introducing an opensource TTS model Masked Generative Codec Transformer (MaskGCT)!!!, https://
arxiv.org/abs/2409.00750 https://
github.com/open-mmlab/Amp
hion/tree/main/models/tts/maskgct
…
@jiqizhixin
-
Open Source TTS Model MaskGCT Released
By
–
-
Non-Autoregressive TTS Model Eliminates Need for Alignment
By
–
a fully non-autoregressive TTS model that eliminates the need for explicit alignment information between text and speech supervision, as well as phone-level duration prediction
-
ROCKET-1: Visual-Temporal Prompting for Open-World Robot Interaction
By
–
ROCKET-1: Master Open-World Interaction with Visual-Temporal Context Prompting https://
arxiv.org/pdf/2410.17856
a low-level policy that predicts actions based on concatenated visual observations and segmentation masks, with real-time object tracking provided by SAM-2. -
MIT KV Cache Framework Reduces LLM Decoding Latency
By
–
MIT HAN Lab introduce a framework that only applies a full KV cache to retrieval heads while using a light-weight, constant-length KV cache for streaming heads, which reduces both LLM's decoding and pre-filling memory and latency without compromising its long-context abilities.
-
DuoAttention: Efficient Long-Context LLM Inference with Retrieval
By
–
DuoAttention: Efficient Long-Context LLM Inference with Retrieval and Streaming Heads https://
arxiv.org/abs/2410.10819 https://
github.com/mit-han-lab/du
o-attention
… #MIT @songhan_mit -
Questioning the Practical Value of Efficient AI Architectures
By
–
These findings raise important questions about the practical value of efficient architectures.
-
Efficient Transformers Struggle with Reasoning Tasks Without More Parameters
By
–
Notably, both papers present negative results, showing that many efficient Transformers will provably struggle with reasoning tasks unless the number of parameters is significantly increased.
-
Efficient Models Show Reasoning Gaps Versus Standard Transformers
By
–
These papers examine the capabilities of efficient models, including Sparse Transformer, Linear Transformer, and Mamba, revealing significant gaps in reasoning tasks compared to standard Transformers.
-
Theoretical Limitations of Efficient Transformers Discovered
By
–
Theoretical limitations of efficient Transformers (Mamba, etc.) are found by PKU and THU groups https://
arxiv.org/abs/2402.13934 (Published at ICML 2024) https://
arxiv.org/abs/2402.18510 -
SageAttention Outperforms FlashAttention2 and Xformers by 2-3x
By
–
The OPS (operations per second) of our approach outperforms FlashAttention2 and xformers by about 2.1x and 2.7x, respectively. SageAttention also achieves superior accuracy performance over FlashAttention3.