Skip to content

Prompt stack

PromptStack

Bases: SerializableMixin

Source code in griptape/utils/prompt_stack.py 
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
@define
class PromptStack(SerializableMixin):
    GENERIC_ROLE = "generic"
    USER_ROLE = "user"
    ASSISTANT_ROLE = "assistant"
    SYSTEM_ROLE = "system"

    @define
    class Input(SerializableMixin):
        content: str = field(metadata={"serializable": True})
        role: str = field(metadata={"serializable": True})

        def is_generic(self) -> bool:
            return self.role == PromptStack.GENERIC_ROLE

        def is_system(self) -> bool:
            return self.role == PromptStack.SYSTEM_ROLE

        def is_user(self) -> bool:
            return self.role == PromptStack.USER_ROLE

        def is_assistant(self) -> bool:
            return self.role == PromptStack.ASSISTANT_ROLE

    inputs: list[Input] = field(factory=list, kw_only=True, metadata={"serializable": True})

    def add_input(self, content: str, role: str) -> Input:
        self.inputs.append(self.Input(content=content, role=role))

        return self.inputs[-1]

    def add_generic_input(self, content: str) -> Input:
        return self.add_input(content, self.GENERIC_ROLE)

    def add_system_input(self, content: str) -> Input:
        return self.add_input(content, self.SYSTEM_ROLE)

    def add_user_input(self, content: str) -> Input:
        return self.add_input(content, self.USER_ROLE)

    def add_assistant_input(self, content: str) -> Input:
        return self.add_input(content, self.ASSISTANT_ROLE)

    def add_conversation_memory(self, memory: BaseConversationMemory, index: Optional[int] = None) -> list[Input]:
        """Add the Conversation Memory runs to the Prompt Stack.

        If autoprune is enabled, this will fit as many Conversation Memory runs into the Prompt Stack
        as possible without exceeding the token limit.

        Args:
            memory: The Conversation Memory to add the Prompt Stack to.
            index: Optional index to insert the Conversation Memory runs at.
                   Defaults to appending to the end of the Prompt Stack.
        """
        num_runs_to_fit_in_prompt = len(memory.runs)

        if memory.autoprune and hasattr(memory, "structure"):
            should_prune = True
            prompt_driver = memory.structure.config.global_drivers.prompt_driver
            temp_stack = PromptStack()

            # Try to determine how many Conversation Memory runs we can
            # fit into the Prompt Stack without exceeding the token limit.
            while should_prune and num_runs_to_fit_in_prompt > 0:
                temp_stack.inputs = self.inputs.copy()

                # Add n runs from Conversation Memory.
                # Where we insert into the Prompt Stack doesn't matter here
                # since we only care about the total token count.
                memory_inputs = memory.to_prompt_stack(num_runs_to_fit_in_prompt).inputs
                temp_stack.inputs.extend(memory_inputs)

                # Convert the prompt stack into tokens left.
                prompt_string = prompt_driver.prompt_stack_to_string(temp_stack)
                tokens_left = prompt_driver.tokenizer.count_input_tokens_left(prompt_string)
                if tokens_left > 0:
                    # There are still tokens left, no need to prune.
                    should_prune = False
                else:
                    # There were not any tokens left, prune one run and try again.
                    num_runs_to_fit_in_prompt -= 1

        if num_runs_to_fit_in_prompt:
            memory_inputs = memory.to_prompt_stack(num_runs_to_fit_in_prompt).inputs
            if index:
                self.inputs[index:index] = memory_inputs
            else:
                self.inputs.extend(memory_inputs)
        return self.inputs

ASSISTANT_ROLE = 'assistant' class-attribute instance-attribute

GENERIC_ROLE = 'generic' class-attribute instance-attribute

SYSTEM_ROLE = 'system' class-attribute instance-attribute

USER_ROLE = 'user' class-attribute instance-attribute

inputs: list[Input] = field(factory=list, kw_only=True, metadata={'serializable': True}) class-attribute instance-attribute

Input

Bases: SerializableMixin

Source code in griptape/utils/prompt_stack.py 
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
@define
class Input(SerializableMixin):
    content: str = field(metadata={"serializable": True})
    role: str = field(metadata={"serializable": True})

    def is_generic(self) -> bool:
        return self.role == PromptStack.GENERIC_ROLE

    def is_system(self) -> bool:
        return self.role == PromptStack.SYSTEM_ROLE

    def is_user(self) -> bool:
        return self.role == PromptStack.USER_ROLE

    def is_assistant(self) -> bool:
        return self.role == PromptStack.ASSISTANT_ROLE
content: str = field(metadata={'serializable': True}) class-attribute instance-attribute
role: str = field(metadata={'serializable': True}) class-attribute instance-attribute
is_assistant()
Source code in griptape/utils/prompt_stack.py 
32
33
def is_assistant(self) -> bool:
    return self.role == PromptStack.ASSISTANT_ROLE
is_generic()
Source code in griptape/utils/prompt_stack.py 
23
24
def is_generic(self) -> bool:
    return self.role == PromptStack.GENERIC_ROLE
is_system()
Source code in griptape/utils/prompt_stack.py 
26
27
def is_system(self) -> bool:
    return self.role == PromptStack.SYSTEM_ROLE
is_user()
Source code in griptape/utils/prompt_stack.py 
29
30
def is_user(self) -> bool:
    return self.role == PromptStack.USER_ROLE

add_assistant_input(content)

Source code in griptape/utils/prompt_stack.py 
51
52
def add_assistant_input(self, content: str) -> Input:
    return self.add_input(content, self.ASSISTANT_ROLE)

add_conversation_memory(memory, index=None)

Add the Conversation Memory runs to the Prompt Stack.

If autoprune is enabled, this will fit as many Conversation Memory runs into the Prompt Stack as possible without exceeding the token limit.

Parameters:

Name Type Description Default
memory BaseConversationMemory

The Conversation Memory to add the Prompt Stack to.

 required
index Optional[int]

Optional index to insert the Conversation Memory runs at. Defaults to appending to the end of the Prompt Stack.

 None
Source code in griptape/utils/prompt_stack.py 
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
def add_conversation_memory(self, memory: BaseConversationMemory, index: Optional[int] = None) -> list[Input]:
    """Add the Conversation Memory runs to the Prompt Stack.

    If autoprune is enabled, this will fit as many Conversation Memory runs into the Prompt Stack
    as possible without exceeding the token limit.

    Args:
        memory: The Conversation Memory to add the Prompt Stack to.
        index: Optional index to insert the Conversation Memory runs at.
               Defaults to appending to the end of the Prompt Stack.
    """
    num_runs_to_fit_in_prompt = len(memory.runs)

    if memory.autoprune and hasattr(memory, "structure"):
        should_prune = True
        prompt_driver = memory.structure.config.global_drivers.prompt_driver
        temp_stack = PromptStack()

        # Try to determine how many Conversation Memory runs we can
        # fit into the Prompt Stack without exceeding the token limit.
        while should_prune and num_runs_to_fit_in_prompt > 0:
            temp_stack.inputs = self.inputs.copy()

            # Add n runs from Conversation Memory.
            # Where we insert into the Prompt Stack doesn't matter here
            # since we only care about the total token count.
            memory_inputs = memory.to_prompt_stack(num_runs_to_fit_in_prompt).inputs
            temp_stack.inputs.extend(memory_inputs)

            # Convert the prompt stack into tokens left.
            prompt_string = prompt_driver.prompt_stack_to_string(temp_stack)
            tokens_left = prompt_driver.tokenizer.count_input_tokens_left(prompt_string)
            if tokens_left > 0:
                # There are still tokens left, no need to prune.
                should_prune = False
            else:
                # There were not any tokens left, prune one run and try again.
                num_runs_to_fit_in_prompt -= 1

    if num_runs_to_fit_in_prompt:
        memory_inputs = memory.to_prompt_stack(num_runs_to_fit_in_prompt).inputs
        if index:
            self.inputs[index:index] = memory_inputs
        else:
            self.inputs.extend(memory_inputs)
    return self.inputs

add_generic_input(content)

Source code in griptape/utils/prompt_stack.py 
42
43
def add_generic_input(self, content: str) -> Input:
    return self.add_input(content, self.GENERIC_ROLE)

add_input(content, role)

Source code in griptape/utils/prompt_stack.py 
37
38
39
40
def add_input(self, content: str, role: str) -> Input:
    self.inputs.append(self.Input(content=content, role=role))

    return self.inputs[-1]

add_system_input(content)

Source code in griptape/utils/prompt_stack.py 
45
46
def add_system_input(self, content: str) -> Input:
    return self.add_input(content, self.SYSTEM_ROLE)

add_user_input(content)

Source code in griptape/utils/prompt_stack.py 
48
49
def add_user_input(self, content: str) -> Input:
    return self.add_input(content, self.USER_ROLE)