Tagged articles
model pruning
12 articles · Page 1 of 1
Analysis of DeepSeek's 92GB pruned model shows it requires 128GB Macs due to macOS GPU memory limits, while multilingual top-1 consistency drops to 65.4%, making it a poor default for 96GB Mac users; the article compares model tiers, pruning trade-offs, SSD offloading penalties, and speculative decoding gains.
DeepSeekGGUFMLX
0 likes · 12 min read

Taote details its end-to-end on-device model pipeline for real-time user intent detection, using Qwen3-0.6B with LoRA fine-tuning, ShortGPT-based layer pruning, and AWQ quantization to achieve 80%+ on-device accuracy, enabling feed re-ranking and retention coupon scenarios with measurable GMV growth.
AWQ quantizationGMV growthLoRA fine-tuning
0 likes · 30 min read

Deploying large AI models to tiny embedded devices like Arduino and Raspberry Pi requires aggressive model slimming through quantization, pruning, and distillation, careful selection of runtimes such as TensorFlow Lite, and addressing power, latency, and debugging challenges to achieve real‑time inference.
ArduinoEmbedded AIRaspberry Pi
0 likes · 7 min read

This article explains the principles, key methods, and practical effects of model quantization, pruning, and knowledge distillation, comparing their advantages and disadvantages, and showing how combining these techniques enables compact, high‑performance AI models on resource‑constrained devices.
edge AIknowledge distillationmodel compression
0 likes · 7 min read

In this detailed engineering guide we diagnose a heavyweight ERNIE‑Base text‑classification service consuming 128 CPU cores and 96 GB RAM, then apply a three‑step optimization—model selection, structured pruning with PaddleSlim, and engine migration to ONNX Runtime—achieving a 75% reduction in resource usage while keeping recall above 99.5% and boosting inference speed by over 20%.
AI Model OptimizationONNX RuntimePaddleSlim
0 likes · 11 min read

Quwan Technology presents its Kaitian social large model, designed for personalized, emotionally rich, multimodal AI interactions, detailing its scene‑specific goals, CPT+SFT+RLHF training pipeline, data desensitization, LoRA fine‑tuning, evaluation methods, pruning, latency trade‑offs, safety mechanisms, and future feedback loops.
AI safetyLoRARLHF
0 likes · 13 min read

This article recounts how JD Retail's young algorithm engineers tackled diverse AI problems—optimizing reward‑model ensembles for ad image generation, building large‑language‑model‑based query expansion, and pruning diffusion models with FFT and RDP—while sharing their technical approaches, code snippets, and growth reflections.
AIAlgorithm Engineeringmodel pruning
0 likes · 14 min read

Young JD algorithm engineers share how they solve tough AI problems—from optimizing large‑model training and reward‑model design for ad image generation, to building LLM‑based query expansion, agent evaluation, and model pruning with FFT and RDP—illustrating practical breakthroughs and personal growth in cutting‑edge AI research.
AIAlgorithm EngineeringReward Modeling
0 likes · 15 min read

This article reviews ten recent papers that tackle the over‑thinking problem in large language models by shortening chain‑of‑thought reasoning, introducing dynamic early‑exit, adaptive thinking triggers, and reinforcement‑learning‑based training, showing how models can maintain or improve accuracy while dramatically reducing token usage and latency.
AI researchadaptive inferencechain-of-thought
0 likes · 38 min read

JD Retail’s engineering team tackles hard AI problems by replacing a monolithic reward model with specialized small models for ad‑image generation, deploying an LLM‑driven query‑expansion pipeline that lifts conversion rates, and pruning text‑to‑image transformers using FFT and RDP to boost throughput 40% without loss, while building comprehensive evaluation tools and a semantic smart‑assistant.
AIReward Modelinglarge models
0 likes · 14 min read

The paper introduces SHARK, a two‑component framework that uses a fast Taylor‑expanded permutation method to prune embedding tables and a frequency‑aware quantization scheme to apply mixed‑precision to embeddings, achieving up to 70% memory reduction and 30% QPS improvement in industrial short‑video and e‑commerce recommendation systems.
Efficiencyembedding compressionlarge-scale AI
0 likes · 8 min read

This article introduces a neuron‑level shared multi‑task learning framework that jointly estimates click‑through rate (CTR) and conversion rate (CVR), discusses the background and advantages of multi‑task learning, reviews classic shared‑bottom models, describes the proposed pruning‑based architecture, and presents experimental results demonstrating its effectiveness in large‑scale recommendation systems.
CTRCVRmodel pruning
0 likes · 11 min read
