[ad_1]
The first open source equivalent of OpenAI ChatGPT has arrived
This week, Philip Wang, the developer responsible for reverse engineering closed-source AI systems including Meta Make-A-Video, PaLM + RLHF published a text generation model that works similarly to ChatGPT. The system integrates Paul MGoogle used a large language model, and a technique called reinforcement learning with human feedback — abbreviated as RLHF — to create a system that can do almost any task that ChatGPT can do, including drafting emails and suggesting computer code.
But PalM + RLHF was not pretrained. That is, the computer is not trained on the example data from the Internet that it needs to actually work. Downloading PalM + RLHF doesn’t magically install a ChatGPT-like experience – it requires compiling gigabytes of text from which the model can learn and finding enough hardware to handle the training workload.
Like ChatGPT, PalM + RLHF is a statistical tool for word prediction. When given a large number of examples from training data – for example, Reddit posts, news articles and e-books – PalM + RLHF learns how likely words are based on patterns such as the semantic context of the surrounding text.
ChatGPT and PalM + RLHF share a special sauce in reinforcement learning with human feedback, a technique that aims to better align language models with what users want to achieve. RLHF involves training a language model – in the case of PalM + RLHF, PalM – and fine-tuning it to a dataset that includes instructions attached to what human volunteers expect. Saying (eg, “Machine learning is a form of AI…”). The aforementioned stimuli are then presented to the fine-tuned model, which generates a number of responses, and volunteers rank all responses from best to worst. Finally, the rankings are used to train a “reward model” that takes the original sample’s responses and ranks them by preference, filtering out the best responses for a given prompt.
It is an expensive process to collect training data. And training doesn’t come cheap. PalM is 540 billion parameters, with “parameters” referring to parts of the language model learned from the training data. A 2020 study Estimates of the costs of building a text generation model with just 1.5 billion parameters are $1.6 million. and to train the open source model Bloom, which has 176 billion parameters, took three months using 384 Nvidia A100 GPUs; An A100 costs thousands of dollars.
Running a trained model on the scale of PalM + RLHF is not trivial. Bloom A dedicated PC with around eight A100 GPUs is required. Cloud alternatives are expensive, with back-of-the-envelope math will find out The cost of running OpenAI’s text generation GPT-3 — which has about 175 billion parameters — is worth about $87,000 per year per Amazon Web Services instance.
Sebastian Raschka, an AI researcher, points out on LinkedIn Mail Regarding PalM + RLHF, scaling the required dev workflow can also be a challenge. “Even if someone gives you 500 GPUs to train this model, you still have to deal with the infrastructure and have a software architecture that can handle it,” he said. “It’s obviously possible, but it’s a huge undertaking at the moment (of course, we’re building the framework to make it easier, but it’s still not trivial).”
PalM + RLHF is not going to replace ChatGPT today – unless a well-funded effort (or person) goes to the trouble of training and making it publicly available.
In better news, many efforts to replicate ChatGPT are progressing at a rapid clip. CarperAI. In partnership with open AI research firm EleutherAI and startups Scale AI and Hugging Face, CarperAI plans to release the first ready-to-run, ChatGPT-like AI model trained with human input.
LAION is a nonprofit that provided the initial dataset used for training Constant diffusionhas Leading A project that replicates ChatGPT using new machine learning techniques. Ambitiously, LAION aims to create a “future assistant” — one that doesn’t just write emails and cover letters but “does meaningful work, uses APIs, processes dynamic information and more.” It is in the early stages. But a GitHub page It went live a few weeks ago with resources for the project.
[ad_2]