6 Commits

Author SHA1 Message Date
juk0de 6672e5ee3a translation: added check for valid document format when using OpenAI 2023-11-08 20:11:15 +01:00
juk0de 75314cd777 translation: speficied / implemented the question format for OpenAI based translations 2023-11-08 19:38:37 +01:00
juk0de 44f7ebe365 translation: some small required refactoring 2023-11-08 18:20:22 +01:00
juk0de 7b8ee56230 added new command 'translation' 2023-11-07 17:35:44 +01:00
Oleksandr Kozachuk 75a123eb72 Fix usage of the dynamic answer is some cases. 2023-10-24 12:59:13 +02:00
juk0de 7c1c67f8ff Merge pull request 'Dynamic Answer class and OpenAI streaming API' (#19) from dynamic_answer into main
Introduces several changes with the main objective of enabling OpenAI's streaming API in the chatmastermind application. This allows for the retrieval of AI responses gradually as a stream, which can significantly improve the user experience in interactions that involve large result sets.

* Added tiktoken import in 'openai.py' and modifications to the OpenAI class to support streaming. This includes the addition of a new class OpenAIAnswer to handle streaming API responses.
* Modified request function in the OpenAI class: the stream=True flag is added to the openai.ChatCompletion.create method to enable streaming API.
* Modified 'question.py' to print the answer parts as they are streamed.
* Replaced the Answer class's string data type with a generator which supports str and Generator[str, None, None] data types. Modifications are made to the Answer class methods to handle both data types accordingly.
* Updated the tests in 'test_ais_openai.py' and 'test_message.py' to reflect and validate these changes.
2023-10-21 15:50:45 +02:00
5 changed files with 194 additions and 60 deletions
+1 -1
View File
@@ -150,7 +150,7 @@ class OpenAI(AI):
for message in chat.messages:
if message.answer:
prompt_tokens += append('user', message.question)
prompt_tokens += append('assistant', message.answer)
prompt_tokens += append('assistant', str(message.answer))
if question:
prompt_tokens += append('user', question.question)
return oai_chat, prompt_tokens
+69
View File
@@ -0,0 +1,69 @@
"""
Contains shared functions for the various CMM subcommands.
"""
import argparse
from pathlib import Path
from ..message import Message, MessageError, source_code
def read_text_file(file: Path) -> str:
with open(file) as r:
content = r.read().strip()
return content
def add_file_as_text(question_parts: list[str], file: str) -> None:
"""
Add the given file as plain text to the question part list.
If the file is a Message, add the answer.
"""
file_path = Path(file)
content: str
try:
message = Message.from_file(file_path)
if message and message.answer:
content = message.answer
except MessageError:
content = read_text_file(Path(file))
if len(content) > 0:
question_parts.append(content)
def add_file_as_code(question_parts: list[str], file: str) -> None:
"""
Add all source code from the given file. If no code segments can be extracted,
the whole content is added as source code segment. If the file is a Message,
extract the source code from the answer.
"""
file_path = Path(file)
content: str
try:
message = Message.from_file(file_path)
if message and message.answer:
content = message.answer
except MessageError:
with open(file) as r:
content = r.read().strip()
# extract and add source code
code_parts = source_code(content, include_delims=True)
if len(code_parts) > 0:
question_parts += code_parts
else:
question_parts.append(f"```\n{content}\n```")
def invert_input_tag_args(args: argparse.Namespace) -> None:
"""
Changes the semantics of the INPUT tags for this command:
* not tags specified on the CLI -> no tags are selected
* empty tags specified on the CLI -> all tags are selected
"""
if args.or_tags is None:
args.or_tags = set()
elif len(args.or_tags) == 0:
args.or_tags = None
if args.and_tags is None:
args.and_tags = set()
elif len(args.and_tags) == 0:
args.and_tags = None
+2 -58
View File
@@ -3,9 +3,10 @@ import argparse
from pathlib import Path
from itertools import zip_longest
from copy import deepcopy
from .common import invert_input_tag_args, add_file_as_code, add_file_as_text
from ..configuration import Config
from ..chat import ChatDB, msg_location
from ..message import Message, MessageFilter, MessageError, Question, source_code
from ..message import Message, MessageFilter, Question
from ..ai_factory import create_ai
from ..ai import AI, AIResponse
@@ -14,47 +15,6 @@ class QuestionCmdError(Exception):
pass
def add_file_as_text(question_parts: list[str], file: str) -> None:
"""
Add the given file as plain text to the question part list.
If the file is a Message, add the answer.
"""
file_path = Path(file)
content: str
try:
message = Message.from_file(file_path)
if message and message.answer:
content = message.answer
except MessageError:
with open(file) as r:
content = r.read().strip()
if len(content) > 0:
question_parts.append(content)
def add_file_as_code(question_parts: list[str], file: str) -> None:
"""
Add all source code from the given file. If no code segments can be extracted,
the whole content is added as source code segment. If the file is a Message,
extract the source code from the answer.
"""
file_path = Path(file)
content: str
try:
message = Message.from_file(file_path)
if message and message.answer:
content = message.answer
except MessageError:
with open(file) as r:
content = r.read().strip()
# extract and add source code
code_parts = source_code(content, include_delims=True)
if len(code_parts) > 0:
question_parts += code_parts
else:
question_parts.append(f"```\n{content}\n```")
def create_msg_args(msg: Message, args: argparse.Namespace) -> argparse.Namespace:
"""
Takes an existing message and CLI arguments, and returns modified args based
@@ -163,22 +123,6 @@ def repeat_messages(messages: list[Message], chat: ChatDB, args: argparse.Namesp
make_request(ai, chat, message, msg_args)
def invert_input_tag_args(args: argparse.Namespace) -> None:
"""
Changes the semantics of the INPUT tags for this command:
* not tags specified on the CLI -> no tags are selected
* empty tags specified on the CLI -> all tags are selected
"""
if args.or_tags is None:
args.or_tags = set()
elif len(args.or_tags) == 0:
args.or_tags = None
if args.and_tags is None:
args.and_tags = set()
elif len(args.and_tags) == 0:
args.and_tags = None
def question_cmd(args: argparse.Namespace, config: Config) -> None:
"""
Handler for the 'question' command.
+105
View File
@@ -0,0 +1,105 @@
import argparse
import mimetypes
from pathlib import Path
from .common import invert_input_tag_args, read_text_file
from ..configuration import Config
from ..message import MessageFilter, Message, Question
from ..chat import ChatDB, msg_location
class TranslationCmdError(Exception):
pass
text_separator: str = 'TEXT:'
def assert_document_type_supported_openai(document_file: Path) -> None:
doctype = mimetypes.guess_type(document_file)
if doctype != 'text/plain':
raise TranslationCmdError("AI 'OpenAI' only supports document type 'text/plain''")
def translation_prompt_openai(source_lang: str, target_lang: str) -> str:
"""
Return the prompt for GPT that tells it to do the translation.
"""
return f"Translate the text below the line {text_separator} from {source_lang} to {target_lang}."
def create_message_openai(chat: ChatDB, args: argparse.Namespace) -> Message:
"""
Create a new message from the given arguments and write it to the cache directory.
Message format
1. Translation prompt (tells GPT to do a translation)
2. Glossary (if specified as an argument)
3. User provided prompt enhancements
4. Translation separator
5. User provided text to be translated
The text to be translated is determined as a follows:
- if a document is provided in the arguments, translate its content
- if no document is provided, translate the last text argument
The other text arguments will be put into the "header" and can be used
to improve the translation prompt.
"""
text_args: list[str] = []
if args.create is not None:
text_args = args.create
elif args.ask is not None:
text_args = args.ask
else:
raise TranslationCmdError("No input text found")
# extract user prompt and user text to be translated
user_text: str
user_prompt: str
if args.input_document is not None:
assert_document_type_supported_openai(Path(args.input_document))
user_text = read_text_file(Path(args.input_document))
user_prompt = '\n\n'.join([str(s) for s in text_args])
else:
user_text = text_args[-1]
user_prompt = '\n\n'.join([str(s) for s in text_args[:-1]])
# build full question string
# FIXME: add glossaries if given
question_text: str = '\n\n'.join([translation_prompt_openai(args.source_lang, args.target_lang),
user_prompt,
text_separator,
user_text])
# create and write the message
message = Message(question=Question(question_text),
tags=args.output_tags,
ai=args.AI,
model=args.model)
# only write the new message to the cache,
# don't add it to the internal list
chat.cache_write([message])
return message
def translation_cmd(args: argparse.Namespace, config: Config) -> None:
"""
Handler for the 'translation' command. Creates and executes translation
requests based on the input and selected AI. Depending on the AI, the
whole process may be significantly different (e.g. DeepL vs OpenAI).
"""
invert_input_tag_args(args)
mfilter = MessageFilter(tags_or=args.or_tags,
tags_and=args.and_tags,
tags_not=args.exclude_tags)
chat = ChatDB.from_dir(cache_path=Path(config.cache),
db_path=Path(config.db),
mfilter=mfilter,
glob=args.glob,
loc=msg_location(args.location))
# if it's a new translation, create and store it immediately
# FIXME: check AI type
if args.ask or args.create:
# message = create_message(chat, args)
create_message_openai(chat, args)
if args.create:
return
+17 -1
View File
@@ -14,6 +14,7 @@ from .commands.tags import tags_cmd
from .commands.config import config_cmd
from .commands.hist import hist_cmd
from .commands.print import print_cmd
from .commands.translation import translation_cmd
from .chat import msg_location
@@ -102,7 +103,7 @@ def create_parser() -> argparse.ArgumentParser:
# 'tags' command parser
tags_cmd_parser = cmdparser.add_parser('tags',
help="Manage tags.",
aliases=['t'])
aliases=['T'])
tags_cmd_parser.set_defaults(func=tags_cmd)
tags_group = tags_cmd_parser.add_mutually_exclusive_group(required=True)
tags_group.add_argument('-l', '--list', help="List all tags and their frequency",
@@ -136,6 +137,21 @@ def create_parser() -> argparse.ArgumentParser:
print_cmd_modes.add_argument('-a', '--answer', help='Only print the answer', action='store_true')
print_cmd_modes.add_argument('-S', '--only-source-code', help='Only print embedded source code', action='store_true')
# 'translation' command parser
translation_cmd_parser = cmdparser.add_parser('translation', parents=[ai_parser, tag_parser],
help="ask, create and repeat translations.",
aliases=['t'])
translation_cmd_parser.set_defaults(func=translation_cmd)
translation_group = translation_cmd_parser.add_mutually_exclusive_group(required=True)
translation_group.add_argument('-a', '--ask', nargs='+', help='Ask to translate the given text', metavar='TEXT')
translation_group.add_argument('-c', '--create', nargs='+', help='Create a translation', metavar='TEXT')
translation_group.add_argument('-r', '--repeat', nargs='*', help='Repeat a translation', metavar='MESSAGE')
translation_cmd_parser.add_argument('-S', '--source-lang', help="Source language", metavar="LANGUAGE", required=True)
translation_cmd_parser.add_argument('-T', '--target-lang', help="Target language", metavar="LANGUAGE", required=True)
translation_cmd_parser.add_argument('-G', '--glossaries', nargs='+', help="List of glossaries", metavar="GLOSSARY")
translation_cmd_parser.add_argument('-d', '--input-document', help="Document to translate", metavar="FILE")
translation_cmd_parser.add_argument('-D', '--output-document', help="Path for the translated document", metavar="FILE")
argcomplete.autocomplete(parser)
return parser