lm-deluge 0.0.26__tar.gz → 0.0.28__tar.gz

{lm_deluge-0.0.26/src/lm_deluge.egg-info → lm_deluge-0.0.28}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: lm_deluge
-Version: 0.0.26
+Version: 0.0.28
 Summary: Python utility for using LLM API models.
 Author-email: Benjamin Anderson <ben@trytaylor.ai>
 Requires-Python: >=3.10
@@ -274,7 +274,7 @@ We support all models in `src/lm_deluge/models.py`. Vertex support is not planne
 
 ## Feature Support
 
-We support structured outputs via `json_mode` parameter provided to `SamplingParams`. Structured outputs with a schema are planned. Reasoning models are supported via the `reasoning_effort` parameter, which is translated to a thinking budget for Claude/Gemini. Image models are supported. We support tool use as documented above. We support logprobs for OpenAI models that return them.
+We support structured outputs via `json_mode` parameter provided to `SamplingParams`. Structured outputs with a schema are planned. Reasoning models are supported via the `reasoning_effort` parameter, which is translated to a thinking budget for Claude/Gemini. Passing `None` (or the string `"none"`) disables Gemini thoughts entirely. Image models are supported. We support tool use as documented above. We support logprobs for OpenAI models that return them.
 
 ## Built‑in tools
 

{lm_deluge-0.0.26 → lm_deluge-0.0.28}/README.md

@@ -247,7 +247,7 @@ We support all models in `src/lm_deluge/models.py`. Vertex support is not planne
 
 ## Feature Support
 
-We support structured outputs via `json_mode` parameter provided to `SamplingParams`. Structured outputs with a schema are planned. Reasoning models are supported via the `reasoning_effort` parameter, which is translated to a thinking budget for Claude/Gemini. Image models are supported. We support tool use as documented above. We support logprobs for OpenAI models that return them.
+We support structured outputs via `json_mode` parameter provided to `SamplingParams`. Structured outputs with a schema are planned. Reasoning models are supported via the `reasoning_effort` parameter, which is translated to a thinking budget for Claude/Gemini. Passing `None` (or the string `"none"`) disables Gemini thoughts entirely. Image models are supported. We support tool use as documented above. We support logprobs for OpenAI models that return them.
 
 ## Built‑in tools
 

{lm_deluge-0.0.26 → lm_deluge-0.0.28}/pyproject.toml

@@ -3,7 +3,7 @@ requires = ["setuptools", "wheel"]
 
 [project]
 name = "lm_deluge"
-version = "0.0.26"
+version = "0.0.28"
 authors = [{ name = "Benjamin Anderson", email = "ben@trytaylor.ai" }]
 description = "Python utility for using LLM API models."
 readme = "README.md"

{lm_deluge-0.0.26 → lm_deluge-0.0.28}/src/lm_deluge/api_requests/gemini.py

@@ -37,14 +37,17 @@ async def _build_gemini_request(
 
     # Handle reasoning models (thinking)
     if model.reasoning_model:
-        request_json["generationConfig"]["thinkingConfig"] = {"includeThoughts": True}
-        if sampling_params.reasoning_effort and "flash" in model.id:
-            budget = {"low": 1024, "medium": 4096, "high": 16384}.get(
-                sampling_params.reasoning_effort
-            )
-            request_json["generationConfig"]["thinkingConfig"]["thinkingBudget"] = (
-                budget
-            )
+        thinking_config = None
+        effort = sampling_params.reasoning_effort
+        if effort is None or effort == "none":
+            # Explicitly disable thoughts when no effort is requested
+            thinking_config = {"includeThoughts": False, "thinkingBudget": 0}
+        else:
+            thinking_config = {"includeThoughts": True}
+            if effort in {"low", "medium", "high"} and "flash" in model.id:
+                budget = {"low": 1024, "medium": 4096, "high": 16384}[effort]
+                thinking_config["thinkingBudget"] = budget
+        request_json["generationConfig"]["thinkingConfig"] = thinking_config
 
     else:
         if sampling_params.reasoning_effort:

{lm_deluge-0.0.26 → lm_deluge-0.0.28}/src/lm_deluge/api_requests/openai.py

@@ -37,7 +37,14 @@ async def _build_oa_chat_request(
     if model.reasoning_model:
         request_json["temperature"] = 1.0
         request_json["top_p"] = 1.0
-        request_json["reasoning_effort"] = sampling_params.reasoning_effort
+        effort = sampling_params.reasoning_effort
+        if effort in [None, "none"]:
+            # Disable reasoning for Gemini models when no effort requested
+            if "gemini" in model.id:
+                effort = "none"
+            else:
+                effort = "low"
+        request_json["reasoning_effort"] = effort
     else:
         if sampling_params.reasoning_effort:
             warnings.warn(

{lm_deluge-0.0.26 → lm_deluge-0.0.28/src/lm_deluge.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: lm_deluge
-Version: 0.0.26
+Version: 0.0.28
 Summary: Python utility for using LLM API models.
 Author-email: Benjamin Anderson <ben@trytaylor.ai>
 Requires-Python: >=3.10
@@ -274,7 +274,7 @@ We support all models in `src/lm_deluge/models.py`. Vertex support is not planne
 
 ## Feature Support
 
-We support structured outputs via `json_mode` parameter provided to `SamplingParams`. Structured outputs with a schema are planned. Reasoning models are supported via the `reasoning_effort` parameter, which is translated to a thinking budget for Claude/Gemini. Image models are supported. We support tool use as documented above. We support logprobs for OpenAI models that return them.
+We support structured outputs via `json_mode` parameter provided to `SamplingParams`. Structured outputs with a schema are planned. Reasoning models are supported via the `reasoning_effort` parameter, which is translated to a thinking budget for Claude/Gemini. Passing `None` (or the string `"none"`) disables Gemini thoughts entirely. Image models are supported. We support tool use as documented above. We support logprobs for OpenAI models that return them.
 
 ## Built‑in tools