global memory, hardware
Main GPU memory.
9,967 technical terms and definitions
Main GPU memory.
Global pooling aggregates all node features into graph-level representations using operations like sum mean or attention-weighted averaging.
Global variation affects entire die uniformly from wafer-level process differences.
Hybrid timing approach.
PyTorch's collective communication library.
I can act as a living glossary: define terms, connect them to each other, and give concrete examples.
Glove boxes provide isolated atmospheres for handling sensitive materials.
Bulk impurity analysis.
Glow-TTS combines normalizing flows with monotonic alignment for fast high-quality speech synthesis.
Different gated activation functions.
Gating mechanism for sequences.
Suite of language understanding tasks.
Standard language understanding evaluation.
General Language Understanding Evaluation tests models on diverse NLP tasks.
Generalized Matrix Factorization applies element-wise product followed by neural layers generalizing linear matrix factorization models.
MLP with gating for language.
Gated MLP for images.
Graph Multiset Transformer uses multiset attention and virtual nodes for expressive graph-level representation learning.
GNN expressiveness theory studies which graph properties can be distinguished by different message passing architectures.
Go-Explore maintains archive of interesting states and returns to them enabling exploration of sparse reward environments.
Goal achievement detection recognizes when objectives are satisfied enabling termination.
Goal stacks organize objectives hierarchically tracking completion dependencies.
Learn to reach specified goals.
Goal-conditioned reinforcement learning trains agents to reach diverse goals specified as part of state representation.
GOAT is arithmetic fine-tuned LLM. Perfect on many arithmetic tasks.
Find classes doing too much.
High-quality reference annotations.
Use gold wire.
Best-performing chamber used as reference.
Reference wafer with known good properties for calibration.
Goodbye! Feel free to come back anytime with questions about AI, chips, or LLMs.
Goodness-of-fit tests compare observed distributions to theoretical expectations.
DeepMind's large language model.
LLM trained to use APIs and tools effectively.
Gorilla is LLM trained for API calling. Reduces hallucination on APIs.
Gowning procedures specify sequence for donning cleanroom apparel.
Protocols for cleanroom garments.
Area where workers change into cleanroom suits before entering fab.
Autoregressive model family for text generation.
GPT Engineer generates codebases from specs. Agentic coding.
GPT-J is EleutherAI 6B model. Popular open model.
GPT-NeoX is EleutherAI 20B model. Open source, open weights.
OpenAI's multimodal large language model.
OpenAI's multimodal GPT-4.
GPT4All runs local models on desktop. Nomic AI. Privacy-focused.
GPTQ is post-training quantization method. Uses calibration data. Popular for 4-bit models.
Parallel processor widely used for deep learning.
Interconnected GPUs for parallel training.
Remote direct memory access for GPUs.
NVIDIA technology for direct data transfer.