mradermacher

/

model_requests

like 140

Also I want to say, if it's too small or isn't useful this is just a test model before I make my first from scratch maybe more useful model

if it works it will be quanted, there's nothing too small or isnt useful and small. Unless you are asking for 1T useless model we dont mind quanting anything of yours =)

You can check for progress at http://hf.tst.eu/status.html or regularly check the model
summary page at https://hf.tst.eu/model#Bye-world-GGUF for quants to appear.

if it works it will be quanted, there's nothing too small or isnt useful and small. Unless you are asking for 1T useless model we dont mind quanting anything of yours =)

You can check for progress at http://hf.tst.eu/status.html or regularly check the model
summary page at https://hf.tst.eu/model#Bye-world-GGUF for quants to appear.

Hmm I looked at the status and it still isn't done so I checked the summary and there are some errors

-6967 1 si Bye-world error/1 ValueError Tokenizer class Toke And a lot of other had similar issues

seems like something is wrong with the tokenizer
ValueError: Tokenizer class TokenizersBackend does not exist or is not currently imported.

Some still work so I guess it's a issue with some models

wait, is that a custom tokenizer? We cannot do that, we can only accept some predefined as far as I know, unless llama cpp changed that and I am not aware of it

wait, is that a custom tokenizer? We cannot do that, we can only accept some predefined as far as I know, unless llama cpp changed that and I am not aware of it

There's a chance I accidently made a custom tokenizer instead of gpt 2 but why does hugging face recognise it as gpt 2?

It might recognize from config, but if tokenizer itself is different, llama cpp cant process it (I think it does hash eval or something). So doublr check your tokenizer and make sure it;s compatible with llama cpp

It might recognize from config, but if tokenizer itself is different, llama cpp cant process it (I think it does hash eval or something). So doublr check your tokenizer and make sure it;s compatible with llama cpp

since its my first model i thought it could have issues, i will try to run it on python like gpt 2

I think i found the issue, i think it was trained to be simular to gpt 2 but not quite like with a normal script to run normal gpt 2 models i get this mess

C:\Users\Administrator\Desktop>python test.py
`torch_dtype` is deprecated! Use `dtype` instead!
Loading weights: 100%|███████████████████████████████████████████████████████████████████████| 76/76 [00:00<00:00, 1528.02it/s, Materializing param=transformer.wte.weight]
Setting `pad_token_id` to `eos_token_id`:50256 for open-end generation.
ĠO n c e Ġu p o n Ġa Ġt i m e ĠBye ĠMy ĠWorld ĠBye ĠMy ĠWorld ĠBye ĠMy ĠWorld ĠBye ĠWorld ĠBye ĠWorld ĠWorld ĠWorld ĠWorld ĠWorld ĠWorld ĠWorld ĠWorld ĠBye ĠWorld ĠWorld ĠWorld ĠWorld ĠWorld ĠWorld ĠWorld ĠWorld ĠWorld ĠWorld ĠWorld ĠWorld ĠWorld ĠWorld ĠWorld ĠWorld ĠWorld ĠWorld ĠWorld ĠWorld ĠWorld ĠWorld ĠWorld ĠWorld ĠWorld ĠWorld ĠWorld ĠWorld ĠWorld ĠWorld

C:\Users\Administrator\Desktop>

I guess i need to retrain it

ah damn, that's sad, well, goodluck! make sure it's the correct tokenizer =)

ah damn, that's sad, well, goodluck! make sure it's the correct tokenizer =)

Ill train one and in a few minutes update it

This time i got Bye Cruel World World World World World World World World World World World World World World World World World World World World World World
i will upload and test it if it works using my other script

This one seems to work for me atleast, i should just tell users to limit the generation to like 18 or 24 max char

Best model ever lol. Just let me know when to requeue

Best model ever lol. Just let me know when to requeue

I think its alredy uploaded, also im working on a model from scratch thats trained on oass1, i guess i will be able to later finetune it

Best model ever lol. Just let me know when to requeue

i should also probably close this, if you want you can requeue

requeued !