Researchers introduced OCGQuant, a post-training quantization method that pairs outlier channels with low-magnitude companions to reduce quantization error in NVFP4 low-bit inference, achieving the lo…
CrowdTensor, an Apache-2.0 open-source protocol for checkpointed volunteer model-training campaigns, completed a feasibility run using Qwen/Qwen2.5-7B-Instruct and openai/gsm8k, achieving an accuracy …
A developer built RasavedaGPT, a 6.4-million-parameter decoder-only transformer from scratch, achieving a loss drop from 2.439 to 0.787 after fine-tuning on 2,139 recipe examples. The model, designed …
A developer trained a 6.4M-parameter decoder-only transformer from scratch in PyTorch on a single Colab T4 to power their recipe app Rasaveda, replacing external APIs with a custom model that runs inf…
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…
Researchers propose CONCORD, an asynchronous sparse aggregation framework for device-cloud retrieval-augmented generation under document isolation, which improves throughput by up to 2.15x and reduces…
Researchers introduced generic triple-latent sequence models that maintain a running token state and compressed pair-memory pathway to capture higher-order token interactions without benchmark-specifi…