SDXL Performance on Low VRAM
Experiments on a Google Colab T4 with a 16 GB GPU show that SDXL's low-VRAM performance depends more on the granularity of runtime offloading than on the nominal memory budget, with finer-grained poli…
Experiments on a Google Colab T4 with a 16 GB GPU show that SDXL's low-VRAM performance depends more on the granularity of runtime offloading than on the nominal memory budget, with finer-grained poli…
Hume AI's new research introduces three tests to quantify benchmark optimization in speech recognition, finding that several of 11 open-source ASR models reproduced benchmark transcripts from VoxPopul…
Hugging Face has built a hybrid search system for Papers with Code that combines PostgreSQL full-text search with pgvector dense embeddings, using reciprocal rank fusion (RRF) to merge results. The sy…
A new approach to generating visual datasets uses script-generated images with reproducible metadata and Adobe Generator tools, enabling controlled evaluation and diagnostics for machine learning mode…
Promptfoo is recommended as the primary evaluation tool for AI agents focused on documentation and RAG tasks, with Ragas and LangSmith suggested for deeper analysis. The guidance from Promptfoo, Huggi…
Liquid AI released DSpark draft model checkpoints for its LFM2.5 family, claiming up to 3.18x throughput improvement on GPU and up to 2.87x on-device, with day-one support for llama.cpp and SGLang. Th…
A Hugging Face forum user asked how to set up an agent team without a Claude subscription, and a responder explained that API-based or local models can be used with a coordinator prompt, worker prompt…
A new measurement study across four models finds that the default LoRA target_modules of q_proj and v_proj underperforms a data-driven placement for capability retention: on Llama-3-8B fine-tuning cod…
Phillip Linstrum, drawing on over a decade in regulated healthcare quality systems, built Project Shadow, an AI governance architecture that separates observation, classification, authority, execution…
A developer is beta-testing an OpenAI-compatible API gateway for Chinese-origin open-source large language models, offering limited free token quota to overseas developers in exchange for feedback and…
A developer survey on Hacker News asks practitioners where they draw the line between retrieval-augmented generation (RAG) and fine-tuning in production large language model (LLM) applications, highli…
Developers report that LLM-powered chatbots still struggle to maintain coherent conversations over multiple turns, with issues such as memory loss and context confusion emerging in longer interactions…
A Hugging Face community member asked developers to share their biggest challenges moving LLM and generative AI projects from prototype to production, citing issues like reliability and evaluation. Th…
Developer kafidog released AI Content Factory, an open-source, offline-first content production pipeline that turns a single topic into research, articles, scripts, storyboards, media plans, and platf…
AIIDE, an AI API platform, is seeking developers and AI builders to test its unified API for accessing multiple AI models. The platform aims to simplify integration for AI SaaS, agents, automation too…
Docker's model pull command fails with an 'insufficient_scope: authorization failed' error because the command resolves against Docker Hub (docker.io) instead of Hugging Face, according to a technical…
Liquid AI released Q4_0 checkpoints for its LFM2.5 language models trained with quantization-aware distillation (QAD), recovering 96.5% to 97.4% of the accuracy lost to quantization across four models…
As AI projects move from experimentation to production, teams are questioning whether API governance for AI and inference workloads requires different handling than traditional web APIs, according to …
A developer is seeking open-source replacements for Llama 3.3 70B Versatile after Groq shut down the model, focusing on smaller parameter sizes with comparable performance for RAG, agentic workflows, …
IBM Research's ALT K-Evolve framework shows that the optimal amount of agentic memory varies by model capability, with strong models like DeepSeek-V3.2 (671B MoE) gaining +9.5 percentage points in tas…