During a continuous 12-day live event, OpenAI released the O1 model API to developers on the 9th day and announced significant upgrades to the real-time API, supporting WebRTC technology. Starting from the release date, OpenAI will provide access to the O1 API for developers at usage level 5. Compared to the previous preview version, this updated O1 model API achieves a 60% reduction in thinking costs and adds advanced visual processing capabilities. Meanwhile, GPT-4o has improved in audio processing.

Recently, a research team from Shanghai Jiao Tong University and Harvard University introduced a novel model fine-tuning method — LoRA-Dash. This new approach claims to be more efficient than existing LoRA methods, particularly in the fine-tuning of specific tasks, achieving the same results with an 8 to 16 times reduction in the number of parameters. This is undoubtedly a major breakthrough for fine-tuning tasks that require substantial computational resources. In the context of the rapid development of large-scale language models, the demand for fine-tuning specific tasks is steadily increasing. However, fine-tuning often

Baidu Intelligent Cloud's Qianfan Platform has recently announced its support for customized fine-tuning services for the flagship large model ERNIE 4.0 Turbo. Enterprise clients can apply for this service through the Baidu Intelligent Cloud official website to meet their specific business needs and data characteristics, thereby training a large model that is more suitable for their business and significantly improving the effectiveness of business applications.

OpenAI has announced that third-party developers can fine-tune the newly introduced multimodal large model GPT-4o. This feature allows developers to adjust the model's behavior based on specific needs, such as changing the tone, following instructions, or improving the accuracy of technical tasks. Fine-tuning can achieve significant performance improvements with only a small amount of training data and is easy to operate. OpenAI is providing developers with a free allowance of 1 million tokens daily until September 23, as well as a daily free allowance of 2 million tokens for the GPT-4o mini model.

PanoFree is an innovative multi-view image generation technology that can produce various panoramic images, including long images, 360-degree panoramas, and skyboxes, without the need for fine-tuning. Its core lies in addressing consistency issues and artifacts during the generation process through an iterative deformation and repair process while enhancing cross-view awareness. PanoFree employs techniques such as cross-view guidance, risk area estimation, erasure, and symmetric bidirectional guidance, combined with guided semantic and density control to ensure the preservation of scene structure. Experiments show that PanoFree achieves ...

OpenAI has announced the launch of GPT-4o mini, offering daily free fine-tuning with 2 million training tokens until September 23rd, aimed at advancing AI technology and fostering developer innovation. This model boasts performance on par with GPT-4o but is priced at only 1/20th of the latter, with significant advantages in training costs, context length, and intelligence levels. After the free period, services will continue to be offered at discounted rates. The fine-tuning capabilities of GPT-4o mini will be initially available to enterprise clients and specific developer tiers, with future expansion planned.

Microsoft has announced the launch of serverless fine-tuning capabilities for the Phi-3 small language model, enabling developers to easily adjust model performance on the Azure AI platform without managing underlying infrastructure, and it is free for initial use. The Phi-3 model, with 3 billion parameters, is suitable for enterprise developers, offering efficient performance at a lower cost. Although it has fewer parameters compared to OpenAI's GPT-3.5, it performs similarly across various applications. Serverless fine-tuning simplifies operations, reduces barriers, and supports the fine-tuning of small to medium models.

Recently, leading artificial intelligence company OpenAI announced the launch of fine-tuning capabilities for its new small-scale model, GPT-4o mini, aimed at enhancing the model's performance in specific application scenarios. This move will provide developers and enterprise users with more customized AI solutions. According to official OpenAI news, from today until September 23, users can fine-tune GPT-4o mini for free up to 2 million training tokens per day. Any excess will be charged at $3.0

In the realm of artificial intelligence, Large Language Models (LLMs) are renowned for their exceptional natural language processing capabilities. However, deploying these models in practical applications faces significant challenges due to their high computational costs and memory footprint during the inference phase. To address this issue, researchers have been exploring ways to enhance the efficiency of LLMs. Recently, a method called Q-Sparse has garnered widespread attention.Q-Sparse is a s

In the field of artificial intelligence, large language models (LLMs) such as GPT-3 and Llama-2 have made significant strides, capable of accurately understanding and generating human language. However, the vast number of parameters in these models poses challenges in terms of the significant computational resources required for both training and deployment, particularly in resource-constrained environments.Paper Access: https://arxiv.org/html/2406.10260v1Traditionally, to achieve a balance betw

On July 11th, Kuaikan Comics announced that they are exploring fine-tuning of open-source large models to train a vertical large model for the two-dimensional (2D) domain. Kuaikan Comics utilizes large language models (LLM) and retrieval-augmented generation (RAG) technology to improve the conversion rate of work searches and inject new vitality into the search process. They have adopted a "fine-tuned large model + RAG enhancement" strategy, building internal PGC (Professional Generated Content)

Korea's major mobile carrier LG Uplus officially launched its generative artificial intelligence model ixi-GEN on Tuesday, joining its two larger competitors KT Corp. and SK Telecom Co. in the race to lead the B2B generative AI market.

Text-based image generation technology has garnered significant attention, allowing models to create realistic images based on textual descriptions. Researchers introduced the Orthogonal Fine-Tuning (OFT) method, greatly enhancing control over image generation models. OFT preserves the model's semantic generation capabilities while improving the quality and efficiency of the generated images. The OFT method has wide-ranging applications in fields such as artistic creation and advertising, offering new creative possibilities.

The aMUSEd model launched by Hugging Face can generate images in seconds, faster than competitors like Stable Diffusion. aMUSEd utilizes a lightweight text-to-image model based on Google's MUSE model. It adopts a Masked Image Model (MIM) architecture, reducing inference steps and enhancing generation speed and interpretability. The model is available in a demo on Hugging Face.

The Alibaba Tongyi Laboratory open sources the AI image generation fine-tuning framework SCEdit. The SCEdit framework supports fine-tuning for image generation tasks and demonstrates high efficiency in generation tasks. It achieves precise control in controllable generation tasks while saving training memory costs.

Cohere announced a new fine-tuning suite, featuring capabilities such as chat, reordering, and multi-label classification to empower personalized AI applications for businesses. The new fine-tuning dashboard allows users to manage and run fine-tuning projects, providing a testing ground, real-time progress monitoring, and a price calculator. Fine-tuning enhances model performance, improving the accuracy of generated solutions by 60%, with chat solutions improved by 40%. Models can be fine-tuned in 30 minutes and deployed in under a minute. Cohere plans to expand its London research center's workforce to meet the demand for customized models and improved performance through fine-tuning.

By using chained density prompting to fine-tune ChatGPT, we achieve performance on news summarization tasks that exceeds GPT-4's zero-shot capabilities and approaches GPT-4's performance with chained prompting. This is accomplished through automatic evaluation chains such as ScoreStringEvalChain and the improved PairwiseStringEvalChain for quantitative assessments at human-level quality. Fine-tuning ChatGPT is 11 times faster and 63% cheaper than GPT-4's zero-shot method, and 33 times faster and 84% cheaper than GPT-4's chained prompting. The results show that fine-tuned ChatGPT excels in news summarization.

RA-DIT improves the ability of language models to utilize retrieval information and provide relevant results from the retriever through a two-stage fine-tuning process. RA-DIT outperforms existing retrieval-augmented models in knowledge-intensive zero-shot and few-shot testing. The RA-DIT65B model shows improved performance in tasks that require knowledge, demonstrating the effectiveness of fine-tuning. The enhancement of retrieval capabilities brought by RA-DIT has shown good results in knowledge-intensive evaluations, confirming the effectiveness of fine-tuning.

The Ma Yi team proposed the EMT framework to evaluate the catastrophic forgetting of fine-tuned Multimodal Large Models (MLLMs). Experiments showed that while fine-tuning MLLMs improves performance on the fine-tuning dataset, it also leads to a decline in performance on other datasets. During the fine-tuning process, MLLMs generate hallucinated text related to the fine-tuning dataset, ignoring the original problem. This research provides a framework and benchmarks for future work, but the design and training techniques of the model still need further optimization. The Ma Yi team systematically assessed the issue of catastrophic forgetting in MLLMs for the first time, balancing different capabilities.