Open simonw opened 10 months ago
Not sure how best to design this one.
I got to this:
@cli.command()
def conversations():
"List conversations stored in the logs"
db = sqlite_utils.Database(logs_db_path())
migrate(db)
for row in db["conversations"].rows_where(order_by="id"):
click.echo(
"{id} {model}\t{name}".format(
id=row["id"], model=row["model"], name=' '.join((row["name"] or "").split())
)
)
I don't like it much though:
01h88dkcfxtkn1hpp5mk8er1q9 openrouter/anthropic/claude-2 import llm from llm.default_plu…
01h88frbzhhz3jwyxarq9m9wkj mlc-chat-Llama-2-7b-chat-hf-q4f16_1 What was the beef between the t…
01h88fxfmdenpksh27ewrn3766 mlc-chat-Llama-2-13b-chat-hf-q4f16_1 What was the beef between the t…
01h88p72dmmw46bnw96mz7we3g gpt-3.5-turbo from typing import List, Dict d…
01h88qmbkg4tnqzgzvdj58hrtc gpt-3.5-turbo ubuntu install build essentials…
01h88qve060a3r93hahhn7bg33 gpt-3.5-turbo ubuntu: tclsh not fonud
01h88r1q9x6sa9tw57g3fajtj1 gpt-3.5-turbo What do you know about OpenAI F…
The design challenge here is how much information to display and how to display it.
I think I want:
llm logs --cid X
to see the actual conversation.Really not sure how best to lay those out in a concise terminal format though.
Docker might be good inspiration
In particular:
I think shell_gpt's solution is pretty good. (It's the sole reason I still use sgpt.)
shell_gpt allows you to name your chats. It's very handy for retrieving them, and providing instant topical context when I have a new question. I have chats for databases, nix, parsing, etc.
If people are trying to resume a conversation, a name beats a cid hands down.
just came here to say the same thing - it'd be cool to be able to to name ones convos then continue them like
llm chat --continue thesis-assistant
A command for listing just conversations - their IDs, titles, number of messages and start date.