AI Briefing#text-generation
The latest AI and developer news about #text-generation, with the original source and a short summary.
The latest AI and developer news about #text-generation, with the original source and a short summary.
NCM uses a two-stage coarse-to-fine framework to generate text in parallel, outperforming diffusion language models in speed and perplexity on standard benchmarks.
Apple researchers implemented 4-step text generation using a 1.7B CFM model.
DiffusionGemma adapts Gemma 4 into a discrete diffusion model, achieving 1,500 TPS on H100.
MDLM improved agentic RL performance as a stronger text-based world model than AR.
NVIDIA has released LogitsProcessorZoo, a Hugging Face-compatible library for controlling LLM generation.
Introducing Self-Speculative Decoding, a technique that leverages the early layers of the same model to improve text generation speed and memory efficiency.
This introduces how to use a text generation pipeline based on Optimum Habana to efficiently run Llama 2 models on Intel Gaudi 2 accelerators.
A guide summarizing Hugging Face's open source LLM models, licenses, serving tools, and PEFT techniques.
It introduces a new decoding technique that reduces text generation latency by up to 10x by solving the memory bandwidth problem, a bottleneck in LLM inference.
Contrastive Search, which solves the text repetition problem, has been introduced to the Transformers library.
By combining TensorFlow and XLA in Hugging Face's `transformers`, text generation speed can be increased by up to 100x.
It introduces the Constrained Beam Search feature in the Transformers library, which controls text generation to include specific words.
AI21 Labs has unveiled its developer platform AI21 Studio and Jurassic-1.