webscout 6.0__py3-none-any.whl → 6.2b0__py3-none-any.whl

webscout/Local/samplers.py

@@ -1,161 +1,125 @@
-
-from ._version import __version__, __llama_cpp_version__
-
-"""Submodule containing SamplerSettings class and some preset samplers"""
-
+from typing import Optional
 from sys import maxsize
 
+from .utils import assert_type, NoneType
 
 MAX_TEMP = float(maxsize)
 
 class SamplerSettings:
     """
-    A SamplerSettings object specifies the sampling parameters that will be
-    used to control text generation
+    Specifies sampling parameters for controlling text generation.
+
+    This class allows you to fine-tune the behavior of text generation
+    models by adjusting various sampling parameters. These settings are
+    passed as an optional parameter to functions like `Thread.__init__()`,
+    `Model.generate()`, `Model.stream()`, and `Model.stream_print()`.
+
+    If a parameter is unspecified, the default value from llama.cpp is used.
+    If all parameters are unspecified, the behavior is equivalent to
+    `DefaultSampling`.
+
+    Setting a parameter explicitly to `None` disables it. When all samplers
+    are disabled, it's equivalent to `NoSampling` (unmodified probability
+    distribution).
+
+    Attributes:
+        max_len_tokens (Optional[int]): Maximum number of tokens to generate.
+                                       Defaults to -1 (no limit).
+        top_k (Optional[int]): Number of highest probability tokens to consider.
+                              Defaults to 40. Set to `None` to disable.
+        top_p (Optional[float]): Nucleus sampling threshold (0.0 - 1.0).
+                               Defaults to 0.95. Set to `None` to disable.
+        min_p (Optional[float]): Minimum probability threshold (0.0 - 1.0).
+                               Defaults to 0.05. Set to `None` to disable.
+        temp (Optional[float]): Temperature for sampling (0.0 - inf).
+                             Defaults to 0.8. Set to `None` to disable.
+        frequency_penalty (Optional[float]): Penalty for repeating tokens.
+                                            Defaults to 0.0.
+        presence_penalty (Optional[float]): Penalty for generating new tokens.
+                                           Defaults to 0.0.
+        repeat_penalty (Optional[float]): Penalty for repeating token sequences.
+                                         Defaults to 1.0.
+
+    Presets:
+        - `GreedyDecoding`: Always chooses the most likely token.
+        - `DefaultSampling`: Uses default parameters from llama.cpp.
+        - `NoSampling`: Unmodified probability distribution (all parameters disabled).
+        - `ClassicSampling`: Reflects old llama.cpp defaults.
+        - `SemiSampling`: Halfway between DefaultSampling and SimpleSampling.
+        - `TikTokenSampling`: For models with large vocabularies.
+        - `LowMinPSampling`, `MinPSampling`, `StrictMinPSampling`: Use `min_p` as the only active sampler.
+        - `ContrastiveSearch`, `WarmContrastiveSearch`: Implement contrastive search.
+        - `RandomSampling`: Outputs completely random tokens (useless).
+        - `LowTempSampling`: Default sampling with reduced temperature.
+        - `HighTempSampling`: Default sampling with increased temperature.
+        - `LowTopPSampling`, `TopPSampling`, `StrictTopPSampling`: Use `top_p` as the primary sampler.
+        - `MidnightMiqu`: For sophosympatheia/Midnight-Miqu-70B-v1.5 model.
+        - `Llama3`: For meta-llama/Meta-Llama-3.1-8B-Instruct model.
+        - `Nemo`: For mistralai/Mistral-Nemo-Instruct-2407 model.
     """
-
-    ParamTypes: dict[str, type] = {
-        'max_len_tokens':    int,
-        'temp':              float,
-        'top_p':             float,
-        'min_p':             float,
-        'frequency_penalty': float,
-        'presence_penalty':  float,
-        'repeat_penalty':    float,
-        'top_k':             int
+    
+    param_types: dict[str, tuple[type]] = {
+        'max_len_tokens': (int, NoneType),
+        'top_k': (int, NoneType),
+        'top_p': (float, NoneType),
+        'min_p': (float, NoneType),
+        'temp': (float, NoneType),
+        'frequency_penalty': (float, NoneType),
+        'presence_penalty': (float, NoneType),
+        'repeat_penalty': (float, NoneType)
     }
 
     def __init__(
         self,
-        max_len_tokens:    int   = -1,
-        temp:              float = 0.8,
-        top_p:             float = 0.95,
-        min_p:             float = 0.05,
-        frequency_penalty: float = 0.0,
-        presence_penalty:  float = 0.0,
-        repeat_penalty:    float = 1.0,
-        top_k:             int   = 40
+        max_len_tokens: Optional[int] = -1,
+        top_k: Optional[int] = 40,
+        top_p: Optional[float] = 0.95,
+        min_p: Optional[float] = 0.05,
+        temp: Optional[float] = 0.8,
+        frequency_penalty: Optional[float] = 0.0,
+        presence_penalty: Optional[float] = 0.0,
+        repeat_penalty: Optional[float] = 1.0
     ):
-        """
-        Construct a new SamplerSettings instance
-        """
+        self.max_len_tokens = max_len_tokens if max_len_tokens is not None else -1
+        self.top_k = top_k if top_k is not None else -1
+        self.top_p = top_p if top_p is not None else 1.0
+        self.min_p = min_p if min_p is not None else 0.0
+        self.temp = temp if temp is not None else 1.0
+        self.frequency_penalty = frequency_penalty if frequency_penalty is not None else 0.0
+        self.presence_penalty = presence_penalty if presence_penalty is not None else 0.0
+        self.repeat_penalty = repeat_penalty if repeat_penalty is not None else 1.0
+
+        # Validate parameters using param_types dictionary
+        for param_name, param_value in self.__dict__.items():
+            assert_type(param_value, self.param_types[param_name],
+                        f'{param_name} parameter', 'SamplerSettings')
 
-        self.max_len_tokens    = max_len_tokens
-        self.temp              = temp
-        self.top_p             = top_p
-        self.min_p             = min_p
-        self.frequency_penalty = frequency_penalty
-        self.presence_penalty  = presence_penalty
-        self.repeat_penalty    = repeat_penalty
-        self.top_k             = top_k
-
-        for sampler_param in SamplerSettings.ParamTypes:
-            expected_type = SamplerSettings.ParamTypes[sampler_param]
-            actual_type = type(getattr(self, sampler_param))
-            if actual_type != expected_type:
-                raise TypeError(
-                    f"wrong type for SamplerSettings parameter '{sampler_param}'"
-                    f" - expected {expected_type}, got {actual_type}"
-                )
-    
     def __repr__(self) -> str:
-        repr_str = 'SamplerSettings('
-        repr_str += f'max_len_tokens={self.max_len_tokens}, '
-        repr_str += f'temp={self.temp}, '
-        repr_str += f'top_p={self.top_p}, '
-        repr_str += f'min_p={self.min_p}, '
-        repr_str += f'frequency_penalty={self.frequency_penalty}, '
-        repr_str += f'presence_penalty={self.presence_penalty}, '
-        repr_str += f'repeat_penalty={self.repeat_penalty}, '
-        repr_str += f'top_k={self.top_k})'
-        return repr_str
-
-# most likely token is always chosen
-GreedyDecoding = SamplerSettings(
-    temp = 0.0,
-)
+        params = ', '.join(
+            f'{name}={value}' for name, value in self.__dict__.items()
+        )
+        return f'SamplerSettings({params})'
 
-# reflects llama.cpp
+# Predefined sampler settings
+GreedyDecoding = SamplerSettings(temp=0.0)
 DefaultSampling = SamplerSettings()
-
-# unmodified probability distribution (i.e. what the model actually thinks)
-SimpleSampling = SamplerSettings(
-    temp = 1.0,
-    top_p = 1.0,
-    min_p = 0.0,
-    top_k = -1
-)
-
-# reflects old llama.cpp defaults
-ClassicSampling = SamplerSettings(
-    min_p=0.0,
-    repeat_penalty = 1.1
-)
-
-# halfway between DefaultSampling and SimpleSampling
-SemiSampling = SamplerSettings(
-    temp=0.9,
-    top_p=0.975,
-    min_p=0.025,
-    top_k=80
-)
-
-# for models with large vocabulary, which tend to run hot
-TikTokenSampling = SamplerSettings(
-    temp=0.6,
-    repeat_penalty=1.1
-)
-
-# use min_p as the only active sampler (more permissive)
-LowMinPSampling = SamplerSettings(
-    temp = 1.0,
-    top_p = 1.0,
-    min_p = 0.05,
-    top_k = -1
-)
-
-# use min_p as the only active sampler (moderate)
-MinPSampling = SamplerSettings(
-    temp = 1.0,
-    top_p = 1.0,
-    min_p = 0.1,
-    top_k = -1
-)
-
-# use min_p as the only active sampler (more restrictive)
-StrictMinPSampling = SamplerSettings(
-    temp = 1.0,
-    top_p = 1.0,
-    min_p = 0.2,
-    top_k = -1
-)
-
-# https://arxiv.org/abs/2210.14140
-ContrastiveSearch = SamplerSettings(
-    temp = 0.0,
-    presence_penalty = 0.4
-)
-
-# https://arxiv.org/abs/2210.14140
-WarmContrastiveSearch = SamplerSettings(
-    temp = 0.0,
-    presence_penalty = 0.8
-)
-
-# outputs completely random tokens from vocab (useless)
-RandomSampling = SamplerSettings(
-    temp = MAX_TEMP,
-    top_p = 1.0,
-    min_p = 0.0,
-    top_k = -1
-)
-
-# default sampling with reduced temperature
-LowTempSampling = SamplerSettings(
-    temp = 0.4
-)
-
-# default sampling with increased temperature
-HighTempSampling = SamplerSettings(
-    temp = 1.2
-)
+NoSampling = SimpleSampling = SamplerSettings(top_k=None, top_p=None, min_p=None, temp=None)
+ClassicSampling = SamplerSettings(min_p=None, repeat_penalty=1.1)
+SemiSampling = SamplerSettings(top_k=80, top_p=0.975, min_p=0.025, temp=0.9)
+TikTokenSampling = SamplerSettings(temp=0.65)
+LowMinPSampling = SamplerSettings(top_k=None, top_p=None, min_p=0.01, temp=None)
+MinPSampling = SamplerSettings(top_k=None, top_p=None, min_p=0.075, temp=None)
+StrictMinPSampling = SamplerSettings(top_k=None, top_p=None, min_p=0.2, temp=None)
+ContrastiveSearch = SamplerSettings(top_k=None, top_p=None, min_p=None, temp=0.0, presence_penalty=0.6)
+WarmContrastiveSearch = SamplerSettings(top_k=None, top_p=None, min_p=None, temp=0.0, presence_penalty=1.0)
+RandomSampling = SamplerSettings(top_k=None, top_p=None, min_p=None, temp=MAX_TEMP)
+LowTempSampling = SamplerSettings(temp=0.4)
+HighTempSampling = SamplerSettings(temp=1.1)
+LowTopPSampling = SamplerSettings(top_k=None, top_p=0.98, min_p=None, temp=None)
+TopPSampling = SamplerSettings(top_k=None, top_p=0.9, min_p=None, temp=None)
+StrictTopPSampling = SamplerSettings(top_k=None, top_p=0.7, min_p=None, temp=None)
+
+# Model-specific samplers
+MidnightMiqu = SamplerSettings(top_k=None, top_p=None, min_p=0.12, temp=1.0, repeat_penalty=1.05)  # sophosympatheia/Midnight-Miqu-70B-v1.5
+Llama3 = SamplerSettings(top_k=None, top_p=0.9, min_p=None, temp=0.6)  # meta-llama/Meta-Llama-3.1-8B-Instruct
+Nemo = MistralNemo = MistralSmall = SamplerSettings(top_k=None, top_p=0.85, min_p=None, temp=0.7)  # mistralai/Mistral-Nemo-Instruct-2407