Noa Roggendorff

nroggendorff

AI & ML interests

Large language models, Illusion Diffusion, Stable Diffusion, Stable Video Diffusion

Articles

Organizations

nroggendorff's activity

posted an update about 10 hours ago
view post
Post
183
What could have been

"{{bos_token}}{% for message in messages %}{% if (message['role'] == 'user') != (loop.index0 % 2 == 0) %}{{ raise_exception('Conversation roles must alternate user/assistant/user/assistant/...') }}{% endif %}{% if message['role'] == 'user' %}{{ '<|user|>\n' + message['content'] + '<|end|>\n' }}{% elif message['role'] == 'assistant' %}{{ '<|bot|>\n' + message['content'] }}{% if loop.last %}{{ '<|end|>' + eos_token }}{% else %}{{ '<|end|>\n' }}{% endif %}{% else %}{{ raise_exception('Only user and assistant roles are supported!') }}{% endif %}{% endfor %}"
  • 1 reply
ยท
replied to not-lain's post 1 day ago
view reply

we know, stop bragging

not that im jealous or anything

jk congrats

replied to their post 1 day ago
view reply

Im pretty sure you can add additional tokens and special tokens, so I suppose

but that doesnt really have anything to do with this article, thats just a transformers feature

replied to their post 1 day ago
view reply

Im pretty sure you can add additional tokens and special tokens, so I suppose

replied to their post 3 days ago
posted an update 3 days ago
replied to their post 10 days ago
posted an update 10 days ago
view post
Post
550
I updated nroggendorff/llava to be even worse. The point of this space is to point out how bad the base model for Llava is, and how without an image it struggles quite a bit. In the new update, from my testing, even if you ask the model about the image, it won't be able to tell you. I experimented with quite a few things, including an image where all the values are 0 (shoutout np.zeros), an image with the most generic portrait photo I could think of (black hair, brown eyes, plain white shirt, etc..) (generated with nroggendorff/epicrealismxl), an image that has text that reads "The image you are looking for is unavailable", and much.. much more.. *sigh*. The image I'm ending up with seems to work best. I don't think it would work with other multi-modal LLMs, but I'm not sure. If you find a problem, feel free to message me on the [huggingface discord](http://hf.co/join/discord), or open a pull request.
  • 2 replies
ยท
posted an update 14 days ago
view post
Post
706
I changed my mind about kaggle nb versions. They're goated
posted an update 18 days ago
replied to their post 22 days ago
replied to their post 22 days ago
posted an update 22 days ago
replied to their post 26 days ago
view reply

I mean like replacing the binary data with decimals, so I could store a message

for example,
The letter "J" would be 74

replied to their post 26 days ago
replied to their post 26 days ago
replied to their post 26 days ago
view reply

we seek that fucking delete form of any subscribing to your platorm ; it is not ficking possible make it so ; because CEO I'll fuck you up personaly big time if you can't fucking get real

What is wrong with you

replied to their post 26 days ago
view reply

where is the delete account button ? we d\wish not to read hiilbillies

what??

replied to their post 26 days ago
replied to their post 26 days ago
posted an update 26 days ago
view post
Post
2466
I'm about to start storing files in my TFLOPS count..
ยท
posted an update about 1 month ago
replied to their post about 1 month ago
posted an update about 1 month ago
view post
Post
3308
I wish I had a cheaper and more appreciated by the masses hobby.
ยท
replied to their post about 2 months ago
posted an update about 2 months ago
posted an update about 2 months ago
posted an update about 2 months ago
posted an update about 2 months ago