File size: 7,411 Bytes
8d959a7 5e37144 8d959a7 ce24f5e 8d959a7 5e37144 8d959a7 a6028d3 8d959a7 a6028d3 8d959a7 5e37144 8d959a7 a12fb0a cf68153 5e37144 cf68153 8d959a7 6045345 81de0ef 5e37144 81de0ef 2bc1a5b 81de0ef 5e37144 81de0ef 8d959a7 a6028d3 8d959a7 a6028d3 8d959a7 5e37144 8d959a7 5e37144 8d959a7 5e37144 8d959a7 5e37144 8d959a7 a6028d3 8d959a7 5e37144 8d959a7 ce24f5e 5e37144 8d43785 8d959a7 a6028d3 8d959a7 5e37144 |
1 2 3 4 5 6 7 8 9 10 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 |
import copy
import dataclasses
from enum import auto, Enum
from typing import List, Tuple, Any, Union, Generator
IGNORE_TOKEN_ID = -100
class AlpacaPrompter:
prompt_input = "Below is an instruction that describes a task, paired with an input that provides further context. Write a response that appropriately completes the request.\n\n### Instruction:\n{instruction}\n\n### Input:\n{input}\n\n### Response:\n"
prompt_no_input = "Below is an instruction that describes a task. Write a response that appropriately completes the request.\n\n### Instruction:\n{instruction}\n\n### Response:\n"
response_split = "### Response:"
def build_prompt(
self,
instruction: str,
input: Union[None, str] = None,
output: Union[None, str] = None,
) -> Generator[str, None, None]:
# returns the full prompt from instruction and optional input
# if a label (=response, =output) is provided, it's also appended.
if input:
res = self.prompt_input.format(instruction=instruction, input=input)
else:
res = self.prompt_no_input.format(instruction=instruction)
if output:
res = f"{res}{output}"
yield res
def get_response(self, output: str) -> str:
return output.split(self.response_split)[1].strip()
class JeopardyPrompter(AlpacaPrompter):
prompt_input = "Below is a Jeopardy clue paired with input providing the category of the clue. Write a concise response that best answers tbe clue given the category.\n\n### Instruction:\n{instruction}\n\n### Input:\n{input}\n\n### Response:\n"
class CompletionPrompter(AlpacaPrompter):
def build_prompt(self, instruction: str, input=None, output=None) -> Generator[str, None, None]:
yield instruction
def get_response(self, output: str) -> str:
return output.strip()
class GPTeacherPrompter(AlpacaPrompter):
...
class NomicGPT4AllPrompter(AlpacaPrompter):
...
class ReflectAlpacaPrompter:
prompt_input = "Below is an instruction that describes a task, paired with an input that provides further context. You, the Assistant, should generate a response as if it were an abstract for an academic or technical paper on the query along with a methodology. Then generate an Agent Reflection where you create a long form response as if from subject matter expert, be verbose, diligent, and creative in your application of knowledge, apply it through the lens of the response generated by the assistant. Look for flawed reasoning, faulty logic, or other mistakes in the method. Finally, generate a final response and method for the user with the Assistant abstract and Reflection analysis as augmentations to the generation\n\n### Instruction:\n{instruction}\n\n### Input:\n{input}\n\n### Response:\n"
prompt_no_input = "Below is an instruction that describes a task. You, the Assistant, should generate a response as if it were an abstract for an academic or technical paper on the query along with a methodology. Then generate an Agent Reflection where you create a long form response as if from subject matter expert, be verbose, diligent, and creative in your application of knowledge, apply it through the lens of the response generated by the assistant. Look for flawed reasoning, faulty logic, or other mistakes in the method. Finally, generate a final response and method for the user with the Assistant abstract and Reflection analysis as augmentations to the generation\n\n### Instruction:\n{instruction}\n\n### Response:\n"
agent_label = "{output}\n\n### Agent Reflection:\n{reflection}\n\n### Final Response:\n{corrected}"
response_split = "### Response:"
def build_prompt(
self,
instruction: str,
input: Union[None, str] = None,
output: Union[None, str] = None,
reflection: Union[None, str] = None,
corrected: Union[None, str] = None,
) -> Generator[str, None, None]:
# returns the full prompt from instruction and optional input
# if a label (=response, =output) is provided, it's also appended.
if input:
res = self.prompt_input.format(instruction=instruction, input=input)
else:
res = self.prompt_no_input.format(instruction=instruction)
if output and reflection and corrected:
label = self.agent_label.format(
output=output, reflection=reflection, corrected=corrected
)
res = f"{res}{label}"
yield res
def get_response(self, output: str) -> str:
return output.split(self.response_split)[1].strip()
class SeparatorStyle(Enum):
"""Different separator style."""
SINGLE = auto()
TWO = auto()
DOLLY = auto()
# TODO clean this 💩 up
@dataclasses.dataclass
class Conversation:
"""A class that keeps all conversation history."""
system: str
roles: List[str]
messages: List[List[str]]
offset: int
sep_style: SeparatorStyle = SeparatorStyle.SINGLE
sep: str = "###"
sep2: str = None
def get_prompt(self) -> Generator[str, None, None]:
seps = [self.sep, self.sep2]
preamble = self.system + seps[0]
for i, (role, message) in enumerate(self.messages):
if message:
yield preamble + role + ": " + message + seps[i % 2]
else:
yield role + ":"
if i == 0:
preamble = ""
def copy(self):
return Conversation(
system=self.system,
roles=self.roles,
messages=[[x, y] for x, y in self.messages],
offset=self.offset,
sep_style=self.sep_style,
sep=self.sep,
sep2=self.sep2,
)
def append_message(self, role, message):
self.messages.append([role, message])
conv_vicuna_v1_1 = Conversation(
system="A chat between a curious user and an artificial intelligence assistant. "
"The assistant gives helpful, detailed, and polite answers to the user's questions.",
roles=["USER", "ASSISTANT"],
messages=[],
offset=0,
sep_style=SeparatorStyle.TWO,
sep=" ",
sep2=" ",
)
class ShareGPTPrompter:
def build_prompt(self, source, tokenizer, sequence_len=2048) -> Generator[str, None, None]:
# ignore the system prompt if provided
if source[0]["from"] == "system":
source.pop(0)
if len(source) < 2:
# If there isn't a back and forth conversation, ignore it
# also happens on the data splitting leaving empty conversations
raise IndexError
conv = conv_vicuna_v1_1.copy()
roles = {"human": conv.roles[0], "gpt": conv.roles[1]}
try:
# Apply prompt templates
if (
source[0]["from"] not in roles
or roles[source[0]["from"]] != conv.roles[0]
):
# Skip the first one if it is not from human
source = source[1:]
except IndexError as e:
# sometimes there is a bing or system chat
raise e
conv.messages = []
for j, sentence in enumerate(source):
role = roles[sentence["from"]]
assert role == conv.roles[j % 2]
conv.append_message(role, sentence["value"])
for part in conv.get_prompt():
yield part
|