QueryConfig.py 3.9 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120
  1. import re
  2. from string import Template
  3. from embedchain.config.BaseConfig import BaseConfig
  4. DEFAULT_PROMPT = """
  5. Use the following pieces of context to answer the query at the end.
  6. If you don't know the answer, just say that you don't know, don't try to make up an answer.
  7. $context
  8. Query: $query
  9. Helpful Answer:
  10. """ # noqa:E501
  11. DEFAULT_PROMPT_WITH_HISTORY = """
  12. Use the following pieces of context to answer the query at the end.
  13. If you don't know the answer, just say that you don't know, don't try to make up an answer.
  14. I will provide you with our conversation history.
  15. $context
  16. History: $history
  17. Query: $query
  18. Helpful Answer:
  19. """ # noqa:E501
  20. DEFAULT_PROMPT_TEMPLATE = Template(DEFAULT_PROMPT)
  21. DEFAULT_PROMPT_WITH_HISTORY_TEMPLATE = Template(DEFAULT_PROMPT_WITH_HISTORY)
  22. query_re = re.compile(r"\$\{*query\}*")
  23. context_re = re.compile(r"\$\{*context\}*")
  24. history_re = re.compile(r"\$\{*history\}*")
  25. class QueryConfig(BaseConfig):
  26. """
  27. Config for the `query` method.
  28. """
  29. def __init__(
  30. self,
  31. template: Template = None,
  32. model=None,
  33. temperature=None,
  34. max_tokens=None,
  35. top_p=None,
  36. history=None,
  37. stream: bool = False,
  38. ):
  39. """
  40. Initializes the QueryConfig instance.
  41. :param template: Optional. The `Template` instance to use as a template for
  42. prompt.
  43. :param model: Optional. Controls the OpenAI model used.
  44. :param temperature: Optional. Controls the randomness of the model's output.
  45. Higher values (closer to 1) make output more random, lower values make it more
  46. deterministic.
  47. :param max_tokens: Optional. Controls how many tokens are generated.
  48. :param top_p: Optional. Controls the diversity of words. Higher values
  49. (closer to 1) make word selection more diverse, lower values make words less
  50. diverse.
  51. :param history: Optional. A list of strings to consider as history.
  52. :param stream: Optional. Control if response is streamed back to user
  53. :raises ValueError: If the template is not valid as template should
  54. contain $context and $query (and optionally $history).
  55. """
  56. if not history:
  57. self.history = None
  58. else:
  59. if len(history) == 0:
  60. self.history = None
  61. else:
  62. self.history = history
  63. if template is None:
  64. if self.history is None:
  65. template = DEFAULT_PROMPT_TEMPLATE
  66. else:
  67. template = DEFAULT_PROMPT_WITH_HISTORY_TEMPLATE
  68. self.temperature = temperature if temperature else 0
  69. self.max_tokens = max_tokens if max_tokens else 1000
  70. self.model = model if model else "gpt-3.5-turbo-0613"
  71. self.top_p = top_p if top_p else 1
  72. if self.validate_template(template):
  73. self.template = template
  74. else:
  75. if self.history is None:
  76. raise ValueError("`template` should have `query` and `context` keys")
  77. else:
  78. raise ValueError(
  79. "`template` should have `query`, `context` and `history` keys"
  80. )
  81. if not isinstance(stream, bool):
  82. raise ValueError("`stream` should be bool")
  83. self.stream = stream
  84. def validate_template(self, template: Template):
  85. """
  86. validate the template
  87. :param template: the template to validate
  88. :return: Boolean, valid (true) or invalid (false)
  89. """
  90. if self.history is None:
  91. return re.search(query_re, template.template) and re.search(
  92. context_re, template.template
  93. )
  94. else:
  95. return (
  96. re.search(query_re, template.template)
  97. and re.search(context_re, template.template)
  98. and re.search(history_re, template.template)
  99. )