iFeeling Daily
Daily curated AI insights you can't miss.
VocabX: Cost-Weighted Vocabulary Adaptation Cuts Inference Cost by 77%
VocabX rebuilds a domain model's tokenizer around inference cost instead of token frequency, cutting measured inference cost by 77% in a production grader.
ZenteiQ scales physical AI inference 4x with GKE and llm-d
With GKE Inference Gateway and llm-d, ZenteiQ scaled inference throughput 4.07x to 69,114 tokens/s per 1,000 accelerators.
Analyzing Unstructured Data in BigQuery with ObjectRef
ObjectRef is now GA in BigQuery, letting teams query unstructured Cloud Storage files as native columns and build multimodal ETL pipelines entirely in SQL.
Profiling TorchNeuron (Part 1): System-Level Profiling on AWS Trainium
System-level profiling with TorchNeuron on Trainium shows how key_averages() and Neuron Explorer trace RMSNorm from framework dispatch to hardware execution.
The GPU Revenue Ladder: Inference Economics From $38 to $2.44
DeepSeek pegs a GPU's theoretical maximum revenue at $38 per card-hour, while SpaceX's realized revenue per card-hour lands at just $2.44 today.
Granite Speech 5.0: New ASR, Translation, and Speaker Attribution
IBM updates Granite Speech with a new 2B model for ASR/translation, a multilingual Turbo CTC, and a revamped 4.1 2B Plus for speaker attribution.
OpenAI Models Are Coming to Google Cloud
OpenAI and Google Cloud are partnering to bring OpenAI frontier models and Codex workflows into Google Cloud’s Model Garden, with general availability planned in the coming weeks.
Google Cloud Infrastructure and Sovereign AI Updates at NVIDIA GTC Berlin 2026
At NVIDIA GTC Berlin 2026, Google Cloud announced A5X instances with NVIDIA Vera Rubin, expanded Sovereign AI with GDC, and new Confidential Computing options.
Confidential G4 GPUs with NVIDIA Blackwell Reach General Availability
Confidential G4 instances powered by NVIDIA RTX PRO 6000 Blackwell GPUs are now generally available on Google Cloud with under 3.5% throughput degradation.