🚩 Report: Fake & Spam
False claims , fake numbers and benchmark, just fishing for likes
Agreed
It's not "spam".
read the code.
that his evaluation methods are not convincing / complete, doesn't render the model meritless.
I downloaded the model and tried it for myself before actually submitting this , and its no where near what it claims .
Comparing 4b model to sota models and claiming to out perform them by a mile is just too far fetched.
It falls under the "omg this small model fine-tuned is better than sota models " hype spam.
It's not "spam".
read the code.
that his evaluation methods are not convincing / complete, doesn't render the model meritless.
No it's spam because it's same model. I'm not even sure there was much tuning. Performs on simple question exactly the same as phi3-mini
It's not "spam".
read the code.
that his evaluation methods are not convincing / complete, doesn't render the model meritless.
No it's spam because it's same model. I'm not even sure there was much tuning. Performs on simple question exactly the same as phi3-mini
I am currently trying to implement the LLAMA version, which is an improvement on the architecture, and I hope my implementation is correct.This is not just a fine-tuning, it is an architectural improvement, very interesting~
It's not "spam".
read the code.
that his evaluation methods are not convincing / complete, doesn't render the model meritless.
No it's spam because it's same model. I'm not even sure there was much tuning. Performs on simple question exactly the same as phi3-mini
Perhaps freezing the original weights, and training the added components.
Although I have said so much, there are always some ignorant and unmotivated people who just deny it.
