clipboard feature

README.md

@@ -66,14 +66,11 @@ https://github.com/ollama/ollama/blob/main/docs/import.md#manually-converting--q
 
 
 ````
-docker
-https://www.youtube.com/watch?v=m0fc6ZPb6NU
-https://www.geoffreylitt.com/2023/03/25/llm-end-user-programming.html
-https://ollama.com/blog/ollama-is-now-available-as-an-official-docker-image
-https://hub.docker.com/r/ollama/ollama
-https://collabnix.com/getting-started-with-ollama-and-docker/
-https://docs.coqui.ai/en/latest/inference.html
-https://github.com/valiantlynx/ollama-docker
+improve
+https://www.geoffreylitt.com/2023/03/25/llm-end-user-programming
+https://blog.waleson.com/2024/05/the-long-long-tail-of-ai-applications.html
+https://www.strangeloopcanon.com/p/what-can-llms-never-do
+https://jxnl.co/writing/2024/05/22/systematically-improving-your-rag/#cluster-and-model-topics
 ````
 
 ````

bin/__main__.py

@@ -0,0 +1,41 @@
+from dataclasses import dataclass
+
+from jsonargparse import CLI
+
+from llmpeg.agent import Agent
+
+
+@dataclass()
+class Main:
+        rational_model: str
+        trigger_model: str
+        speech_model_size: str
+        hear_model_size: str
+
+        def __post_init__(self):
+                self.agent = Agent(
+                        self.rational_model, self.trigger_model, self.speech_model_size, self.hear_model_size
+                )
+
+        def run(self):
+                # NOTE: [EDITABLE]
+
+                self.agent.chat()
+                # self.agent.summarize_search('https://news.mit.edu/2020/brain-reading-computer-code-1215')
+                # self.agent.explain_search('https://news.mit.edu/2020/brain-reading-computer-code-1215')
+
+                # ----------------
+
+
+def main():
+        try:
+                CLI(Main)
+                return 0
+        except KeyboardInterrupt:
+                return 0
+        except ValueError:
+                return 0
+
+
+if __name__ == '__main__':
+        exit(main())

llmpeg/actions/actions.py

@@ -11,15 +11,15 @@
 
 @dataclass
 class Actions:
-  cache_dir: Path
-  rational_model: str  # ollama/llama3
-  trigger_model: str  # ollama/gemma:2b
-  speech_model_size: str  # tts_models/en/jenny/jenny
-  hear_model_size: str  # openai-whisper/base
+        cache_dir: Path
+        rational_model: str  # ollama/llama3
+        trigger_model: str  # ollama/gemma:2b
+        speech_model_size: str  # tts_models/en/jenny/jenny
+        hear_model_size: str  # openai-whisper/base
 
-  def __post_init__(self) -> None:
-    self.rational = BrainRational(self.rational_model)
-    self.trigger = BrainTrigger(self.trigger_model, self.cache_dir)
-    self.hear = Hear(self.speech_model_size, self.cache_dir)
-    self.speech = Speech(self.hear_model_size, self.cache_dir)
-    self.vision = Vision()
+        def __post_init__(self) -> None:
+                self.rational = BrainRational(self.rational_model)
+                self.trigger = BrainTrigger(self.trigger_model, self.cache_dir)
+                self.hear = Hear(self.speech_model_size, self.cache_dir)
+                self.speech = Speech(self.hear_model_size, self.cache_dir)
+                self.vision = Vision()

llmpeg/actions/brain/rational.py

@@ -5,33 +5,34 @@
 
 
 # TODO: have a conversation with preprompted character roleplay and play songs on request
-# TODO: this should be a in front of browser and call it todo stuff instead of bypassing this and using capabilities directly
+# TODO: this should be a in front of browser and call it todo stuff
+# TODO: instead of bypassing this and using capabilities directly
 @dataclass
 class BrainRational:
-  model: str  # NOTE: e.g. "gemma:2b"
-  explain_prompt: str = 'Explain the following data which was extracted from a webpage in your own words'
-  summarize_prompt: str = 'Summarize the following data which was extracted from a webpage'
+        model: str  # NOTE: e.g. "gemma:2b"
+        explain_prompt: str = 'Explain the following data which was extracted from a webpage in your own words'
+        summarize_prompt: str = 'Summarize the following data which was extracted from a webpage'
 
-  # TODO: sqlite3 for storing chat history
-  def __post_init__(self) -> None:
-    self.chat_messages = []
-    self.llm = LLM(self.model)
+        # TODO: sqlite3 for storing chat history
+        def __post_init__(self) -> None:
+                self.chat_messages = []
+                self.llm = LLM(self.model)
 
-  def summarize(self, prompt: str) -> str:
-    return self.llm.generate(f'{self.summarize_prompt}\n{prompt}')
+        def summarize(self, prompt: str) -> str:
+                return self.llm.generate(f'{self.summarize_prompt}\n{prompt}')
 
-  def explain(self, prompt: str) -> str:
-    return self.llm.generate(f'{self.explain_prompt}\n{prompt}')
+        def explain(self, prompt: str) -> str:
+                return self.llm.generate(f'{self.explain_prompt}\n{prompt}')
 
-  def respond(self, prompt: str) -> str:
-    return self.llm.generate(prompt)
+        def respond(self, prompt: str) -> str:
+                return self.llm.generate(prompt)
 
-  def clear_chat(self) -> None:
-    self.chat_messages = []
+        def clear_chat(self) -> None:
+                self.chat_messages = []
 
-  def _add_message(self, prompt) -> None:
-    return self.chat_messages.append({'role': 'user', 'content': prompt})
+        def _add_message(self, prompt) -> None:
+                return self.chat_messages.append({'role': 'user', 'content': prompt})
 
-  def chat(self, prompt: str) -> Union[str, list[str]]:
-    self._add_message(prompt)
-    return self.llm.recall_generate(self.chat_messages[-1]['content'], self.chat_messages)
+        def chat(self, prompt: str) -> Union[str, list[str]]:
+                self._add_message(prompt)
+                return self.llm.recall_generate(self.chat_messages[-1]['content'], self.chat_messages)

llmpeg/actions/brain/trigger.py

@@ -10,49 +10,85 @@
 
 @dataclass
 class BrainTrigger:
-  model_name: str
-  cache_dir: Path
-
-  # TODO: LLM for NLP
-  def __post_init__(self):
-    self.cache_dir = self.cache_dir / 'triggers'
-    Path.mkdir(self.cache_dir, exist_ok=True)
-    self.llm = LLM(self.model_name)
-    os.environ['NLTK_DATA'] = str(self.cache_dir / 'nltk_data')
-    nltk.download(self.model_name)  # NOTE: e.g. "punkt"
-
-  def _find_intent(self, instruction: str) -> str:
-    return (
-      f'Complete the following text: Given the instruction "{instruction}". '
-      + 'Categorize it as "play music", "research web". "chat with me". The answer is '
-    )
-
-  def find_intent(self, prompt: str) -> str:
-    response = self.llm.generate(self._find_intent(prompt))
-    if self.check_audio_request(response):
-      return 'play'
-    elif self.check_browse_request(response):
-      return 'browse'
-    elif self.check_explain_request(response):
-      return 'chat'
-
-  def check_greeting(self, prompt: str) -> bool:
-    return any(keyword in nltk.tokenize.word_tokenize(prompt.lower()) for keyword in TriggerList.greeting)
-
-  def check_goodbye(self, text: str) -> bool:
-    return any(keyword in nltk.tokenize.word_tokenize(text.lower()) for keyword in TriggerList.goodbye) or all(
-      keyword in nltk.tokenize.word_tokenize(text.lower()) for keyword in TriggerList.goodbye_default_phrase
-    )
-
-  def check_browse_request(self, text: str) -> bool:
-    tokens = nltk.tokenize.word_tokenize(text.lower())
-    return True if TriggerList.browse_start in tokens and any(keyword in tokens for keyword in TriggerList.browse_check) else False
-
-  def check_explain_request(self, text: str) -> bool:
-    tokens = nltk.tokenize.word_tokenize(text.lower())
-    return True if TriggerList.explain_start in tokens and any(keyword in tokens for keyword in TriggerList.explain_check) else False
-
-  # TODO: find sentiment to play song or not
-  def check_audio_request(self, text: str) -> bool:
-    tokens = nltk.tokenize.word_tokenize(text.lower())
-    return True if TriggerList.audio_start in tokens and any(keyword in tokens for keyword in TriggerList.audio_check) else False
+        model_name: str
+        cache_dir: Path
+
+        # TODO: LLM for NLP
+        def __post_init__(self):
+                self.cache_dir = self.cache_dir / 'triggers'
+                Path.mkdir(self.cache_dir, exist_ok=True)
+                self.llm = LLM(self.model_name)
+                os.environ['NLTK_DATA'] = str(self.cache_dir / 'nltk_data')
+                nltk.download(self.model_name)  # NOTE: e.g. "punkt"
+
+        def _find_intent(self, instruction: str) -> str:
+                return (
+                        f'Complete the following text: Given the instruction "{instruction}". '
+                        + 'Categorize it as "play music", "research web". "chat with me". The answer is '
+                )
+
+        def find_intent(self, prompt: str) -> str:
+                response = self.llm.generate(self._find_intent(prompt))
+                if self.check_audio_request(response):
+                        return 'play'
+                elif self.check_browse_request(response):
+                        return 'browse'
+                elif self.check_explain_request(response):
+                        return 'chat'
+
+        def check_greeting(self, prompt: str) -> bool:
+                return any(keyword in nltk.tokenize.word_tokenize(prompt.lower()) for keyword in TriggerList.greeting)
+
+        def check_goodbye(self, text: str) -> bool:
+                return any(
+                        keyword in nltk.tokenize.word_tokenize(text.lower()) for keyword in TriggerList.goodbye
+                ) or all(
+                        keyword in nltk.tokenize.word_tokenize(text.lower())
+                        for keyword in TriggerList.goodbye_default_phrase
+                )
+
+        def check_chat_request(self, text: str) -> bool:
+                tokens = nltk.tokenize.word_tokenize(text.lower())
+                return (
+                        True
+                        if TriggerList.chat_start in tokens
+                        and any(keyword in tokens for keyword in TriggerList.chat_check)
+                        else False
+                )
+
+        def check_browse_request(self, text: str) -> bool:
+                tokens = nltk.tokenize.word_tokenize(text.lower())
+                return (
+                        True
+                        if TriggerList.browse_start in tokens
+                        and any(keyword in tokens for keyword in TriggerList.browse_check)
+                        else False
+                )
+
+        def check_summarize_request(self, text: str) -> bool:
+                tokens = nltk.tokenize.word_tokenize(text.lower())
+                return (
+                        True
+                        if TriggerList.summarize_start in tokens
+                        and any(keyword in tokens for keyword in TriggerList.summarize_check)
+                        else False
+                )
+
+        def check_explain_request(self, text: str) -> bool:
+                tokens = nltk.tokenize.word_tokenize(text.lower())
+                return (
+                        True
+                        if TriggerList.explain_start in tokens
+                        and any(keyword in tokens for keyword in TriggerList.explain_check)
+                        else False
+                )
+
+        # TODO: find sentiment to play song or not
+        def check_audio_request(self, text: str) -> bool:
+                tokens = nltk.tokenize.word_tokenize(text.lower())
+                return (
+                        True
+                        if TriggerList.audio_start in tokens
+                        and any(keyword in tokens for keyword in TriggerList.audio_check)
+                        else False
+                )