AI Briefing#lora
The latest AI and developer news about #lora, with the original source and a short summary.
The latest AI and developer news about #lora, with the original source and a short summary.
We propose the K-Merge framework, which continually expands an LLM's capabilities without data even in storage-constrained on-device environments.
PorTAL is a base-agnostic technology that allows task adaptation learned once to be easily transferred to a new LLM.
MiCA, which improves training efficiency by leveraging the minor singular subspace of weight matrices, has been integrated into the PEFT library.
Despite LoRA's high market share, a variety of PEFT techniques should be considered to improve performance.
Research has been published on reducing the gap between LLMs' internal confidence and verbal expression through probe-targeted fine-tuning.
PopuLoRA co-evolved a verifiable self-play curriculum with a teacher-student LoRA population.
Stability AI has released Stable Audio 3.0, an open-weight audio model trained on licensed data.
Stability AI released the Stable Audio 3.0 model family, trained on fully licensed data.
Fine-tuning NVIDIA Cosmos Predict 2.5 with LoRA and DoRA achieves efficient robot video generation in a single-GPU environment.
PINA improves the performance of differentially private clustered FL through LoRA sketching and normality-based aggregation.
AutoAdapt automates the selection and tuning of RAG and fine-tuning within given constraints.
PAW has been released, turning natural language function descriptions into a neural program for local execution.
A 35-domain LoRA routing model based on Qwen3.5-35B-A3B has been released.
An abliteration version targeting the MoE characteristics of Qwen3.6-35B-A3B has been uploaded to HF.
A summary of the performance and limitations of local music generation on Apple Silicon by running ACE-Step 1.5 with MLX.
A diffusion LLM has achieved both AR-level quality and parallel generation speed together.
An experiment interpreting reasoning degradation and catastrophic forgetting as the same kind of structural collapse.
Added 8 Indian languages to Chatterbox-Multilingual using only LoRA and tokenizer expansion.
I-DLM simultaneously boosts DLM quality and speed through introspective consistency.
A tool has been released that supports multimodal fine-tuning of Gemma models, including text, image, and audio, on Apple Silicon environments.
Fujitsu has released OneComp, an open-source Python library that improves the quantization accuracy of LLMs.
LoRA, CFG, and chain-of-thought reduced accent leakage, hallucination, and truncation in LLM-based TTS.
Based on Nova 2.0 Lite, attaching LoRA to only **o_proj** gave the best balance of efficiency and accuracy.
By attaching CPT and BM25 RAG to Qwen3-30B-A3B, the team won the Excellence Award while running in a 24GB environment.