search

meta-llama / codellama

Inference code for CodeLlama models
Other
15.92k stars 1.85k forks source link

Will there be a Codellama based on Llama 3? #227

Open PhilKes opened 5 months ago

PhilKes commented 5 months ago

Is there going to be an updated version of Codellama based on Meta's new LLaMa 3 ?

hyperbolic-c commented 5 months ago

Same question！Looking forward to the model based on llama3!

bgruszka commented 5 months ago

It would be great to have it based on llama 3 😄

bmaciag commented 5 months ago

Same question.

TheLapinMalin commented 5 months ago

Apparently Llama 3 has already been trained on a lot more code than Llama 2.

So, do we need a full blown Codellama 3 model, or do you think a FIM fine-tune of Llama 3 would be sufficient?

PhilKes commented 4 months ago

Apparently Llama 3 has already been trained on a lot more code than Llama 2.

So, do we need a full blown Codellama 3 model, or do you think a FIM fine-tune of Llama 3 would be sufficient?

Would love to see a FIM fine-tune of Llama 3, I dont have any insights on how the training process differed from Llama 2. Is there anyone working on a FIM fine-tune? Haven't seen one yet on HF

reneleonhardt commented 4 months ago

Would love to see a FIM fine-tune of Llama 3, I dont have any insights on how the training process differed from Llama 2. Is there anyone working on a FIM fine-tune? Haven't seen one yet on HF

Supposedly StarCoder2 should be perfect for FIM 😅

hschaeufler commented 4 months ago

Would also love to see a Code Llama Version based on LLama 3. Llama 3 70b seems to perform better on coding-Tasks than Code Llama 70b.

DevopsDood commented 2 months ago

3.1:70b will be :fire:

ShahNewazKhan commented 3 weeks ago

+1

ogbozoyan commented 2 weeks ago

+1 there'e even fresher model which impress