Researchers released the official Multi-Teacher On-Policy Distillation (MOPD) paper, which composes multiple capabilities into a single large language model by training domain-expert teachers independ…
Researchers introduced TF-Engram, a train-free memory system that stores phrase-specific semantic knowledge offline across GPU, DRAM, and SSD tiers, and uses predictive prefetching to reduce latency. …
AWS introduced Reverse Direct Preference Optimization (rDPO) as a selective unlearning method for Amazon Nova, enabling customizable content moderation settings that reduce over-deflection in approved…
Researchers developed a vision-language framework using Low-Rank Adaptation (LoRA) on a frozen BiomedCLIP backbone to detect severe adverse events in wound monitoring. The system integrates clinical n…
Researchers developed a spoken question answering system for Luxembourgish using text-to-speech augmentation, training a SLAM-style architecture with frozen Whisper and multilingual LLM backends. Mult…
Researchers propose a reinforcement learning with verifiable rewards (RLVR) method for data-efficient adaptation of audio-language models to code-switched automatic speech recognition (ASR). Using Qwe…
MarkTechPost published a tutorial on training Gemma-3 for structured mathematical reasoning using Tunix GRPO, LoRA adapters, and GSM8K rewards. The workflow includes environment setup, prompt formatti…
A developer implemented a multi-agent architecture using the Orchestrator Worker pattern, where an orchestrator delegates tasks to specialized workers for search and quality checking. The approach, in…
Google researchers have developed the Scalable Cluster Termination System (S-CTS), a multimodal defense that uses LLMs enhanced with LoRA and APO to detect and terminate coordinated bot-nets producing…
Researchers released a benchmark for Arabic-Russian scientific translation, including a 27,000-sentence parallel corpus and fine-tuned multilingual models. The Qwen2.5-7B model achieved BLEU 23.15, ou…
Researchers introduced a three-phase deep reinforcement learning system for personalized portfolio management that overcomes ticker lock-in, monolithic objectives, and static user models. Phase 1 uses…
A developer details practical QLoRA fine-tuning using Axolotl and Unsloth, explaining how parameter-efficient methods like LoRA and QLoRA enable training multi-billion parameter models on a single con…
LoRA (Low-Rank Adaptation) and QLoRA have become widely adopted methods for efficiently fine-tuning large language models with a fraction of the parameters, solving the problem of massive GPU requirem…
The CUHKSZ team submitted a simultaneous speech translation system to IWSLT 2026, built on Qwen3-Omni-30B-A3B with LoRA adaptation, achieving 40.5 BLEU for English→Chinese and 27.7 BLEU for English→Ge…
A new physics-informed AI pipeline uses supervised LoRA fine-tuning on a small language model to reliably output structured engineering data, which a deterministic physics checker then validates. The …
Researchers introduced Hankel Reduced order Model (HRM) adapters, a state space model-based fine-tuning method that outperforms LoRA on long-context tasks like QuALITY and QMSum by leveraging FFT-base…
Researchers developed a methodology to transform Hindi WordNet into 1.25 million instruction-response pairs, fine-tuning a 12B-parameter language model to create a specialized conversational AI for lo…
Researchers at Goodfire demonstrated that fine-tuning a single scalar prefactor on a German-related rank-1 parameter subcomponent of a 67M-parameter language model can destroy its ability to predict G…
NVIDIA has released a GenAI LLM Certification Lab that guides developers through building a production-ready fine-tuning and optimization pipeline. The lab covers data preparation, LoRA fine-tuning wi…
Deltatensors, a new open-source tool, compresses fine-tuned neural network model deltas into small .wdelta files, achieving near-lossless compression with sub-1% perplexity difference. Tested on Qwen2…