🔥The 9th Round of Easy Loan, Earn $40 Reward is in progress❗️ ⏰ Promotion Period: January 15th - Feburary 15th, 2025 👉 Register now and check more details at gate.io/campaigns/358

1/n. The Machine Learning group at Microsoft Research Asia is hiring researchers on deep learning, generative models, machine translation and language processing, speech/audio synthesis and speech processing, etc. If you have interests, welcome to contact me [email protected] :)

How Good Are GPT Models at Machine Translation? A Comprehensive Evaluation abs: arxiv.org/abs/2302.09210

📢 New paper alert! We evaluate the performance of ChatGPT / GPT-3.5 for machine translation across 18 high & low resource language pairs. Can combining GPT models with conventional NMT systems enhance translation quality? Our study suggests it's possible 🧵

joint work with a great team Microsoft: Amr Hendy, Mohamed Abdelrehim , Vikas Raunak, Mohamed Gabr, Hitokazu Matsushita, Young , Mohamed Afify, Hany Awadalla

👋 Check out our new work on a comprehensive evaluation of GPT models for MT. Lots of goodies in the paper, including a thorough analysis of the differential aspects of the translations produced by GPT. 👉 Manuscript: arxiv.org/abs/2302.09210 👉 GitHub: github.com/microsoft/gpt-…

Interested in using large language models for translating literary works (e.g., novels)? We hire human translators to annotate errors in LLM translations. In sum, LLMs w/ doc-level context >>> Google Translate, but they still make major errors (across 18 diff language pairs)!

Nice work and thorough evaluation on document-level translation capabilities of ChatGPT.

Deliberate then Generate: Enhanced Prompting Framework for Text Generation Large language models (LLMs) have shown remarkable success across a wide range of natural language generation tasks, where proper prompt designs make great impacts. While existing prompting methods are

For machine translation, how does GPT stack up against MT-specific models such as Microsoft/Google Translate? We dug around and found: * MT models are better on average and are more robust * GPT is better at some things, like ambiguous expressions * Price is comparable Details↓

We examined the Microsoft GPT-MT dataset (github.com/microsoft/gpt-…). The dataset consists of outputs from GPT-3 with Microsoft Translate and the best systems from the Conference on Machine Translation (www2.statmt.org/wmt23/) over 18 language pairs.

Is your LLaMA still struggling with translation? 🚀ALMA, built on LLaMA-2 with our new MT training recipe, achieves over 12 BLEU & COMET than LLaMA-2's zero-shot and beats all prior works, even NLLB-54B & GPT-3.5 with 7B/13B parameters! Paper: arxiv.org/pdf/2309.11674…

A Paradigm Shift in Machine Translation: Boosting Translation Performance of Large Language Models paper page: huggingface.co/papers/2309.11… Generative Large Language Models (LLMs) have achieved remarkable advancements in various NLP tasks. However, these advances have not been

Very proud that ALMA has been accepted at ICLR 2024! JHU CLSP Microsoft Research arxiv.org/abs/2309.11674

I am super excited about BLIP3, a very strong multimodal model that can reason about interleaved image text sequences! I had a ton of fun working with Le Xue and team on this.

Updated phi-3 tech report with final numbers for 7B/14B and a new section on phi-3-V (e.g., MMMU at 40.4, in the ballpark of Claude 3-haiku and Gemini-1.0 pro) : arxiv.org/abs/2404.14219

Introducing Phi-4-multimodal and Phi-4-mini! Phi-4-multimodal integrates speech, vision, and text processing, while Phi-4-mini excels in text-based tasks. Discover these models on Azure AI Foundry: msft.it/6019U7H4D

More big updates today for our Phi family of SLMs: Phi-4 multimodal and Phi-4 mini. Can't wait to see what you build. azure.microsoft.com/en-us/blog/emp…

We released Phi-4-mini (3.8B base in LLM), a new SLM excelling in language, vision, and audio through a mixture-of-LoRA, uniting three modalities in one model. I am so impressed with its new audio capability. I hope you can play with it and share with us your feedback. We also

Congratulations to Microsoft on the launch of Phi-4-multimodal and Phi-4-mini. 🎉 This new generation of SLMs introduces image and audio capabilities to the Phi family and provides high quality, cost-effective models to run in more places. Read ➡️ nvda.ws/4h39SUi

Phi-4-Mini Technical Report Compact yet Powerful Multimodal Language Models via Mixture-of-LoRAs