AI Weekly Malaysia

IBM Research published a Granite Timeseries Ensemble recipe on Hugging Face that combines four Granite time-series models behind one interface and returns point plus quantile forecasts, with no retraining and no fitted ensemble weights. You use it via `pip install "granite-tsfm>=0.3.11"` and `QuantileEnsembleForecaster.from_pretrained("ibm-granite/granite-timeseries-ensemble-r1")`, with the demo running context_length 1024 and prediction_length 24 on the ETTh1 dataset. Two variants are offered: a default Probabilistic Ensemble using uniform probability pooling, and an adaptive IQR Ensemble; the recipe stores only the ensemble configuration, so the four member checkpoints are downloaded at runtime on first run.

Why: If you're shipping forecasting into a product, this gives you a permissively licensed default that removes the 'which model do I pick' step for smooth industrial signals, noisy retail demand, and long seasonal horizons alike. The practical catch is stated in the post itself: because the recipe holds configuration rather than merged weights, your first invocation pulls four checkpoints, so budget for that in cold-start, serverless, or CI environments rather than assuming a single small download. The text contains no Malaysian or Southeast Asian angle, so treat this purely as a tooling decision.

IBM's Granite team published three speech models: the new granite-speech-5.0-2b, plus updated granite-speech-5.0-470m-turboctc and granite-speech-4.1-2b-plus. The 2B model handles transcription in English, French, German, Spanish, Portuguese and Japanese, translates to and from English in those languages plus English-to-Italian and Mandarin, and was trained on 390,000 hours of public audio with synthetic data added for Japanese ASR, keyword biasing and translation. New controls include language-ID prompting, keyword lists of up to 2,000 terms, and improved long-form and noisy/distant-microphone performance, while the 4.1-2b-plus update targets speaker attribution and timestamps for multi-speaker recordings.

Why: Check the language list before you plan anything: there is no Bahasa Malaysia, Indonesian, Thai or Vietnamese here, so a Malaysian-facing transcription feature would need fine-tuning or a commercial API instead. If your audio is English or Japanese, the two concrete levers are the 2,000-term keyword biasing (useful for product names, acronyms and local brand terms) and the 470M turboctc variant for on-device or high-throughput work where a 2B model is too heavy. The post gives no benchmark numbers, licence terms or pricing, so pilot it on your own audio before committing rather than trusting the 'more accurate and faster' framing.