AI Factory Glossary

torchscript, infrastructure

PyTorch's JIT compiler.

torchscript, model optimization

TorchScript creates serializable and optimizable representations of PyTorch models.

torchscript,jit,trace

TorchScript compiles PyTorch to graph. Trace or script. Enables deployment without Python.

torchserve,pytorch,serving

TorchServe serves PyTorch models. Production deployment.

total cost ownership, supply chain & logistics

Total cost of ownership includes purchase price plus logistics inventory quality and risk costs.

total jitter, signal & power integrity

Total jitter combines random and deterministic components at specified bit error rate.

total productive maintenance, tpm, production

Holistic maintenance approach.

total reflection x-ray fluorescence, txrf, metrology

Surface metal contamination.

total thickness variation, ttv, metrology

Maximum thickness difference on wafer.

touchdown detection, advanced test & probe

Touchdown detection senses when probe tips contact wafer surface through sudden resistance or capacitance change.

toxic exhaust,facility

Separate exhaust for toxic or flammable gases.

toxicity bias, evaluation

Toxicity bias occurs when models generate more toxic content for certain groups.

toxicity classifier,ai safety

Model trained to detect harmful language.

toxicity detection models, ai safety

Identify toxic content.

toxicity detection, ai safety

Toxicity detection identifies offensive abusive or hateful language in text.

toxicity detection,ai safety

Classify text for hate speech offensive language or harmful content.

toxicity filtering, data quality

Remove harmful content.

toxicity prediction, healthcare ai

Predict toxic effects of compounds.

tpm, tpm, manufacturing operations

Total Productive Maintenance maximizes equipment effectiveness through operator involvement and proactive maintenance.

tpu (tensor processing unit),tpu,tensor processing unit,hardware

Google's custom ASIC for ML workloads.

tpu,google,tensor

TPUs are Google custom chips for ML. Available on GCP. Optimized for large batch training and inference.

trace analysis, optimization

Examine execution timeline.

trace data,automation

Detailed time-series data from tool sensors during processing.

traceability (measurement),traceability,measurement,quality

Link measurements to standards.

traceability,quality

Track chips from wafer to customer for quality and recalls.

tracin, explainable ai

Efficient influence computation.

tracin, interpretability

TracIn traces model predictions back to influential training examples through gradient similarity.

trades, trades, ai safety

Balance accuracy and robustness.

traffic splitting,deployment

Route percentage of traffic to different versions.

trailing edge / mature node,industry

Older larger process nodes still used for cost-sensitive products.

trailing-edge node, business & strategy

Trailing-edge nodes are mature processes offering stability and cost advantages.

training compute budget, planning

Total FLOPS for training.

training cost estimation, planning

Predict computational cost.

training cost,model training

Total compute and time required to train a model.

training data attribution, interpretability

Training data attribution identifies which training examples most influenced specific predictions.

training data extraction attack,ai safety

Attempt to extract memorized training examples from model.

training data quality vs quantity, data quality

Tradeoff between data aspects.

training efficiency metrics, optimization

Measure computational efficiency.

training job orchestration, infrastructure

Manage complex training workflows.

training on thousands of gpus, distributed training

Massive-scale distributed training.

training pipeline optimization, optimization

Optimize end-to-end training workflow.

training time prediction, planning

Estimate training duration.

training verification, quality & reliability

Training verification confirms personnel understand and can execute procedures.

trajectory buffer, reinforcement learning advanced

Trajectory buffers in offline RL store complete episodes preserving temporal structure for algorithms requiring sequential context.

trajectory convolution, video understanding

Convolution along motion trajectories.

trajectory prediction,computer vision

Predict future paths of agents.

transductive learning,few-shot learning

Use test set structure during prediction.

transductive transfer learning, transfer learning

Use unlabeled target data during training.

transe, graph neural networks

TransE embeds knowledge graph triples so that head plus relation approximates tail in vector space.

transe,graph neural networks

Translational embedding for KGs.