Merge pull request #276 from griptape-ai:131-add-min_p-and-top_k-for-…

…lmstudio-and-ollama

131-add-min_p-and-top_k-for-lmstudio-and-ollama

CHANGELOG.md

@@ -14,6 +14,22 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 ### Fixed 
 ### Security   -->
 
+## [2.2.11] - 2025-28-02
+### Added
+- Added `min_p` and/or `top_k` to prompt drivers that support them. 
+
+  - Top-k: Controls the variety of words the AI can choose from. A lower number (like 10) makes responses more focused and predictable, while a higher number (like 50) allows for more creativity and surprise.
+
+  - Min-p: Sets a quality threshold for word choices. Only words with at least a certain percentage of confidence compared to the best option are considered. Lower values (like 0.05) allow more variety, while higher values (like 0.3) stick closer to the most predictable options.
+
+  For example, if you ask the question "Give me one unexpected use for a paperclip"..
+  - With low top_k/high min_p: You'll get common answers (holding papers, makeshift hook)
+
+  - With high top_k/low min_p: You'll get more creative or unusual answers (lockpick, tiny sculpture material)
+
+  ℹ️ Note: Some models take `top_p` instead of `min_p`. To keep the parameters persistent, I'm still using `min_p` in the node, but setting it to `1-min_p`, so it _acts_ like `top_p`.
+
+
 ## [2.2.10] - 2025-27-02
 ### Added
 - Added ability to use key value pair replacement with Griptape Cloud Assistant as well.

nodes/drivers/gtUIAmazonBedrockPromptDriver.py

@@ -104,7 +104,8 @@ def build_params(self, **kwargs):
             params["aws_secret_access_key"] = secret_access_key
         if api_key:
             params["aws_access_key_id"] = api_key
-
+        params["min_p"] = kwargs.get("min_p", 0.1)
+        params["top_k"] = kwargs.get("top_k", 40)
         return params
 
     def create(self, **kwargs):

nodes/drivers/gtUIAmazonSageMakerJumpstartPromptDriver.py

@@ -18,6 +18,8 @@ def INPUT_TYPES(cls):
         inputs = super().INPUT_TYPES()
 
         inputs["required"].update()
+        del inputs["optional"]["min_p"]
+        del inputs["optional"]["top_k"]
         inputs["optional"].update(
             {
                 "model": (
@@ -100,6 +102,7 @@ def create(self, **kwargs):
             params["endpoint"] = endpoint
         if use_native_tools:
             params["use_native_tools"] = use_native_tools
+
         try:
             driver = AmazonSageMakerJumpstartPromptDriver(**params)
             return (driver,)

nodes/drivers/gtUIAnthropicPromptDriver.py

@@ -23,7 +23,7 @@ def INPUT_TYPES(cls):
 
         # Add the optional inputs
         inputs["optional"].update(base_optional_inputs)
-
+        del inputs["optional"]["min_p"]
         # Set model default
         inputs["optional"]["model"] = (
             models,

nodes/drivers/gtUIAzureOpenAiChatPromptDriver.py

@@ -27,6 +27,8 @@ def INPUT_TYPES(cls):
 
         # Add the optional inputs
         inputs["optional"].update(base_optional_inputs)
+        del inputs["optional"]["min_p"]
+        del inputs["optional"]["top_k"]
         inputs["optional"].update(
             {
                 "model": (models, {"default": models[0]}),
@@ -78,6 +80,10 @@ def build_params(self, **kwargs):
             "temperature": temperature,
             "max_attempts": max_attempts_on_fail,
             "use_native_tools": use_native_tools,
+            # "extra_params": {
+            #     # "min_p": min_p,
+            #     "top_k": top_k,
+            # },
         }
         if response_format == "json_object":
             response_format = {"type": "json_object"}

nodes/drivers/gtUIBasePromptDriver.py

@@ -55,6 +55,23 @@ def INPUT_TYPES(cls):
                         "tooltip": "Maximum tokens to generate. If <=0, it will use the default based on the tokenizer.",
                     },
                 ),
+                "min_p": (
+                    "FLOAT",
+                    {
+                        "default": 0.1,
+                        "min": 0.0,
+                        "max": 1.0,
+                        "step": 0.01,
+                        "tooltip": "Minimum probability for sampling. Lower values will be more random.",
+                    },
+                ),
+                "top_k": (
+                    "INT",
+                    {
+                        "default": 40,
+                        "tooltip": "Top k for sampling. Lower values are more deterministic.",
+                    },
+                ),
             },
         )
         return inputs

nodes/drivers/gtUICoherePromptDriver.py

@@ -51,11 +51,17 @@ def build_params(self, **kwargs):
         max_attempts = kwargs.get("max_attempts_on_fail", None)
         use_native_tools = kwargs.get("use_native_tools", False)
         max_tokens = kwargs.get("max_tokens", None)
+        min_p = kwargs.get("min_p", None)
+        top_k = kwargs.get("top_k", None)
         params = {
             "api_key": api_key,
             "model": model,
             "max_attempts": max_attempts,
             "use_native_tools": use_native_tools,
+            "extra_params": {
+                "p": 1 - min_p,
+                "k": top_k,
+            },
         }
         if max_tokens > 0:
             params["max_tokens"] = max_tokens

nodes/drivers/gtUIGooglePromptDriver.py

@@ -18,7 +18,7 @@ class gtUIGooglePromptDriver(gtUIBasePromptDriver):
     @classmethod
     def INPUT_TYPES(cls):
         inputs = super().INPUT_TYPES()
-
+        del inputs["optional"]["min_p"]
         inputs["optional"].update(
             {
                 "model": (models, {"default": models[0]}),

nodes/drivers/gtUIGriptapeCloudPromptDriver.py

@@ -28,7 +28,7 @@ def INPUT_TYPES(cls):
 
         # Add the optional inputs
         inputs["optional"].update(base_optional_inputs)
-
+        del inputs["optional"]["top_k"]
         # Set model default
         inputs["optional"]["model"] = (models, {"default": DEFAULT_MODEL})
         inputs["optional"].update(
@@ -83,6 +83,7 @@ def build_params(self, **kwargs):
             params["use_native_tools"] = use_native_tools
         if max_tokens > 0:
             params["max_tokens"] = max_tokens
+        params["extra_params"]["top_p"] = 1 - kwargs.get("min_p", None)
 
         return params
 

nodes/drivers/gtUIGrokPromptDriver.py

@@ -29,6 +29,8 @@ def INPUT_TYPES(cls):
         # Add the optional inputs
         inputs["optional"].update(base_optional_inputs)
 
+        del inputs["optional"]["top_k"]
+
         # Set model default
         inputs["optional"]["model"] = (models, {"default": DEFAULT_MODEL})
         inputs["optional"].update(
@@ -83,6 +85,7 @@ def build_params(self, **kwargs):
             params["use_native_tools"] = use_native_tools
         if max_tokens > 0:
             params["max_tokens"] = max_tokens
+        params["extra_params"]["top_p"] = 1 - kwargs.get("min_p", None)
 
         return params
 

nodes/drivers/gtUIGroqChatPromptDriver.py

@@ -77,6 +77,10 @@ def build_params(self, **kwargs):
             "temperature": temperature,
             "use_native_tools": use_native_tools,
             "max_attempts": max_attempts,
+            "modalities": [],
+            "extra_params": {
+                "top_p": 1 - kwargs.get("min_p", None),
+            },
         }
         if response_format == "json_object":
             params["response_format"] = {"type": response_format}

nodes/drivers/gtUIHuggingFaceHubPromptDriver.py

@@ -15,6 +15,8 @@ def INPUT_TYPES(cls):
         inputs = super().INPUT_TYPES()
 
         inputs["required"].update({})
+        del inputs["optional"]["min_p"]
+        del inputs["optional"]["top_k"]
         inputs["optional"].update(
             {
                 "model": (

nodes/drivers/gtUILMStudioChatPromptDriver.py

@@ -69,14 +69,18 @@ def build_params(self, **kwargs):
         max_attempts = kwargs.get("max_attempts_on_fail", None)
         use_native_tools = kwargs.get("use_native_tools", False)
         max_tokens = kwargs.get("max_tokens", None)
-
+        top_p = 1 - kwargs.get("min_p", None)
         params = {
             "model": model,
             "base_url": f"{base_url}:{port}/v1",
             "api_key": api_key,
             "temperature": temperature,
             "use_native_tools": use_native_tools,
             "max_attempts": max_attempts,
+            "extra_params": {
+                "top_p": top_p,
+                # "top_k": top_k,
+            },
         }
         if response_format == "json_object":
             params["response_format"] = {"type": response_format}

nodes/drivers/gtUIOllamaPromptDriver.py

@@ -85,7 +85,8 @@ def build_params(self, **kwargs):
         use_native_tools = kwargs.get("use_native_tools", False)
         max_tokens = kwargs.get("max_tokens", None)
         keep_alive = kwargs.get("keep_alive", 240)
-
+        min_p = kwargs.get("min_p")
+        top_k = kwargs.get("top_k")
         params = {
             "model": model,
             "temperature": temperature,
@@ -96,7 +97,13 @@ def build_params(self, **kwargs):
             params["host"] = f"{base_url}:{port}"
         if max_tokens > 0:
             params["max_tokens"] = max_tokens
-        params["extra_params"] = {"keep_alive": int(keep_alive)}
+        params["extra_params"] = {
+            "keep_alive": int(keep_alive),
+            "options": {
+                "min_p": min_p,
+                "top_k": top_k,
+            },
+        }
         return params
 
     def create(self, **kwargs):

nodes/drivers/gtUIOpenAiChatPromptDriver.py

@@ -32,6 +32,8 @@ def INPUT_TYPES(cls):
         # Add the optional inputs
         inputs["optional"].update(base_optional_inputs)
 
+        del inputs["optional"]["top_k"]
+
         # Set model default
         inputs["optional"]["model"] = (models, {"default": DEFAULT_MODEL})
         inputs["optional"].update(
@@ -85,7 +87,9 @@ def build_params(self, **kwargs):
             params["use_native_tools"] = use_native_tools
         if max_tokens > 0:
             params["max_tokens"] = max_tokens
-
+        params["extra_params"] = {
+            "top_p": 1 - kwargs.get("min_p", None),
+        }
         return params
 
     def create(self, **kwargs):

pyproject.toml

@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "comfyui-griptape"
-version = "2.2.10"
+version = "2.2.11"
 description = "Griptape LLM(Large Language Model) Nodes for ComfyUI."
 authors = ["Jason Schleifer <[email protected]>"]
 readme = "README.md"
@@ -9,7 +9,7 @@ readme = "README.md"
 [project]
 name = "comfyui-griptape"
 description = "Griptape LLM(Large Language Model) Nodes for ComfyUI."
-version = "2.2.10" 
+version = "2.2.11" 
 license = {file = "LICENSE"}
 dependencies = ["attrs>=24.3.0,<26.0.0", "openai>=1.58.1,<2.0.0", "griptape[all]>=1.4.0", "python-dotenv", "poetry==1.8.5", "griptape-black-forest @ git+https://github.com/griptape-ai/griptape-black-forest.git", "griptape_serper_driver_extension @ git+https://github.com/mertdeveci5/griptape-serper-driver-extension.git"]