Skip to content

Build AI methods with your coding agent, then run them anywhere: as an MCP for chatbots, as a webapp for people, or via API for your software.

Moved settings for structure prompt suffixe and instructor dumps to cogt config

Louis Choquelc8ffaf195036
diff --git a/docs/pages/configuration/config-technical/cogt-config.md b/docs/pages/configuration/config-technical/cogt-config.md
index ea0d9e368..8896096ab 100644
--- a/docs/pages/configuration/config-technical/cogt-config.md
+++ b/docs/pages/configuration/config-technical/cogt-config.md
@@ -31,6 +31,7 @@ Configuration for all Language Model interactions:
```toml
[cogt.llm_config]
default_max_images = 100 # Maximum number of images in prompts
+is_structure_prompt_enabled = false
# Job configuration
[cogt.llm_config.llm_job_config]
@@ -172,6 +173,7 @@ is_auto_setup_preset_ocr = true
[cogt.llm_config]
default_max_images = 100
+is_structure_prompt_enabled = false
[cogt.llm_config.llm_job_config]
is_streaming_enabled = false
diff --git a/pipelex/cogt/config_cogt.py b/pipelex/cogt/config_cogt.py
index 478e108e3..adcaed1c1 100644
--- a/pipelex/cogt/config_cogt.py
+++ b/pipelex/cogt/config_cogt.py
@@ -21,12 +21,15 @@ class ImgGenConfig(ConfigModel):
class InstructorConfig(ConfigModel):
is_openai_structured_output_enabled: bool
+ is_dump_kwargs_enabled: bool
+ is_dump_response_enabled: bool
+ is_dump_error_enabled: bool
class LLMConfig(ConfigModel):
instructor_config: InstructorConfig
llm_job_config: LLMJobConfig
-
+ is_structure_prompt_enabled: bool
default_max_images: int
diff --git a/pipelex/pipe_operators/img_gen/pipe_img_gen.py b/pipelex/pipe_operators/img_gen/pipe_img_gen.py
index be9036289..d4d3684c8 100644
--- a/pipelex/pipe_operators/img_gen/pipe_img_gen.py
+++ b/pipelex/pipe_operators/img_gen/pipe_img_gen.py
@@ -138,44 +138,20 @@ class PipeImgGen(PipeOperator[PipeImgGenOutput]):
default_reaction = static_validation_config.default_reaction
reactions = static_validation_config.reactions
# check that we have either an img_gen_prompt passed as attribute or as a single text input
+ nb_inputs = self.inputs.nb_inputs
if self.img_gen_prompt:
- if self.inputs.items:
- msg = "img_gen_prompt_var_name must be None if img_gen_prompt is provided"
+ if nb_inputs > 0:
+ msg = "There must be no inputs if img_gen_prompt is provided"
raise PipeDefinitionError(msg)
# we're good with the prompt provided as attribute
return
- candidate_prompt_var_names: list[str] = []
- for input_name, requirement in self.inputs.items:
- log.debug(f"Validating input '{input_name}' with concept code '{requirement.concept.code}'")
- if concept_provider.is_compatible(
- tested_concept=requirement.concept,
- wanted_concept=ConceptFactory.make_native_concept(native_concept_data=NATIVE_CONCEPTS_DATA[NativeConceptEnum.TEXT]),
- ):
- self.img_gen_prompt_var_name = input_name
- candidate_prompt_var_names.append(input_name)
- else:
- inadequate_input_concept_error = StaticValidationError(
- error_type=StaticValidationErrorType.INADEQUATE_INPUT_CONCEPT,
- domain=self.domain,
- pipe_code=self.code,
- variable_names=[input_name],
- provided_concept_code=requirement.concept.code,
- explanation="Only a text input can be provided for image gen prompt",
- )
- match reactions.get(StaticValidationErrorType.INADEQUATE_INPUT_CONCEPT, default_reaction):
- case StaticValidationReaction.IGNORE:
- pass
- case StaticValidationReaction.LOG:
- log.error(inadequate_input_concept_error.desc())
- case StaticValidationReaction.RAISE:
- raise inadequate_input_concept_error
- if len(candidate_prompt_var_names) > 1:
+ if nb_inputs > 1:
too_many_candidate_inputs_error = StaticValidationError(
error_type=StaticValidationErrorType.TOO_MANY_CANDIDATE_INPUTS,
domain=self.domain,
pipe_code=self.code,
- variable_names=candidate_prompt_var_names,
+ variable_names=self.inputs.variables,
explanation="Only one text input can be provided for image gen prompt",
)
match reactions.get(StaticValidationErrorType.TOO_MANY_CANDIDATE_INPUTS, default_reaction):
@@ -185,7 +161,7 @@ class PipeImgGen(PipeOperator[PipeImgGenOutput]):
log.error(too_many_candidate_inputs_error.desc())
case StaticValidationReaction.RAISE:
raise too_many_candidate_inputs_error
- elif len(candidate_prompt_var_names) == 0:
+ elif nb_inputs < 1:
missing_input_var_error = StaticValidationError(
error_type=StaticValidationErrorType.MISSING_INPUT_VARIABLE,
domain=self.domain,
@@ -199,8 +175,32 @@ class PipeImgGen(PipeOperator[PipeImgGenOutput]):
log.error(missing_input_var_error.desc())
case StaticValidationReaction.RAISE:
raise missing_input_var_error
+
+ # We have confirmed right above that we have exactly one input
+ # get input_name, requirement from single item in inputs
+ input_name, requirement = self.inputs.items[0]
+ log.debug(f"Validating input '{input_name}' with concept code '{requirement.concept.code}'")
+ if concept_provider.is_compatible(
+ tested_concept=requirement.concept,
+ wanted_concept=ConceptFactory.make_native_concept(native_concept_data=NATIVE_CONCEPTS_DATA[NativeConceptEnum.TEXT]),
+ ):
+ self.img_gen_prompt_var_name = input_name
else:
- self.img_gen_prompt_var_name = candidate_prompt_var_names[0]
+ inadequate_input_concept_error = StaticValidationError(
+ error_type=StaticValidationErrorType.INADEQUATE_INPUT_CONCEPT,
+ domain=self.domain,
+ pipe_code=self.code,
+ variable_names=[input_name],
+ provided_concept_code=requirement.concept.code,
+ explanation="For PipeImgGen you must provide a text input or a concept that refines text",
+ )
+ match reactions.get(StaticValidationErrorType.INADEQUATE_INPUT_CONCEPT, default_reaction):
+ case StaticValidationReaction.IGNORE:
+ pass
+ case StaticValidationReaction.LOG:
+ log.error(inadequate_input_concept_error.desc())
+ case StaticValidationReaction.RAISE:
+ raise inadequate_input_concept_error
@override
async def _run_operator_pipe(
diff --git a/pipelex/pipe_operators/llm/pipe_llm.py b/pipelex/pipe_operators/llm/pipe_llm.py
index da33d67d0..10cfab882 100644
--- a/pipelex/pipe_operators/llm/pipe_llm.py
+++ b/pipelex/pipe_operators/llm/pipe_llm.py
@@ -388,10 +388,12 @@ class PipeLLM(PipeOperator[PipeLLMOutput]):
else:
llm_prompt_2_factory = None
- output_structure_prompt: str | None = PipeLLM.get_output_structure_prompt(
- concept_string=pipe_run_params.dynamic_output_concept_code or output_concept.concept_string,
- is_with_preliminary_text=is_with_preliminary_text,
- )
+ output_structure_prompt: str | None = None
+ if get_config().cogt.llm_config.is_structure_prompt_enabled:
+ output_structure_prompt = PipeLLM.get_output_structure_prompt(
+ concept_string=pipe_run_params.dynamic_output_concept_code or output_concept.concept_string,
+ is_with_preliminary_text=is_with_preliminary_text,
+ )
llm_prompt_1_for_object = await self.llm_prompt_spec.make_llm_prompt(
output_concept_string=output_concept.concept_string,
context_provider=working_memory,
@@ -528,7 +530,6 @@ class PipeLLM(PipeOperator[PipeLLMOutput]):
@staticmethod
def get_output_structure_prompt(concept_string: str, is_with_preliminary_text: bool) -> str | None:
- return None
concept = get_required_concept(concept_string=concept_string)
class_name = concept.structure_class_name
output_class = get_class_registry().get_class(class_name)
diff --git a/pipelex/pipe_operators/ocr/pipe_ocr.py b/pipelex/pipe_operators/ocr/pipe_ocr.py
index 1d9838d8e..9b2c94c47 100644
--- a/pipelex/pipe_operators/ocr/pipe_ocr.py
+++ b/pipelex/pipe_operators/ocr/pipe_ocr.py
@@ -111,8 +111,6 @@ class PipeOcr(PipeOperator[PipeOcrOutput]):
# We have confirmed right above that we have exactly one input
# get input_name, requirement from single item in inputs
input_name, requirement = self.inputs.items[0]
- log.debug(f"{input_name=}")
- log.debug(f"{requirement=}")
log.debug(f"Validating input '{input_name}' with concept code '{requirement.concept.code}'")
if concept_provider.is_compatible(
tested_concept=requirement.concept,
@@ -133,7 +131,7 @@ class PipeOcr(PipeOperator[PipeOcrOutput]):
pipe_code=self.code,
variable_names=[input_name],
provided_concept_code=requirement.concept.code,
- explanation="For OCR you must provide either a pdf or an image or a concept that refines one of them",
+ explanation="For PipeOcr you must provide either a pdf or an image or a concept that refines one of them",
)
match reactions.get(StaticValidationErrorType.INADEQUATE_INPUT_CONCEPT, default_reaction):
case StaticValidationReaction.IGNORE:
diff --git a/pipelex/pipelex.toml b/pipelex/pipelex.toml
index 352b8b794..3df6c9fb3 100644
--- a/pipelex/pipelex.toml
+++ b/pipelex/pipelex.toml
@@ -87,9 +87,13 @@ is_auto_setup_preset_ocr = true
[cogt.llm_config]
default_max_images = 100
+is_structure_prompt_enabled = false
[cogt.llm_config.instructor_config]
is_openai_structured_output_enabled = false
+is_dump_kwargs_enabled = true
+is_dump_response_enabled = false
+is_dump_error_enabled = false
[cogt.llm_config.llm_job_config]
max_retries = 3
@@ -234,4 +238,3 @@ structure_field_ordering = ["type", "definition", "choices", "required"]
[pipelex.plx_config.pipes]
field_ordering = ["type", "definition", "inputs", "output"]
-
diff --git a/pipelex/plugins/anthropic/anthropic_llm_worker.py b/pipelex/plugins/anthropic/anthropic_llm_worker.py
index af0ffcf57..bc533de71 100644
--- a/pipelex/plugins/anthropic/anthropic_llm_worker.py
+++ b/pipelex/plugins/anthropic/anthropic_llm_worker.py
@@ -4,23 +4,21 @@ import instructor
from anthropic import AsyncAnthropic, AsyncAnthropicBedrock, omit
from typing_extensions import override
-from pipelex import log, pretty_print
+from pipelex import log
from pipelex.cogt.exceptions import LLMCompletionError, SdkTypeError
from pipelex.cogt.llm.llm_job import LLMJob
from pipelex.cogt.llm.llm_worker_internal_abstract import LLMWorkerInternalAbstract
from pipelex.cogt.llm.structured_output import StructureMethod
from pipelex.cogt.model_backends.model_spec import InferenceModelSpec
+from pipelex.config import get_config
from pipelex.plugins.anthropic.anthropic_exceptions import AnthropicWorkerConfigurationError
from pipelex.plugins.anthropic.anthropic_factory import AnthropicFactory, AnthropicSdkVariant
from pipelex.reporting.reporting_protocol import ReportingProtocol
+from pipelex.tools.misc.instructor_utils import dump_error, dump_kwargs, dump_response
from pipelex.tools.typing.pydantic_utils import BaseModelTypeVar
from pipelex.types import StrEnum
-def dump_kwargs(*_: Any, **kwargs: Any) -> None:
- pretty_print(kwargs, title="Instructor about to send to Anthropic")
-
-
class AnthropicExtraField(StrEnum):
CLAUDE_4_TOKENS_LIMIT = "claude_4_tokens_limit"
@@ -79,7 +77,13 @@ class AnthropicLLMWorker(LLMWorkerInternalAbstract):
else:
self.instructor_for_objects = instructor.from_anthropic(client=sdk_instance)
- self.instructor_for_objects.on(hook_name="completion:kwargs", handler=dump_kwargs)
+ instructor_config = get_config().cogt.llm_config.instructor_config
+ if instructor_config.is_dump_kwargs_enabled:
+ self.instructor_for_objects.on(hook_name="completion:kwargs", handler=dump_kwargs)
+ if instructor_config.is_dump_response_enabled:
+ self.instructor_for_objects.on(hook_name="completion:response", handler=dump_response)
+ if instructor_config.is_dump_error_enabled:
+ self.instructor_for_objects.on(hook_name="completion:error", handler=dump_error)
#########################################################
# Instance methods
diff --git a/pipelex/plugins/openai/openai_llm_worker.py b/pipelex/plugins/openai/openai_llm_worker.py
index 26907e0e0..6fd8b2b04 100644
--- a/pipelex/plugins/openai/openai_llm_worker.py
+++ b/pipelex/plugins/openai/openai_llm_worker.py
@@ -9,30 +9,20 @@ if TYPE_CHECKING:
from openai.types.chat import ChatCompletionMessage
from typing_extensions import override
-from pipelex import log, pretty_print
+from pipelex import log
from pipelex.cogt.exceptions import LLMCompletionError, LLMModelNotFoundError, SdkTypeError
from pipelex.cogt.llm.llm_job import LLMJob
from pipelex.cogt.llm.llm_worker_internal_abstract import LLMWorkerInternalAbstract
from pipelex.cogt.llm.structured_output import StructureMethod
from pipelex.cogt.model_backends.model_constraints import ModelConstraints
from pipelex.cogt.model_backends.model_spec import InferenceModelSpec
+from pipelex.config import get_config
from pipelex.plugins.openai.openai_factory import OpenAIFactory
from pipelex.reporting.reporting_protocol import ReportingProtocol
+from pipelex.tools.misc.instructor_utils import dump_error, dump_kwargs, dump_response
from pipelex.tools.typing.pydantic_utils import BaseModelTypeVar
-def dump_kwargs(*_: Any, **kwargs: Any) -> None:
- pretty_print(kwargs, title="Instructor about to send to OpenAI")
-
-
-def dump_response(response: Any) -> None:
- pretty_print(response, title="Instructor response from OpenAI")
-
-
-def dump_error(error: Exception) -> None:
- pretty_print(error, title="Instructor error from OpenAI")
-
-
class OpenAILLMWorker(LLMWorkerInternalAbstract):
def __init__(
self,
@@ -60,9 +50,13 @@ class OpenAILLMWorker(LLMWorkerInternalAbstract):
else:
self.instructor_for_objects = instructor.from_openai(client=sdk_instance)
- self.instructor_for_objects.on(hook_name="completion:kwargs", handler=dump_kwargs)
- self.instructor_for_objects.on(hook_name="completion:response", handler=dump_response)
- self.instructor_for_objects.on(hook_name="completion:error", handler=dump_error)
+ instructor_config = get_config().cogt.llm_config.instructor_config
+ if instructor_config.is_dump_kwargs_enabled:
+ self.instructor_for_objects.on(hook_name="completion:kwargs", handler=dump_kwargs)
+ if instructor_config.is_dump_response_enabled:
+ self.instructor_for_objects.on(hook_name="completion:response", handler=dump_response)
+ if instructor_config.is_dump_error_enabled:
+ self.instructor_for_objects.on(hook_name="completion:error", handler=dump_error)
#########################################################
@override
diff --git a/pipelex/tools/misc/instructor_utils.py b/pipelex/tools/misc/instructor_utils.py
new file mode 100644
index 000000000..b64340270
--- /dev/null
+++ b/pipelex/tools/misc/instructor_utils.py
@@ -0,0 +1,15 @@
+from typing import Any
+
+from pipelex import pretty_print
+
+
+def dump_kwargs(*_: Any, **kwargs: Any) -> None:
+ pretty_print(kwargs, title="Instructor about to send to LLM provider")
+
+
+def dump_response(response: Any) -> None:
+ pretty_print(response, title="Instructor response from LLM provider")
+
+
+def dump_error(error: Exception) -> None:
+ pretty_print(error, title="Instructor error from LLM provider")