squad
PulseAugur coverage of squad — every cluster mentioning squad across labs, papers, and developer communities, ranked by signal.
1 day(s) with sentiment data
-
User trained small LLM with custom dataset 1-2 years ago
A user reflects on training a small LLM approximately one to two years ago using a custom dataset comprising their Mastodon posts, personal writings, code, and Q&A factoids, supplemented by open datasets like SQuAD. The…
-
SQuad framework slashes Video Transformer compute costs with sub-quadratic attention
Researchers have developed SQuad, a Sub-Quadratic Attention Distillation framework designed to improve the efficiency of Video Diffusion Transformers (DiTs). This new method reduces the computational cost of the self-at…
-
BERT-based QA models assessed for reliability; RoBERTa shows most stability
A new study published on arXiv evaluates the reliability of several BERT-based models, including RoBERTa, ALBERT, and DistilBERT, when applied to question-answering tasks. Researchers assessed model stability by introdu…
-
New RAG verification method improves multi-hop question answering
A new research paper proposes a novel approach to improve verification in retrieval-augmented generation (RAG) systems, particularly for multi-hop question answering. The study demonstrates that traditional per-chunk fi…
-
New cMoLLM architecture scales LLMs via dynamic convolutions
Researchers have introduced cMoLLM, a novel approach to scaling large language models by incorporating a mixture-of-experts (MoE) style throughout the entire model pipeline, rather than just in the feed-forward networks…
-
Microsoft details Agent Framework for disposable AI agents
Microsoft has detailed its Agent Framework, a layered SDK designed for building AI agents. The framework emphasizes disposable agents that can leverage durable memory, enabling complex task execution. This architecture …
-
New RAG research separates context length from semantic competition
A new research paper proposes a method to distinguish between context length and semantic competition as causes for errors in retrieval-augmented generation (RAG) systems. The study introduces a matched-control protocol…
-
New AS-LoRA method improves privacy in federated learning
Researchers have developed AS-LoRA, a novel framework for adaptive selection of LoRA components in privacy-preserving federated learning. This method addresses aggregation errors common in such setups by allowing each l…
-
PACZero enables PAC-private fine-tuning of language models with usable utility
Researchers have developed PACZero, a novel method for fine-tuning large language models that offers strong privacy guarantees. This approach utilizes sign quantization of gradients to achieve a privacy regime where mem…
-
Researchers release Faithfulness-QA dataset to train context-faithful RAG models
Researchers have developed Faithfulness-QA, a new dataset containing nearly 100,000 samples designed to train Retrieval-Augmented Generation (RAG) models to prioritize retrieved context over their internal knowledge. Th…
-
New hardware design offers efficient Softmax and LayerNorm for edge AI
Researchers have developed new hardware-efficient approximations for Softmax and Layer Normalization operations, crucial for Transformer models on edge devices. These methods ensure guaranteed normalization, which is vi…