = {"parts": [{"text": system_message}]}
 
     # Handle reasoning models (thinking)
-    if model.reasoning_model:
-        thinking_config: dict[str, Any] | None = None
-        effort = sampling_params.reasoning_effort
-        is_gemini_3 = "gemini-3" in model.name.lower()
+    is_gemini_3 = "gemini-3" in model.name.lower()
+    if is_gemini_3:
+        # gemini3 MUST think
+        if not sampling_params.reasoning_effort:
+            maybe_warn("WARN_GEMINI3_NO_REASONING")
+            effort = "low"
+        else:
+            level_map = {
+                "none": "low",
+                "minimal": "low",
+                "low": "low",
+                "medium": "high",  # change when supported
+                "high": "high",
+            }
+            effort = level_map[sampling_params.reasoning_effort]
+        thinking_config = {"thinkingLevel": effort}
+        request_json["generationConfig"]["thinkingConfig"] = thinking_config
 
-        if is_gemini_3:
-            # Gemini 3 uses thinkingLevel instead of thinkingBudget
-            if effort in {"none", "minimal"}:
-                thinking_config = {"thinkingLevel": "low"}
-            elif effort is None:
-                # Default to high when reasoning is enabled but no preference was provided
-                thinking_config = {"thinkingLevel": "high"}
-            else:
-                # Map reasoning_effort to thinkingLevel
-                level_map = {
-                    "minimal": "low",
-                    "low": "low",
-                    "medium": "medium",  # Will work when supported
-                    "high": "high",
-                }
-                thinking_level = level_map.get(effort, "high")
-                thinking_config = {"thinkingLevel": thinking_level}
+    elif model.reasoning_model:
+        if (
+            sampling_params.thinking_budget is not None
+            and sampling_params.reasoning_effort is not None
+        ):
+            maybe_warn("WARN_THINKING_BUDGET_AND_REASONING_EFFORT")
+
+        if (
+            sampling_params.thinking_budget is not None
+            and sampling_params.thinking_budget > 0
+        ):
+            thinking_config = {
+                "includeThoughts": True,
+                "thinkingBudget": sampling_params.thinking_budget,
+            }
+        elif sampling_params.thinking_budget == -1:
+            # dynamic thinking
+            thinking_config = {"includeThoughts": True, "thinkingBudget": -1}
+        elif sampling_params.reasoning_effort not in [None, "none"]:
+            level_map = {
+                "minimal": 256,
+                "low": 1024,
+                "medium": 4096,
+                "high": 16384,
+            }
+            assert sampling_params.reasoning_effort in level_map
+            budget = level_map[sampling_params.reasoning_effort]
+            if "flash-lite" in model.id:
+                budget = max(budget, 512)
+            thinking_config = {"includeThoughts": True, "thinkingBudget": budget}
+        elif "2.5-pro" in model.id:
+            # 2.5 pro must think.
+            thinking_config = {"includeThoughts": True, "thinkingBudget": 128}
         else:
-            # Gemini 2.5 uses thinkingBudget (legacy)
-            if effort is None or effort == "none":
-                budget = 128 if "2.5-pro" in model.id else 0
-                # Explicitly disable thoughts when no effort is requested
-                thinking_config = {"includeThoughts": False, "thinkingBudget": budget}
-            else:
-                thinking_config = {"includeThoughts": True}
-                if (
-                    effort in {"minimal", "low", "medium", "high"}
-                    and "flash" in model.id
-                ):
-                    budget = {
-                        "minimal": 256,
-                        "low": 1024,
-                        "medium": 4096,
-                        "high": 16384,
-                    }[effort]
-                    thinking_config["thinkingBudget"] = budget
+            # no thoughts head empty
+            thinking_config = {"includeThoughts": False, "thinkingBudget": 0}
+
         request_json["generationConfig"]["thinkingConfig"] = thinking_config
 
     else:

lm_deluge/client.py

@@ -79,7 +79,7 @@ class _LLMClient(BaseModel):
     background: bool = False
     # sampling params - if provided, and sampling_params is not,
     # these override the defaults
-    temperature: float = 0.75
+    temperature: float = 1.0
     top_p: float = 1.0
     json_mode: bool = False
     max_new_tokens: int = 512
@@ -337,7 +337,7 @@ class _LLMClient(BaseModel):
         if "sampling_params" not in data or len(data.get("sampling_params", [])) == 0:
             data["sampling_params"] = [
                 SamplingParams(
-                    temperature=data.get("temperature", 0.75),
+                    temperature=data.get("temperature", 1.0),
                     top_p=data.get("top_p", 1.0),
                     json_mode=data.get("json_mode", False),
                     max_new_tokens=data.get("max_new_tokens", 512),
@@ -1067,7 +1067,7 @@ def LLMClient(
     extra_headers: dict[str, str] | None = None,
     use_responses_api: bool = False,
     background: bool = False,
-    temperature: float = 0.75,
+    temperature: float = 1.0,
     top_p: float = 1.0,
     json_mode: bool = False,
     max_new_tokens: int = 512,
@@ -1096,7 +1096,7 @@ def LLMClient(
     extra_headers: dict[str, str] | None = None,
     use_responses_api: bool = False,
     background: bool = False,
-    temperature: float = 0.75,
+    temperature: float = 1.0,
     top_p: float = 1.0,
     json_mode: bool = False,
     max_new_tokens: int = 512,
@@ -1124,7 +1124,7 @@ def LLMClient(
     extra_headers: dict[str, str] | None = None,
     use_responses_api: bool = False,
     background: bool = False,
-    temperature: float = 0.75,
+    temperature: float = 1.0,
     top_p: float = 1.0,
     json_mode: bool = False,
     max_new_tokens: int = 512,

lm_deluge/config.py

@@ -4,11 +4,13 @@ from pydantic import BaseModel
 
 
 class SamplingParams(BaseModel):
-    temperature: float = 0.0
+    temperature: float = 1.0  # more typical for new models
     top_p: float = 1.0
     json_mode: bool = False
     max_new_tokens: int = 2_048
+    global_effort: Literal["low", "medium", "high"] = "high"  # for opus-4.5
     reasoning_effort: Literal["low", "medium", "high", "minimal", "none", None] = None
+    thinking_budget: int | None = None
     logprobs: bool = False
     top_logprobs: int | None = None
     strict_tools: bool = True

lm_deluge/models/anthropic.py

@@ -10,6 +10,19 @@ ANTHROPIC_MODELS = {
     #                                                                   ░███
     #                                                                   █████
     #
+    "claude-4.5-opus": {
+        "id": "claude-4.5-opus",
+        "name": "claude-opus-4-5-20251101",
+        "api_base": "https://api.anthropic.com/v1",
+        "api_key_env_var": "ANTHROPIC_API_KEY",
+        "supports_json": False,
+        "api_spec": "anthropic",
+        "input_cost": 5.0,
+        "cached_input_cost": 0.50,
+        "cache_write_cost": 6.25,
+        "output_cost": 25.0,
+        "reasoning_model": True,
+    },
     "claude-4.5-haiku": {
         "id": "claude-4.5-haiku",
         "name": "claude-haiku-4-5-20251001",
@@ -21,6 +34,7 @@ ANTHROPIC_MODELS = {
         "cached_input_cost": 0.10,
         "cache_write_cost": 1.25,
         "output_cost": 3.0,
+        "reasoning_model": True,
     },
     "claude-4.5-sonnet": {
         "id": "claude-4.5-sonnet",
@@ -33,6 +47,7 @@ ANTHROPIC_MODELS = {
         "cached_input_cost": 0.30,
         "cache_write_cost": 3.75,
         "output_cost": 15.0,
+        "reasoning_model": True,
     },
     "claude-4.1-opus": {
         "id": "claude-4.1-opus",

lm_deluge/warnings.py

@@ -11,6 +11,8 @@ WARNINGS: dict[str, str] = {
     "WARN_MINIMAL_TO_NONE": "GPT-5.1 models don't support 'minimal' reasoning effort. Converting to 'none' for {model_name}.",
     "WARN_MEDIA_RESOLUTION_UNSUPPORTED": "media_resolution parameter is only supported for Gemini 3 models, ignoring for {model_name}.",
     "WARN_GEMINI3_MISSING_SIGNATURE": "Gemini 3 thought signature missing in {part_type}, injecting dummy signature 'context_engineering_is_the_way_to_go' to avoid API error.",
+    "WARN_GEMINI3_NO_REASONING": "Gemini 3 requires reasoning (thinkingConfig). Setting thinkingConfig to low.",
+    "WARN_THINKING_BUDGET_AND_REASONING_EFFORT": "`reasoning_effort` and `thinking_budget` both provided. `thinking_budget` will take priority.",
 }
 
 

{lm_deluge-0.0.79.dist-info → lm_deluge-0.0.80.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: lm_deluge
-Version: 0.0.79
+Version: 0.0.80
 Summary: Python utility for using LLM API models.
 Author-email: Benjamin Anderson <ben@trytaylor.ai>
 Requires-Python: >=3.10

{lm_deluge-0.0.79.dist-info → lm_deluge-0.0.80.dist-info}/RECORD

@@ -2,8 +2,8 @@ lm_deluge/__init__.py,sha256=zF5lAitfgJ8A28IXJ5BE9OUCqGOqSnGOWn3ZIlizNyY,822
 lm_deluge/batches.py,sha256=Km6QM5_7BlF2qEyo4WPlhkaZkpzrLqf50AaveHXQOoY,25127
 lm_deluge/cache.py,sha256=xO2AIYvP3tUpTMKQjwQQYfGRJSRi6e7sMlRhLjsS-u4,4873
 lm_deluge/cli.py,sha256=Ilww5gOw3J5v0NReq_Ra4hhxU4BCIJBl1oTGxJZKedc,12065
-lm_deluge/client.py,sha256=ZwDD4qkPFJsPxDMCijD6lz2s5ULL-hW58tGFN00BmSI,44796
-lm_deluge/config.py,sha256=7pTfqlg4qHf68qpckr21deVtCuao9b0ypiXT2k-nHUE,1210
+lm_deluge/client.py,sha256=VqCuFXM_ylO4v-lev85HMPFRHeU69tZo70favz-I2Uk,44791
+lm_deluge/config.py,sha256=C-_rVwAFL5sivLfKSkaa2ANMqqxKbyDCW86KfQB_Lck,1357
 lm_deluge/embed.py,sha256=CO-TOlC5kOTAM8lcnicoG4u4K664vCBwHF1vHa-nAGg,13382
 lm_deluge/errors.py,sha256=oHjt7YnxWbh-eXMScIzov4NvpJMo0-2r5J6Wh5DQ1tk,209
 lm_deluge/file.py,sha256=PTmlJQ-IaYcYUFun9V0bJ1NPVP84edJrR0hvCMWFylY,19697
@@ -15,14 +15,14 @@ lm_deluge/rerank.py,sha256=-NBAJdHz9OB-SWWJnHzkFmeVO4wR6lFV7Vw-SxG7aVo,11457
 lm_deluge/tool.py,sha256=ipgNy4OpfH3CA9OPQq5zfn1xO8H08GMvDynB8ZPQ5mA,30617
 lm_deluge/tracker.py,sha256=aeS9GUJpgOSQRVXAnGDvlMO8qYpSxpTNLYj2hrMg0m8,14757
 lm_deluge/usage.py,sha256=xz9tAw2hqaJvv9aAVhnQ6N1Arn7fS8Shb28VwCW26wI,5136
-lm_deluge/warnings.py,sha256=bAG9UXPnppk_oWGIsWpY3k5lWin4tganYFw0U7OEvJQ,2062
+lm_deluge/warnings.py,sha256=12RseSa9mYAFkbY783FQTP0x9RapRBErIQt4o7hzVnM,2321
 lm_deluge/api_requests/__init__.py,sha256=AbpHGcgLb-kRsJGnwFEktk7uzpZOCcBY74-YBdrKVGs,1
-lm_deluge/api_requests/anthropic.py,sha256=OvkciXTHyrG1cFyC1vv6nYyCFTqtMgt1r15Q-pbHiUQ,10411
+lm_deluge/api_requests/anthropic.py,sha256=ytNeADgGeflmlm5gVQ0cJ5bgchJ_EZvKJIIt7Imxf2A,11338
 lm_deluge/api_requests/base.py,sha256=mXEM85mcU_5LD-ugELpCl28tv-tpHKcaxerTIVLQZVo,10436
 lm_deluge/api_requests/bedrock.py,sha256=mY1xTvgfCLyqLlfFFmu_baKgkVq1Df1_MJXeN_G1jWQ,15597
 lm_deluge/api_requests/chat_reasoning.py,sha256=sJvstvKFqsSBUjYcwxzGt2_FH4cEp3Z6gKcBPyPjGwk,236
 lm_deluge/api_requests/common.py,sha256=BZ3vRO5TB669_UsNKugkkuFSzoLHOYJIKt4nV4sf4vc,422
-lm_deluge/api_requests/gemini.py,sha256=gHmIfEY48B-MYlJYxYc8hT8ojmK16XSETcvfljRKAH0,10813
+lm_deluge/api_requests/gemini.py,sha256=FjYKisAjD6rW2fA6WyXnnRn3oqJBXMod1_8HtGWIyEU,11099
 lm_deluge/api_requests/mistral.py,sha256=8JZP2CDf1XZfaPcTk0WS4q-VfYYj58ptpoH8LD3MQG4,4528
 lm_deluge/api_requests/openai.py,sha256=E0oakhcb2T5Swfn6ATMjRZKuLyRrx4Zj5SREo1JILfc,28841
 lm_deluge/api_requests/response.py,sha256=vG194gAH5p7ulpNy4qy5Pryfb1p3ZV21-YGoj__ru3E,7436
@@ -49,7 +49,7 @@ lm_deluge/llm_tools/subagents.py,sha256=srJ7On7YR0Y8WuNvf5TJl_7IUfEtG3zlxZeLgmn_
 lm_deluge/llm_tools/todos.py,sha256=doKJZWLZlh4J_k6HkdwonWHfZTZaxEI9_XHAoNFnfQo,14906
 lm_deluge/llm_tools/translate.py,sha256=iXyYvQZ8bC44FWhBk4qpdqjKM1WFF7Shq-H2PxhPgg4,1452
 lm_deluge/models/__init__.py,sha256=54H24K_eADbfdEH9aNORrNEXvDLZCQ4TEekeLiWljSE,4619
-lm_deluge/models/anthropic.py,sha256=sFkS-g0OWgRnVoFMKxWkSUt0qy2LVrcO5KtbYAG26iY,6283
+lm_deluge/models/anthropic.py,sha256=X92EYIapos-8LXnIYiypPJcFhI0tqmXja_w8e9H4CF8,6781
 lm_deluge/models/bedrock.py,sha256=g1PbfceSRH2lWST3ja0mUlF3oTq4e4T-si6RMe7qXgg,4888
 lm_deluge/models/cerebras.py,sha256=u2FMXJF6xMr0euDRKLKMo_NVTOcvSrrEpehbHr8sSeE,2050
 lm_deluge/models/cohere.py,sha256=iXjYtM6jy_YL73Op8OfNsrMNopwae9y-Sw-4vF9cEBw,3406
@@ -74,8 +74,8 @@ lm_deluge/util/schema.py,sha256=q6uwhA4s1lM2dHT1Kwc46E7OY1VecMOtTEI0PTFn6tA,1320
 lm_deluge/util/spatial.py,sha256=BsF_UKhE-x0xBirc-bV1xSKZRTUhsOBdGqsMKme20C8,4099
 lm_deluge/util/validation.py,sha256=hz5dDb3ebvZrZhnaWxOxbNSVMI6nmaOODBkk0htAUhs,1575
 lm_deluge/util/xml.py,sha256=Ft4zajoYBJR3HHCt2oHwGfymGLdvp_gegVmJ-Wqk4Ck,10547
-lm_deluge-0.0.79.dist-info/licenses/LICENSE,sha256=uNNXGXPCw2TC7CUs7SEBkA-Mz6QBQFWUUEWDMgEs1dU,1058
-lm_deluge-0.0.79.dist-info/METADATA,sha256=wqNdfbJ_BIJT-uZMOvwX9RWgqqzUFM4rZ_a4KblAFus,13705
-lm_deluge-0.0.79.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
-lm_deluge-0.0.79.dist-info/top_level.txt,sha256=hqU-TJX93yBwpgkDtYcXyLr3t7TLSCCZ_reytJjwBaE,10
-lm_deluge-0.0.79.dist-info/RECORD,,
+lm_deluge-0.0.80.dist-info/licenses/LICENSE,sha256=uNNXGXPCw2TC7CUs7SEBkA-Mz6QBQFWUUEWDMgEs1dU,1058
+lm_deluge-0.0.80.dist-info/METADATA,sha256=LJ2nPTs9WzdiP3kU5KPKUdOy_SuuiHRJCz9PINHEvZk,13705
+lm_deluge-0.0.80.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
+lm_deluge-0.0.80.dist-info/top_level.txt,sha256=hqU-TJX93yBwpgkDtYcXyLr3t7TLSCCZ_reytJjwBaE,10
+lm_deluge-0.0.80.dist-info/RECORD,,