tocmo0nlord/axolotl
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

optionally be able to specify alpaca or chat style prompts

Browse Source
This commit is contained in:
Wing Lian
2023-05-20 18:16:22 -04:00
parent fa8bd14be4
commit 1d5ab84486
6 changed files with 223 additions and 53 deletions
Download Patch File Download Diff File Expand all files Collapse all files

72
src/axolotl/prompters.py
View File

@@ -1,15 +1,34 @@
import copy
import dataclasses
import logging
from enum import auto, Enum
from typing import List, Tuple, Any, Union, Generator
IGNORE_TOKEN_ID = -100
class PromptStyle(Enum):
instruct = "instruct"
chat = "chat"
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:"
system_prompt = "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"
system_no_input_prompt = "Below is an instruction that describes a task. Write a response that appropriately completes the request.\n\n"
prompt_style = None
def __init__(self, prompt_style="instruct"):
self.prompt_style = prompt_style
self.match_prompt_style()
def match_prompt_style(self):
if self.prompt_style == PromptStyle.instruct.value:
self.prompt_input = self.system_prompt + "### Instruction:\n{instruction}\n\n### Input:\n{input}\n\n### Response:\n"
self.prompt_no_input = self.system_no_input_prompt + "### Instruction:\n{instruction}\n\n### Response:\n"
self.response_split = "### Response:"
if self.prompt_style == PromptStyle.chat.value:
self.prompt_input = self.system_prompt + "USER: {instruction}\n{input}\nASSISTANT:"
self.prompt_no_input = self.system_no_input_prompt + "USER: {instruction}\nASSISTANT:"
self.response_split = "ASSISTANT:"
def build_prompt(
self,
@@ -36,7 +55,7 @@ class JeopardyPrompter(AlpacaPrompter):
class MultipleChoiceExplainPrompter(AlpacaPrompter):
prompt_input = "Choose the answer that best answers the question. Explain your reasoning.\n\n### Question:\n{instruction}\n\n### Choices:\n{input}\n\n### Response:\n"
system_prompt = "Choose the answer that best answers the question. Explain your reasoning."
class MultipleChoiceConcisePrompter(AlpacaPrompter):
@@ -64,11 +83,30 @@ 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}"
system_prompt = "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"
system_no_input_prompt = "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"
prompt_input = "### Instruction:\n{instruction}\n\n### Input:\n{input}\n\n### Response:\n"
prompt_no_input = "### Instruction:\n{instruction}\n\n### Response:\n"
agent_label = "### Thought:\n{output}\n\n### Agent Reflection:\n{reflection}\n\n### Final Response:\n{corrected}"
response_split = "### Response:"
def __init__(self, prompt_style="instruct"):
self.prompt_style = prompt_style
self.match_prompt_style()
def match_prompt_style(self):
if self.prompt_style == PromptStyle.instruct.value:
self.prompt_input = self.system_prompt + "### Instruction:\n{instruction}\n\n### Input:\n{input}\n\n### Response:\n"
self.prompt_no_input = self.system_no_input_prompt + "### Instruction:\n{instruction}\n\n### Response:\n"
self.agent_label = "### Thought:\n{output}\n\n### Agent Reflection:\n{reflection}\n\n### Final Response:\n{corrected}"
self.response_split = "### Final Response:"
if self.prompt_style == PromptStyle.chat.value:
self.prompt_input = self.system_prompt + "USER: {instruction}\n{input}\nASSISTANT:"
self.prompt_no_input = self.system_no_input_prompt + "USER: {instruction}\nASSISTANT:"
self.agent_label = "\nTHOUGHT: {output}\nASSISTANT REFLECTION: {reflection}\nASSISTANT:"
self.response_split = "ASSISTANT:"
def build_prompt(
self,
instruction: str,
@@ -118,13 +156,13 @@ class Conversation:
def get_prompt(self) -> Generator[str, None, None]:
seps = [self.sep, self.sep2]
preamble = self.system + seps[0]
yield preamble
for i, (role, message) in enumerate(self.messages):
if message:
yield preamble + role + ": " + message + seps[i % 2]
yield (role + ":", " " + message)
else:
yield role + ":"
if i == 0:
preamble = ""
logging.warning("role with empty message: " + role)
yield (role + ":", )
def copy(self):
return Conversation(
@@ -154,7 +192,17 @@ conv_vicuna_v1_1 = Conversation(
class ShareGPTPrompter:
def build_prompt(self, source, tokenizer, sequence_len=2048) -> Generator[str, None, None]:
def __init__(self, prompt_style=None):
if prompt_style != PromptStyle.chat.value:
raise Exception(f"unsupported prompt_style for ShareGPTPrompter({prompt_style})")
# def match_prompt_style(self):
# if self.prompt_style == PromptStyle.chat.value:
# self.prompt_input = self.system_prompt + "USER: {instruction}\n{input}\nASSISTANT:"
# self.prompt_no_input = self.system_no_input_prompt + "USER: {instruction}\nASSISTANT:"
# self.response_split = "ASSISTANT:"
def build_prompt(self, source, *args, **kwargs) -> Generator[str, None, None]:
# ignore the system prompt if provided
if source[0]["from"] == "system":
source.pop(0)