feat(gui): enhance temperature and top_p control precision (#112)

* feat(gui): enhance temperature and top_p control precision * fix: correct max_temperature handling as float across app * fix: removed unused `SetName` method in `FloatSpinTextCtrlAccessible` class * fix: standardize temperature settings across models * fix: reorder import statements
SigmaNight · Jul 7, 2024 · b9d0bb1 · b9d0bb1
1 parent 0e6fad0
commit b9d0bb1
Show file tree

Hide file tree

Showing 4 changed files with 57 additions and 13 deletions.
diff --git a/basilisk/gui/conversation_tab.py b/basilisk/gui/conversation_tab.py
@@ -9,6 +9,7 @@
 from uuid import UUID
 
 import wx
+from wx.lib.agw.floatspin import FloatSpin
 
 import basilisk.config as config
 from basilisk import global_vars
@@ -31,6 +32,17 @@
 log = logging.getLogger(__name__)
 
 
+class FloatSpinTextCtrlAccessible(wx.Accessible):
+	def __init__(self, win: wx.Window = None, name: str = None):
+		super().__init__(win)
+		self._name = name
+
+	def GetName(self, childId):
+		if self._name:
+			return (wx.ACC_OK, self._name)
+		return super().GetName(childId)
+
+
 class ConversationTab(wx.Panel):
 	def __init__(self, parent: wx.Window):
 		wx.Panel.__init__(self, parent)
@@ -161,9 +173,20 @@ def init_ui(self):
 			label=_("&Temperature:"),
 		)
 		sizer.Add(self.temperature_label, proportion=0, flag=wx.EXPAND)
-		self.temperature_spinner = wx.SpinCtrl(
-			self, value="100", min=0, max=200
-		)
+		self.temperature_spinner = FloatSpin(
+			self,
+			min_val=0.0,
+			max_val=2.0,
+			increment=0.01,
+			value=0.5,
+			digits=2,
+			name="temperature",
+		)
+		float_spin_accessible = FloatSpinTextCtrlAccessible(
+			win=self.temperature_spinner._textctrl,
+			name=self.temperature_label.GetLabel().replace("&", ""),
+		)
+		self.temperature_spinner._textctrl.SetAccessible(float_spin_accessible)
 		sizer.Add(self.temperature_spinner, proportion=0, flag=wx.EXPAND)
 
 		self.top_p_label = wx.StaticText(
@@ -172,7 +195,20 @@ def init_ui(self):
 			label=_("Probabilit&y Mass (top P):"),
 		)
 		sizer.Add(self.top_p_label, proportion=0, flag=wx.EXPAND)
-		self.top_p_spinner = wx.SpinCtrl(self, value="100", min=0, max=100)
+		self.top_p_spinner = FloatSpin(
+			self,
+			min_val=0.0,
+			max_val=1.0,
+			increment=0.01,
+			value=1.0,
+			digits=2,
+			name="Top P",
+		)
+		float_spin_accessible = FloatSpinTextCtrlAccessible(
+			win=self.top_p_spinner._textctrl,
+			name=self.top_p_label.GetLabel().replace("&", ""),
+		)
+		self.top_p_spinner._textctrl.SetAccessible(float_spin_accessible)
 		sizer.Add(self.top_p_spinner, proportion=0, flag=wx.EXPAND)
 
 		self.stream_mode = wx.CheckBox(
@@ -424,10 +460,8 @@ def on_model_change(self, event: wx.CommandEvent):
 		if model_index == wx.NOT_FOUND:
 			return
 		model = self.current_engine.models[model_index]
-		self.temperature_spinner.SetMax(int(model.max_temperature * 100))
-		self.temperature_spinner.SetValue(
-			str(int(model.max_temperature / 2 * 100))
-		)
+		self.temperature_spinner.SetMax(model.max_temperature)
+		self.temperature_spinner.SetValue(model.default_temperature)
 		max_tokens = model.max_output_tokens
 		if max_tokens < 1:
 			max_tokens = model.context_window
@@ -760,8 +794,8 @@ def on_submit(self, event: wx.CommandEvent):
 				),
 			),
 			model=model,
-			temperature=self.temperature_spinner.GetValue() / 100,
-			top_p=self.top_p_spinner.GetValue() / 100,
+			temperature=self.temperature_spinner.GetValue(),
+			top_p=self.top_p_spinner.GetValue(),
 			max_tokens=self.max_tokens_spin_ctrl.GetValue(),
 			stream=self.stream_mode.GetValue(),
 		)

diff --git a/basilisk/provider_ai_model.py b/basilisk/provider_ai_model.py
@@ -13,8 +13,8 @@ class ProviderAIModel:
 	description: str | None = field(default=None)
 	context_window: int = field(default=0)
 	max_output_tokens: int = field(default=-1)
-	max_temperature: float = field(default=2)
-	default_temperature: float = field(default=1)
+	max_temperature: float = field(default=1.0)
+	default_temperature: float = field(default=1.0)
 	vision: bool = field(default=False)
 	preview: bool = field(default=False)
 	extra_info: dict[str, Any] = field(default_factory=dict)

diff --git a/basilisk/provider_engine/openai_engine.py b/basilisk/provider_engine/openai_engine.py
@@ -67,6 +67,7 @@ def models(self) -> list[ProviderAIModel]:
 				context_window=128000,
 				max_output_tokens=4096,
 				vision=True,
+				max_temperature=2.0,
 			),
 			ProviderAIModel(
 				id="gpt-4-turbo",
@@ -77,6 +78,7 @@ def models(self) -> list[ProviderAIModel]:
 				context_window=128000,
 				max_output_tokens=4096,
 				vision=True,
+				max_temperature=2.0,
 			),
 			ProviderAIModel(
 				id="gpt-3.5-turbo",
@@ -86,6 +88,7 @@ def models(self) -> list[ProviderAIModel]:
 				),
 				context_window=16385,
 				max_output_tokens=4096,
+				max_temperature=2.0,
 			),
 			ProviderAIModel(
 				id="gpt-3.5-turbo-0125",
@@ -95,6 +98,7 @@ def models(self) -> list[ProviderAIModel]:
 				),
 				context_window=16385,
 				max_output_tokens=4096,
+				max_temperature=2.0,
 			),
 			ProviderAIModel(
 				id="gpt-4-turbo-preview",
@@ -104,6 +108,7 @@ def models(self) -> list[ProviderAIModel]:
 				),
 				context_window=128000,
 				max_output_tokens=4096,
+				max_temperature=2.0,
 			),
 			ProviderAIModel(
 				id="gpt-4-0125-preview",
@@ -113,6 +118,7 @@ def models(self) -> list[ProviderAIModel]:
 				),
 				context_window=128000,
 				max_output_tokens=4096,
+				max_temperature=2.0,
 			),
 			ProviderAIModel(
 				id="gpt-4-1106-preview",
@@ -122,6 +128,7 @@ def models(self) -> list[ProviderAIModel]:
 				),
 				context_window=128000,
 				max_output_tokens=4096,
+				max_temperature=2.0,
 			),
 			ProviderAIModel(
 				id="gpt-4-vision-preview",
@@ -132,6 +139,7 @@ def models(self) -> list[ProviderAIModel]:
 				context_window=128000,
 				max_output_tokens=4096,
 				vision=True,
+				max_temperature=2.0,
 			),
 			ProviderAIModel(
 				id="gpt-4-0613",
@@ -140,6 +148,7 @@ def models(self) -> list[ProviderAIModel]:
 					"More capable than any GPT-3.5 ProviderAIModel, able to do more complex tasks, and optimized for chat"
 				),
 				max_output_tokens=8192,
+				max_temperature=2.0,
 			),
 			ProviderAIModel(
 				id="gpt-4-32k-0613",
@@ -149,6 +158,7 @@ def models(self) -> list[ProviderAIModel]:
 				),
 				context_window=32768,
 				max_output_tokens=8192,
+				max_temperature=2.0,
 			),
 		]
 

diff --git a/basilisk/provider_engine/openrouter_engine.py b/basilisk/provider_engine/openrouter_engine.py
@@ -39,7 +39,7 @@ def models(self) -> list[ProviderAIModel]:
 							"max_completion_tokens"
 						)
 						or -1,
-						max_temperature=2,
+						max_temperature=2.0,
 						vision="#multimodal" in model['description'],
 						preview="-preview" in model['id'],
 						extra_info={