main: improved parameter descriptions

README.md

@@ -65,7 +65,7 @@ cmm question [-t OTAGS]... [-k ATAGS]... [-x XTAGS]... [-o OUTTAGS]... [-A AI_ID
 * `-O, --overwrite`: Overwrite existing messages when repeating them
 * `-s, --source-text FILE`: Add content of a file to the query
 * `-S, --source-code FILE`: Add source code file content to the chat history
-* `-l, --location {cache,db,all}`: Use given location when building the chat history (default: 'db')
+* `-l, --location {disk,cache,db,all}`: Use given location when building the chat history (default: 'db')
 * `-g, --glob GLOB`: Filter message files using the given glob pattern
 
 #### Hist
@@ -85,8 +85,6 @@ cmm hist [--print | --convert FORMAT] [-t OTAGS]... [-k ATAGS]... [-x XTAGS]...
 * `-S, --source-code-only`: Only print embedded source code
 * `-A, --answer SUBSTRING`: Filter for answer substring
 * `-Q, --question SUBSTRING`: Filter for question substring
-* `-l, --location {cache,db,all}`: Use given location when building the chat history (default: 'db')
-* `-g, --glob GLOB`: Filter message files using the given glob pattern
 
 #### Tags
 

chatmastermind/ais/openai.py

@@ -2,8 +2,7 @@
 Implements the OpenAI client classes and functions.
 """
 import openai
-import tiktoken
-from typing import Optional, Union, Generator
+from typing import Optional, Union
 from ..tags import Tag
 from ..message import Message, Answer
 from ..chat import Chat
@@ -13,52 +12,6 @@ from ..configuration import OpenAIConfig
 ChatType = list[dict[str, str]]
 
 
-class OpenAIAnswer:
-    def __init__(self,
-                 idx: int,
-                 streams: dict[int, 'OpenAIAnswer'],
-                 response: openai.ChatCompletion,
-                 tokens: Tokens,
-                 encoding: tiktoken.core.Encoding) -> None:
-        self.idx = idx
-        self.streams = streams
-        self.response = response
-        self.position: int = 0
-        self.encoding = encoding
-        self.data: list[str] = []
-        self.finished: bool = False
-        self.tokens = tokens
-
-    def stream(self) -> Generator[str, None, None]:
-        while True:
-            if not self.next():
-                continue
-            if len(self.data) <= self.position:
-                break
-            yield self.data[self.position]
-            self.position += 1
-
-    def next(self) -> bool:
-        if self.finished:
-            return True
-        try:
-            chunk = next(self.response)
-        except StopIteration:
-            self.finished = True
-        if not self.finished:
-            found_choice = False
-            for choice in chunk.choices:
-                if not choice.finish_reason:
-                    self.streams[choice.index].data.append(choice.delta.content)
-                    self.tokens.completion += len(self.encoding.encode(choice.delta.content))
-                    self.tokens.total = self.tokens.prompt + self.tokens.completion
-                if choice.index == self.idx:
-                    found_choice = True
-            if not found_choice:
-                return False
-        return True
-
-
 class OpenAI(AI):
     """
     The OpenAI AI client.
@@ -68,10 +21,7 @@ class OpenAI(AI):
         self.ID = config.ID
         self.name = config.name
         self.config = config
-        self.client = openai.OpenAI(api_key=self.config.api_key)
-
-    def _completions(self, *args, **kw):  # type: ignore
-        return self.client.chat.completions.create(*args, **kw)
+        openai.api_key = config.api_key
 
     def request(self,
                 question: Message,
@@ -83,42 +33,39 @@ class OpenAI(AI):
         chat history. The nr. of requested answers corresponds to the
         nr. of messages in the 'AIResponse'.
         """
-        self.encoding = tiktoken.encoding_for_model(self.config.model)
-        oai_chat, prompt_tokens = self.openai_chat(chat, self.config.system, question)
-        tokens: Tokens = Tokens(prompt_tokens, 0, prompt_tokens)
-        response = self._completions(
+        oai_chat = self.openai_chat(chat, self.config.system, question)
+        response = openai.ChatCompletion.create(
             model=self.config.model,
             messages=oai_chat,
             temperature=self.config.temperature,
             max_tokens=self.config.max_tokens,
             top_p=self.config.top_p,
             n=num_answers,
-            stream=True,
             frequency_penalty=self.config.frequency_penalty,
             presence_penalty=self.config.presence_penalty)
-        streams: dict[int, OpenAIAnswer] = {}
-        for n in range(num_answers):
-            streams[n] = OpenAIAnswer(n, streams, response, tokens, self.encoding)
-        question.answer = Answer(streams[0].stream())
+        question.answer = Answer(response['choices'][0]['message']['content'])
         question.tags = set(otags) if otags is not None else None
         question.ai = self.ID
         question.model = self.config.model
         answers: list[Message] = [question]
-        for idx in range(1, num_answers):
+        for choice in response['choices'][1:]:  # type: ignore
             answers.append(Message(question=question.question,
-                                   answer=Answer(streams[idx].stream()),
+                                   answer=Answer(choice['message']['content']),
                                    tags=otags,
                                    ai=self.ID,
                                    model=self.config.model))
-        return AIResponse(answers, tokens)
+        return AIResponse(answers, Tokens(response['usage']['prompt_tokens'],
+                                          response['usage']['completion_tokens'],
+                                          response['usage']['total_tokens']))
 
     def models(self) -> list[str]:
         """
         Return all models supported by this AI.
         """
         ret = []
-        for engine in sorted(self.client.models.list().data, key=lambda x: x.id):
-            ret.append(engine.id)
+        for engine in sorted(openai.Engine.list()['data'], key=lambda x: x['id']):
+            if engine['ready']:
+                ret.append(engine['id'])
         ret.sort()
         return ret
 
@@ -126,30 +73,34 @@ class OpenAI(AI):
         """
         Print all models supported by the current AI.
         """
-        for model in self.models():
-            print(model)
+        not_ready = []
+        for engine in sorted(openai.Engine.list()['data'], key=lambda x: x['id']):
+            if engine['ready']:
+                print(engine['id'])
+            else:
+                not_ready.append(engine['id'])
+        if len(not_ready) > 0:
+            print('\nNot ready: ' + ', '.join(not_ready))
 
     def openai_chat(self, chat: Chat, system: str,
-                    question: Optional[Message] = None) -> tuple[ChatType, int]:
+                    question: Optional[Message] = None) -> ChatType:
         """
         Create a chat history with system message in OpenAI format.
         Optionally append a new question.
         """
         oai_chat: ChatType = []
-        prompt_tokens: int = 0
 
-        def append(role: str, content: str) -> int:
+        def append(role: str, content: str) -> None:
             oai_chat.append({'role': role, 'content': content.replace("''", "'")})
-            return len(self.encoding.encode(', '.join(['role:', oai_chat[-1]['role'], 'content:', oai_chat[-1]['content']])))
 
-        prompt_tokens += append('system', system)
+        append('system', system)
         for message in chat.messages:
             if message.answer:
-                prompt_tokens += append('user', message.question)
-                prompt_tokens += append('assistant', str(message.answer))
+                append('user', message.question)
+                append('assistant', message.answer)
         if question:
-            prompt_tokens += append('user', question.question)
-        return oai_chat, prompt_tokens
+            append('user', question.question)
+        return oai_chat
 
     def tokens(self, data: Union[Message, Chat]) -> int:
         raise NotImplementedError

chatmastermind/chat.py

@@ -325,8 +325,6 @@ class ChatDB(Chat):
           when reading them.
         * 'loc': read messages from given location instead of 'db_path'
         """
-        if loc == msg_location.MEM:
-            raise ChatError(f"Can't build ChatDB from message location '{loc}'")
         messages: list[Message] = []
         if loc in [msg_location.DB, msg_location.DISK, msg_location.ALL]:
             messages.extend(read_dir(db_path, glob, mfilter))

chatmastermind/commands/question.py

@@ -101,7 +101,7 @@ def create_message(chat: ChatDB, args: argparse.Namespace) -> Message:
         if code_file is not None and len(code_file) > 0:
             add_file_as_code(question_parts, code_file)
 
-    full_question = '\n\n'.join([str(s) for s in question_parts])
+    full_question = '\n\n'.join(question_parts)
 
     message = Message(question=Question(full_question),
                       tags=args.output_tags,
@@ -129,16 +129,13 @@ def make_request(ai: AI, chat: ChatDB, message: Message, args: argparse.Namespac
                                       args.output_tags)
     # only write the response messages to the cache,
     # don't add them to the internal list
+    chat.cache_write(response.messages)
     for idx, msg in enumerate(response.messages):
-        print(f"=== ANSWER {idx+1} ===", flush=True)
-        if msg.answer:
-            for piece in msg.answer:
-                print(piece, end='', flush=True)
-            print()
+        print(f"=== ANSWER {idx+1} ===")
+        print(msg.answer)
     if response.tokens:
         print("===============")
         print(response.tokens)
-    chat.cache_write(response.messages)
 
 
 def repeat_messages(messages: list[Message], chat: ChatDB, args: argparse.Namespace, config: Config) -> None:

chatmastermind/main.py

@@ -67,10 +67,10 @@ def create_parser() -> argparse.ArgumentParser:
     question_group.add_argument('-r', '--repeat', nargs='*', help='Repeat a question', metavar='MESSAGE')
     question_group.add_argument('-p', '--process', nargs='*', help='Process existing questions', metavar='MESSAGE')
     question_cmd_parser.add_argument('-l', '--location',
-                                     choices=[x.value for x in msg_location if x not in [msg_location.MEM, msg_location.DISK]],
+                                     choices=[x.value for x in msg_location],
                                      default='db',
-                                     help='Use given location when building the chat history (default: \'db\')')
-    question_cmd_parser.add_argument('-g', '--glob', help='Filter message files using the given glob pattern')
+                                     help='Select message location, default is \'db\'')
+    question_cmd_parser.add_argument('-g', '--glob', help='Glob for message file names')
     question_cmd_parser.add_argument('-O', '--overwrite', help='Overwrite existing messages when repeating them',
                                      action='store_true')
     question_cmd_parser.add_argument('-s', '--source-text', nargs='+', help='Add content of a file to the query', metavar='FILE')
@@ -94,7 +94,7 @@ def create_parser() -> argparse.ArgumentParser:
     hist_cmd_parser.add_argument('-d', '--tight', help='Print without message separators', action='store_true')
     hist_cmd_parser.add_argument('-P', '--no-paging', help='Print without paging', action='store_true')
     hist_cmd_parser.add_argument('-l', '--location',
-                                 choices=[x.value for x in msg_location if x not in [msg_location.MEM, msg_location.DISK]],
+                                 choices=[x.value for x in msg_location],
                                  default='db',
                                  help='Use given location when building the chat history (default: \'db\')')
     hist_cmd_parser.add_argument('-g', '--glob', help='Filter message files using the given glob pattern')

chatmastermind/message.py

@@ -5,9 +5,7 @@ import pathlib
 import yaml
 import tempfile
 import shutil
-import io
 from typing import Type, TypeVar, ClassVar, Optional, Any, Union, Final, Literal, Iterable, Tuple
-from typing import Generator, Iterator
 from typing import get_args as typing_get_args
 from dataclasses import dataclass, asdict, field
 from .tags import Tag, TagLine, TagError, match_tags, rename_tags
@@ -51,7 +49,7 @@ def source_code(text: str, include_delims: bool = False) -> list[str]:
     code_lines: list[str] = []
     in_code_block = False
 
-    for line in str(text).split('\n'):
+    for line in text.split('\n'):
         if line.strip().startswith('```'):
             if include_delims:
                 code_lines.append(line)
@@ -144,100 +142,30 @@ class Answer(str):
     txt_header: ClassVar[str] = '==== ANSWER ===='
     yaml_key: ClassVar[str] = 'answer'
 
-    def __init__(self, data: Union[str, Generator[str, None, None]]) -> None:
-        # Indicator of whether all of data has been processed
-        self.is_exhausted: bool = False
-
-        # Initialize data
-        self.iterator: Iterator[str] = self._init_data(data)
-
-        # Set up the buffer to hold the 'Answer' content
-        self.buffer: io.StringIO = io.StringIO()
-
-    def _init_data(self, data: Union[str, Generator[str, None, None]]) -> Iterator[str]:
+    def __new__(cls: Type[AnswerInst], string: str) -> AnswerInst:
         """
-        Process input data (either a string or a string generator)
+        Make sure the answer string does not contain the header as a whole line.
         """
-        if isinstance(data, str):
-            yield data
-        else:
-            yield from data
-
-    def __str__(self) -> str:
-        """
-        Output all content when converted into a string
-        """
-        # Ensure all data has been processed
-        for _ in self:
-            pass
-        # Return the 'Answer' content
-        return self.buffer.getvalue()
-
-    def __repr__(self) -> str:
-        return repr(str(self))
-
-    def __iter__(self) -> Generator[str, None, None]:
-        """
-        Allows the object to be iterable
-        """
-        # Generate content if not all data has been processed
-        if not self.is_exhausted:
-            yield from self.generator_iter()
-        else:
-            yield self.buffer.getvalue()
-
-    def generator_iter(self) -> Generator[str, None, None]:
-        """
-        Main generator method to process data
-        """
-        for piece in self.iterator:
-            # Write to buffer and yield piece for the iterator
-            self.buffer.write(piece)
-            yield piece
-        self.is_exhausted = True  # Set the flag that all data has been processed
-        # If the header occurs in the 'Answer' content, raise an error
-        if f'\n{self.txt_header}' in self.buffer.getvalue() or self.buffer.getvalue().startswith(self.txt_header):
-            raise MessageError(f"Answer {repr(self.buffer.getvalue())} contains the header {repr(Answer.txt_header)}")
-
-    def __eq__(self, other: object) -> bool:
-        """
-        Comparing the object to a string or another object
-        """
-        if isinstance(other, str):
-            return str(self) == other  # Compare the string value of this object to the other string
-        # Default behavior for comparing non-string objects
-        return super().__eq__(other)
-
-    def __hash__(self) -> int:
-        """
-        Generate a hash for the object based on its string representation.
-        """
-        return hash(str(self))
-
-    def __format__(self, format_spec: str) -> str:
-        """
-        Return a formatted version of the string as per the format specification.
-        """
-        return str(self).__format__(format_spec)
+        if cls.txt_header in string.split('\n'):
+            raise MessageError(f"Answer '{string}' contains the header '{cls.txt_header}'")
+        instance = super().__new__(cls, string)
+        return instance
 
     @classmethod
     def from_list(cls: Type[AnswerInst], strings: list[str]) -> AnswerInst:
         """
-        Build Answer from a list of strings. Make sure strings do not contain the header.
+        Build Question from a list of strings. Make sure strings do not contain the header.
         """
-        def _gen() -> Generator[str, None, None]:
-            if len(strings) > 0:
-                yield strings[0]
-                for s in strings[1:]:
-                    yield '\n'
-                    yield s
-        return cls(_gen())
+        if cls.txt_header in strings:
+            raise MessageError(f"Question contains the header '{cls.txt_header}'")
+        instance = super().__new__(cls, '\n'.join(strings).strip())
+        return instance
 
     def source_code(self, include_delims: bool = False) -> list[str]:
         """
         Extract and return all source code sections.
         """
-        return source_code(str(self), include_delims)
+        return source_code(self, include_delims)
 
 
 class Question(str):
@@ -513,7 +441,7 @@ class Message():
         output.append(self.question)
         if self.answer:
             output.append(Answer.txt_header)
-            output.append(str(self.answer))
+            output.append(self.answer)
         return '\n'.join(output)
 
     def to_file(self, file_path: Optional[pathlib.Path]=None, mformat: MessageFormat = message_default_format) -> None:  # noqa: 11
@@ -563,7 +491,7 @@ class Message():
                 temp_fd.write(f'{ModelLine.from_model(self.model)}\n')
             temp_fd.write(f'{Question.txt_header}\n{self.question}\n')
             if self.answer:
-                temp_fd.write(f'{Answer.txt_header}\n{str(self.answer)}\n')
+                temp_fd.write(f'{Answer.txt_header}\n{self.answer}\n')
         shutil.move(temp_file_path, file_path)
 
     def __to_file_yaml(self, file_path: pathlib.Path) -> None:
@@ -632,7 +560,7 @@ class Message():
            or (mfilter.ai and (not self.ai or mfilter.ai != self.ai))  # noqa: W503
            or (mfilter.model and (not self.model or mfilter.model != self.model))  # noqa: W503
            or (mfilter.question_contains and mfilter.question_contains not in self.question)  # noqa: W503
-           or (mfilter.answer_contains and (not self.answer or mfilter.answer_contains not in str(self.answer)))  # noqa: W503
+           or (mfilter.answer_contains and (not self.answer or mfilter.answer_contains not in self.answer))  # noqa: W503
            or (mfilter.answer_state == 'available' and not self.answer)  # noqa: W503
            or (mfilter.ai_state == 'available' and not self.ai)  # noqa: W503
            or (mfilter.model_state == 'available' and not self.model)  # noqa: W503

requirements.txt

@@ -2,4 +2,3 @@ openai
 PyYAML
 argcomplete
 pytest
-tiktoken

tests/test_ais_openai.py

@@ -9,32 +9,33 @@ from chatmastermind.configuration import OpenAIConfig
 
 class OpenAITest(unittest.TestCase):
 
-    @mock.patch('chatmastermind.ais.openai.OpenAI._completions')
+    @mock.patch('openai.ChatCompletion.create')
     def test_request(self, mock_create: mock.MagicMock) -> None:
         # Create a test instance of OpenAI
         config = OpenAIConfig()
         openai = OpenAI(config)
 
         # Set up the mock response from openai.ChatCompletion.create
-        class mock_obj:
-            pass
-        mock_chunk1 = mock_obj()
-        mock_chunk1.choices = [mock_obj(), mock_obj()]  # type: ignore
-        mock_chunk1.choices[0].index = 0  # type: ignore
-        mock_chunk1.choices[0].delta = mock_obj()  # type: ignore
-        mock_chunk1.choices[0].delta.content = 'Answer 1'  # type: ignore
-        mock_chunk1.choices[0].finish_reason = None  # type: ignore
-        mock_chunk1.choices[1].index = 1  # type: ignore
-        mock_chunk1.choices[1].delta = mock_obj()  # type: ignore
-        mock_chunk1.choices[1].delta.content = 'Answer 2'  # type: ignore
-        mock_chunk1.choices[1].finish_reason = None  # type: ignore
-        mock_chunk2 = mock_obj()
-        mock_chunk2.choices = [mock_obj(), mock_obj()]  # type: ignore
-        mock_chunk2.choices[0].index = 0  # type: ignore
-        mock_chunk2.choices[0].finish_reason = 'stop'  # type: ignore
-        mock_chunk2.choices[1].index = 1  # type: ignore
-        mock_chunk2.choices[1].finish_reason = 'stop'  # type: ignore
-        mock_create.return_value = iter([mock_chunk1, mock_chunk2])
+        mock_response = {
+            'choices': [
+                {
+                    'message': {
+                        'content': 'Answer 1'
+                    }
+                },
+                {
+                    'message': {
+                        'content': 'Answer 2'
+                    }
+                }
+            ],
+            'usage': {
+                'prompt_tokens': 10,
+                'completion_tokens': 20,
+                'total_tokens': 30
+            }
+        }
+        mock_create.return_value = mock_response
 
         # Create test data
         question = Message(Question('Question'))
@@ -56,9 +57,9 @@ class OpenAITest(unittest.TestCase):
         self.assertIsNotNone(response.tokens)
         self.assertIsInstance(response.tokens, Tokens)
         assert response.tokens
-        self.assertEqual(response.tokens.prompt, 53)
-        self.assertEqual(response.tokens.completion, 6)
-        self.assertEqual(response.tokens.total, 59)
+        self.assertEqual(response.tokens.prompt, 10)
+        self.assertEqual(response.tokens.completion, 20)
+        self.assertEqual(response.tokens.total, 30)
 
         # Assert the mock call to openai.ChatCompletion.create
         mock_create.assert_called_once_with(
@@ -75,7 +76,6 @@ class OpenAITest(unittest.TestCase):
             max_tokens=config.max_tokens,
             top_p=config.top_p,
             n=2,
-            stream=True,
             frequency_penalty=config.frequency_penalty,
             presence_penalty=config.presence_penalty
         )

tests/test_message.py

@@ -91,7 +91,7 @@ class QuestionTestCase(unittest.TestCase):
 class AnswerTestCase(unittest.TestCase):
     def test_answer_with_header(self) -> None:
         with self.assertRaises(MessageError):
-            str(Answer(f"{Answer.txt_header}\nno"))
+            Answer(f"{Answer.txt_header}\nno")
 
     def test_answer_with_legal_header(self) -> None:
         answer = Answer(f"This is a line contaning '{Answer.txt_header}'\nIt is what it is.")