File size: 9,079 Bytes
42da440 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 | import re
import os
import traceback
import logging
import yaml
from augmentoolkit.generation_functions.safe_formatter import safe_format
class OutputProcessorError(Exception):
"""Exception raised when an output processor function fails.
This error is used to provide more context when the output processing
function encounters an error during execution.
"""
class GenerationStep:
def __init__(
self,
prompt_path="", # relative to the Inputs directory
regex=re.compile(r".*", re.DOTALL), # take whole completion
sampling_params={
"temperature": 1,
"top_p": 1,
"max_tokens": 3000,
"stop": [
"### Response",
"\n\n\n\n\n",
"</s>",
"# Input:",
"[INST]",
"### Instruction",
"### Information",
"## Information",
"## Instruction",
"Name:",
"<|eot_id|>",
"<|start_header_id|>",
"<|end_header_id|>",
],
},
completion_mode=True, # Chat vs completion mode
retries=0,
engine_wrapper=None,
logging_level=logging.INFO, # Default logging level
output_processor=lambda x: x, # to ensure that control flow does not need to have decision code handling the outputs of the LLM, you can pass in a function to handle and modify the outputs (post regex) here. By default it's just the identity function and does nothing.
return_input_too=True,
default_prompt_folder="prompts",
prompt_folder="prompts",
use_stop=True,
messages=None,
):
self.prompt_path = prompt_path
self.regex = regex
self.sampling_params = sampling_params
if not use_stop:
if "stop" in self.sampling_params:
del self.sampling_params["stop"]
self.completion_mode = completion_mode
self.retries = retries
self.logging_level = logging_level
self.output_processor = output_processor
self.return_input_too = return_input_too
if not engine_wrapper:
raise Exception("Engine wrapper not passed in!")
self.engine_wrapper = engine_wrapper
self.prompt_folder = prompt_folder
self.default_prompt_folder = default_prompt_folder
logging.basicConfig(
level=self.logging_level, format="%(asctime)s - %(levelname)s - %(message)s"
)
self.messages = messages
async def generate(self, **kwargs):
# Current file directory
current_dir = os.path.dirname(os.path.abspath(__file__))
if not self.messages:
# Get the full path of the prompt file
ideal_path = os.path.join(
current_dir, "..", "..", self.prompt_folder, self.prompt_path
)
if os.path.exists(ideal_path):
full_prompt_path = ideal_path
else:
full_prompt_path = os.path.join(
current_dir,
"..",
"..",
self.default_prompt_folder,
self.prompt_path,
)
with open(full_prompt_path, "r", encoding="utf-8", errors="ignore") as pf:
prompt = pf.read()
# Submit generation and return response, retrying as needed
times_tried = 0
if self.completion_mode:
prompt_formatted = safe_format(prompt, **kwargs)
while times_tried <= self.retries:
try:
response, timeout = await self.engine_wrapper.submit_completion(
prompt_formatted, self.sampling_params
)
filtered_response = re.search(self.regex, response).group(1)
try:
ret = self.output_processor(filtered_response)
except Exception as e:
traceback.print_exc()
print(f"Output processor failed to execute on output: {e}")
raise OutputProcessorError()
if self.return_input_too:
return (
ret,
prompt_formatted + filtered_response,
filtered_response,
prompt_formatted,
)
return ret, timeout
except Exception as e:
# logging.error(f"Error in Generation Step: {e}")
try:
if not self.engine_wrapper.mode == "llamacpp":
print("Response:")
print(response)
except Exception as e:
print(f"Error {e}")
traceback.print_exc()
pass
traceback.print_exc()
times_tried += 1
raise Exception("Generation step failed -- too many retries!")
else:
if not self.messages:
messages = yaml.safe_load(prompt)
else:
messages = self.messages
new_messages = []
for message in messages:
try:
new_messages.append(
{
"role": message["role"],
"content": safe_format(message["content"], **kwargs),
}
)
except Exception as e:
new_messages.append(
{"role": message["role"], "content": message["content"]}
)
messages = new_messages
# messages = [{
# "role": message["role"],
# "content": safe_format(message["content"],**arguments)
# }
# for message in messages]
while times_tried <= self.retries:
try:
# strip whitespace added by yaml load
messages = [
{
"role": message["role"],
"content": message["content"].strip(),
}
for message in messages
]
# print("\n\n\nBEGIN DEBUG")
# print(messages)
# print("END DEBUG\n\n\n")
response, timeout = await self.engine_wrapper.submit_chat(
messages, self.sampling_params
)
try:
ret = self.output_processor(response)
except Exception as e:
traceback.print_exc()
print(f"Output processor failed to execute on output: {e}")
raise OutputProcessorError()
if self.return_input_too:
return (
ret,
yaml.dump(
messages
+ [
{
"role": "assistant",
"content": response,
"timeout": timeout,
}
],
default_flow_style=False,
allow_unicode=True,
),
response,
messages,
)
return ret, timeout
except Exception as e:
logging.error(f"Error in Generation Step: {e}")
if self.completion_mode:
print("Prompt:")
print(prompt)
else:
print("Messages:")
print(
yaml.dump(
messages, default_flow_style=False, allow_unicode=True
)
)
try:
print("\n\nResponse:\n-----")
print(response)
except UnboundLocalError:
print("No response to print")
pass
# if prompt_formatted:
# print(prompt_formatted)
logging.error(
f"Above prompt resulted in error, probably the model's fault: {e}"
)
traceback.print_exc()
times_tried += 1
raise Exception("Generation step failed -- too many retries!")
|