In line with our commitment to continually enhance Generative AI tools for staff and students, we have provisioned additional open-source large language models and refreshed ChatGPT to GPT-4o in July. GPT-4o comes with a reduced token use fee and improved vision capability. With the availability of Alibaba Qwen, Mistral AI Mixtral, and Meta Llama models, staff and students should be able to leverage Generative AI tools with less concern about reaching the monthly usage entitlement threshold.
Open AI GPT-4o
OpenAI released the GPT-4o in May 2024. Access via Azure Cloud regional hosting centers for text chat and image-to-text was released to Enterprise customers in June 2024. GPT-4o matches the performance of GPT-4 Turbo on English text and code benchmarks, but only costs half the token fee for both input and output.
GPT-4o has been available on the PolyU GenAI app since 3 July 2024, replacing the GPT-4 Turbo 0125 preview, the GPT-4 0613 model, as well as image-to-text using the Azure Cognitive Vision service.
Alibaba Tongyi Qianwen 通义千问 (Qwen1.5-110B-chat)
Alibaba Cloud released Tongyi Qianwen large language model Qwen1.5-110B in February 2024. The Qwen1.5-110B-chat model improved alignment with human preference during training and enhanced multilingual support covering English, Chinese, French, Spanish, German, Russian, Korean, Japanese and Arabic.
Qwen1.5-110B-chat was released on PolyU GenAI on 1 June 2024, replacing Qwen-72B model. The use of Qwen1.5-110B-chat model will not consume the monthly credit usage entitlement.
Mistral AI Mixtral-8x22B-instruct-v.01
Mistral AI released the Mixtral-8x22B Mixture of Expert model in April 2024. The Mixtral-8x22B model outperforms Mistral-8x7B in common sense and reasoning benchmarks, making it one of the best open-source models for complex reasoning tasks.
It has been available on PolyU GenAI since 3 July 2024, replacing the Mixtral-8x7B model. The use of the Mixtral-8x22B model will not consume the monthly credit usage entitlement.
Meta Llama 3 70B
Meta released Llama 3 in April 2024. The Llama 3-70B-instruct model has been pre-trained and fine-tuned to support a broad range of use cases with improved reasoning, code generation and instruction following. Llama 3 70B has adopted both input and output safeguards in addition to model-level risks mitigations to improve use safety.
The Meta llama 3 70B model has been available on PolyU GenAI since 3 July 2024. The use of Llama 3 70B will not consume the monthly credit entitlement.
Stable Diffusion 3.0
Stable Diffusion 3.0 was released on Stability.ai Cloud in the first half of 2024. Stable Diffusion 3.0 leverages a family of models to generate better text / font styles within images, as well as improved image captioning, thus following user's input prompts closely.
The quality of images generated by Stable Diffusion 3.0 is much better compared to the SDXL 1.0 models.
Stable Diffusion 3.0 has been available on PolyU GenAI since 3 July 2024.
We encourage staff and students to explore their capabilities and consider how they can enhance their work. Let us embrace these tools with open minds and a spirit of discovery, as we continue to strive for excellence in teaching and learning.
For further information, please contact the IT HelpCentre (Tel: 2766 5900, WhatsApp / WeChat: 6577 9669) or reach out to us via the IT Online ServiceDesk.
