HierSpeech++: All the Amazing Things It Could Do

20 Dec 2024

In this work, we propose HierSpeech++, which achieves a human-level high-quality zero-shot speech synthesis performance.

The Limitations of HierSpeech++ and a Quick Fix

20 Dec 2024

Although our model improves the zero-shot speech synthesis performance significantly, our model also synthesizes the noisy environmental information.

HierSpeech++: How Does It Compare to Vall-E, Natural Speech 2, and StyleTTS2?

20 Dec 2024

We compared the zero-shot TTS performance of our model with Vall-E, NaturalSpeech 2, and StyleTTS 2.

Batching Techniques for LLMs

14 Dec 2024

By reducing the queueing delay and the inefficiencies from padding, the fine-grained batching mechanisms significantly increase the throughput of LLM serving.

LLM Service & Autoregressive Generation: What This Means

14 Dec 2024

Once trained, LLMs are often deployed as a conditional generation service (e.g., completion API [34] or chatbot.

The Generation and Serving Procedures of Typical LLMs: A Quick Explanation

14 Dec 2024

In this section, we describe the generation and serving procedures of typical LLMs and the iteration-level scheduling used in LLM serving.

PagedAttention: An Attention Algorithm Inspired By the Classical Virtual Memory in Operating Systems

14 Dec 2024

To address this problem, we propose PagedAttention, an attention algorithm inspired by the classical virtual memory and paging techniques in operating systems.

The Limitations and Failure Cases of DreamLLM: How Far Can it Go?

28 Nov 2024

While DREAMLLM has made significant strides toward the development of versatile, creative, and foundational MLLMs, it still has several limitations.

DreamLLM: Additional Related Works to Look Out For

28 Nov 2024

This breakthrough garnered a lot of attention and paved the way for further research and development in the field.