Which Llama LLM model to use for AI friend app?
- Thread starter BradyX
- Start date
- Dec 24, 2022
- 152
- 79
I would highly recommend choosing OpenAI over open-source LLMs, despite the potential higher cost.
Overall, OpenAI is superior in performance and reliability.
As a software developer, I have found that using a variety of models is better. Each model has its own advantages.
For instance, if you have a particularly "difficult" task, GPT-4 would be the ideal choice.
However, for simpler tasks like checking if a message is on-topic, one of the smaller and more cheaper models would be fine.
I suggest using LangChain and LangGraph as they are incredibly useful tools. You need to know TypeScript or Python.
Overall, OpenAI is superior in performance and reliability.
As a software developer, I have found that using a variety of models is better. Each model has its own advantages.
For instance, if you have a particularly "difficult" task, GPT-4 would be the ideal choice.
However, for simpler tasks like checking if a message is on-topic, one of the smaller and more cheaper models would be fine.
I suggest using LangChain and LangGraph as they are incredibly useful tools. You need to know TypeScript or Python.
- May 27, 2016
- 11,895
- 2,732
Use chatgpt for your needs.
WesleyTheBlick
Newbie
- Sep 13, 2024
- 5
- 1
Wrong answer. Maybe six months ago this may of been the case.
Sure. OpenAI has first mover advantage. But both Anthropic's Claude is fantastic IMO. But who wants to be a sucker and let these fuckers suck up your key, business critical for data.
Because with a former NSA chief administrator now at OpenAI .. its not in our interest.
Meta's llama 3.1 is FANTASTIC. I have been using uncensored forks with extremely impressive results .
The key aspect is to place your data into vector database. I would start with small text arguments. It's key to note that "tokens" .. are akin to segments of words (like n-grams in linguistics.. which compromise parts of words) .. vRAM is CRITICAL for performance. And the baseline model is held in vRAM along then the vectored data AND then the results.
This is why responses are limited and the workflow concept of RAG is very popular.
If you are going to pour cash into other people's tools when you can buy a GPU (or rent bare metal by the hour.
In closing, anyone serious about using open models on small scale runs (20 to 30 4090 GPUs) .. hit me up.
jskinnerd
Regular Member
- Jul 21, 2023
- 205
- 299
This!
DarkerAds
BANNED
- May 1, 2023
- 568
- 1,042
there are plenty of open source LLMs with several flavors
I've seen some that are dedicated to role-playing or creating stories and i'm dead sure that openai isn't fine-tuned for roleplay or creating stories