3 ms·
The best options are probably GPT-J and BLOOM but they are both so big they need at least a 24GB of VRAM and they are not as suffisticated.
by theCrowing 4y ago
The best options are probably GPT-J and BLOOM but they are both so big they need at least a 24GB of VRAM and they are not as suffisticated.
- musicale 4y ago> not as suffisticated ChatGPT in a nutshell. (I hadn't heard the slang before, but it seems descriptive for a false air of sophistication, or [possibly over-]sophisticated complexity that nevertheless leads to bad results.)
- canadianfella 4y ago
- tough 4y agoyou can fine-tune in 8bits weights for use in collab or similarly more restrained environments https://github.com/huggingface/transformers/issues/14839 https://github.com/huggingface/transformers/issues/14839 https://huggingface.co/hivemind/gpt-j-6B-8bit https://huggingface.co/hivemind/gpt-j-6B-8bit by fine-tuning you should be able to attain niche-specific better results than default chatGPT theoretically, if you're lucky and thorough I guess
- jakearmitage 4y agoHow do you fine-tune to a specific use-case, though? You train it again, with smaller weights?