PaLM-rlhf-pytorch VS GLM-130B

ChatGPT isn't open sourced so we don't know what the actual implementation is. I think you can read Open Assistant's source code for application design. If that is too much, try Open Chat Toolkit's source code for developer tools . If you need very bare implementation, you should go for lucidrains/PaLM-rlhf-pytorch.

AFAIK, it is not. They are using the open-source re-implementation of Phil Wang (aka lucidrains), which is available here: https://github.com/lucidrains/PaLM-rlhf-pytorch

Not sure what do you mean by putting source code in double quote, but I don't think the source code is petabytes of text. GPT-2 implementation is few hundred lines of Python (in HuggingFace). PaLM + RLHF - Pytorch (Basically ChatGPT but with PaLM) is less than 1000 lines.

GitHub Link: https://github.com/lucidrains/PaLM-rlhf-pytorch

Well... Not necessarily. Actually, if you want to be extra thrifty, you could even go without an ML expert. Just use an open-source one, like LaMDA or PaLM. After that, use chatGPT to build you a basic front end (which would still be better than CAI lol).

and PaLM with RLHF from Phil Wang (open model, needs to be trained): https://github.com/lucidrains/PaLM-rlhf-pytorch

Closest you can get is probably with Google T5-Flan [1].

It is not the size of the model or the text it was trained on that makes ChatGPT so performant. It is the additional human assisted training to make it respond well to instructions. Open source versions of that are just starting to see the light of day [2].

[1] https://huggingface.co/google/flan-t5-xxl

[2] https://github.com/lucidrains/PaLM-rlhf-pytorch

The https://github.com/THUDM/GLM-130B model is trained on The Pile and can run on 4x3090 when quantized to INT4. I'm wondering if anyone knows if this model could (or has) been quantized using GPTQ, which gives some impressive performance gains over traditional quantization, and I'm also wondering if anyone has tried a 3-bit or 2-bit quantization of such a massive model (using GPTQ). Are there any inherent limitations in this? Is there anything about this model that prevents it from being run on text-generation-webui?

> GLM-130B, a model comparable with GPT-3, has 130 billion parameters in FP16 precision, a total of 260G of GPU memory is required to store model weights. The DGX-A100 server has 8 A100s and provides an amount of 320G of GPU memory (640G for 80G A100 version) so it suits GLM-130B well.

https://github.com/THUDM/GLM-130B/blob/main/docs/low-resourc...

GLM-130B[1] (a 130 billion parameter model vs GPT-3's 175 billion parameter model) is able to run optimally on consumer level high-end hardware, 4xRTX 3090 in particular. That's < $4k at current prices, and as hardware prices go one can only imagine what it'll be in a year or two. It also enables running with degraded performance on lesser systems.

It's a whole lot cheaper to run neural net style systems than to train them. "Somebody on Twitter"[2] got it setup, and broke down the costs, demonstrated some prompts, and what not. Cliff notes being a fraction of a penny per query, with each taking about 16s to generate. The output's pretty terrible, but it's unclear to me whether that's inherent or a result of priority. I expect OpenAI spent a lot of manpower on supervised training, whereas this system probably had minimal, especially in English (it's from a Chinese university).

[1] - https://github.com/THUDM/GLM-130B

[2] - https://twitter.com/alexjc/status/1617152800571416577

GLM-130B in 4 bit mode is better than GPT3 and can run on 4 RTX-3090s. Still expensive but it’s getting closer. https://github.com/THUDM/GLM-130B

https://github.com/THUDM/GLM-130B might be a useful place to look