EXPLORE / 06
Terms index
Specialized AI and software-development terms with concise explanations
RECORDS
Entity records
3380 records · 53/85
- 2081
FlashAttention
An IO-aware efficient attention algorithm that significantly reduces memory usage.
Website pending - 2082
Sequence Parallelism
A distributed strategy that splits input sequences across multiple devices for parallel computation.
Website pending - 2083
Rollout
The process of generating trajectory data by interacting with the environment using the current policy during RL training.
Website pending - 2084
Group Relative Policy Optimization
An RL algorithm variant that estimates advantages via intra-group relative comparisons without requiring a value network.
Website pending - 2085
Optimizer State
Internal variables recording gradient momentum and other auxiliary info in DL training
Website pending - 2086
Checkpoint
A snapshot of model parameters and optimizer states saved during training
Website pending - 2087
Autoregressive Generation
A generation method that predicts the next element or action step-by-step based on preceding sequences.
Website pending - 2088
VLA
End-to-end multimodal models combining vision, language understanding, and physical action control.
Website pending - 2089
Spearman Correlation
A non-parametric measure of rank correlation that assesses how well the relationship between two variables can be described using a monotonic function.
Website pending - 2090
End-to-End Evaluation
Comprehensive testing method assessing performance and quality from input to final output.
Website pending - 2091
llama.cpp
A C++ library for efficiently running large language models on CPUs.
Website pending - 2092
Inference Providers
A service layer that calls various LLM backends through a unified interface.
Website pending - 2093
Codon Optimization
The process of modifying synonymous codons in a gene sequence to improve expression efficiency or stability.
Website pending - 2094
Perplexity
A metric measuring how well a probability model predicts a sample; lower values indicate better prediction accuracy.
Website pending - 2095
API Stability
A commitment that software interfaces remain backward compatible and unchanged
Website pending - 2096
Hybrid Attention Mask
A masking strategy combining different attention mechanism patterns to optimize context understanding for specific tasks
Website pending - 2097
Open-Vocabulary Segmentation
Identifying and segmenting objects based on arbitrary natural language descriptions, not limited to predefined classes
Website pending - 2098
SSE
An HTTP-based unidirectional communication mechanism often used for real-time updates
Website pending - 2099
Open-Vocabulary
Ability to recognize categories not present in training data
Website pending - 2100
Early Fusion
Combining different modalities at the input stage before processing
Website pending - 2101
Tensor
Multi-dimensional array data structure, basic unit of deep learning computation
Website pending - 2102
Pickle
Python object serialization module with arbitrary code execution risks
Website pending - 2103
Reranker Model
A model that re-scores initial retrieval results to improve relevance ranking.
Website pending - 2104
Real-time Rendering
Technology generating images quickly enough to support immediate user interaction.
Website pending - 2105
Visual Document Retrieval
Retrieving relevant pages or snippets directly from document images.
Website pending - 2106
Reranker
A model that refines the ranking of retrieved candidates to improve precision.
Website pending - 2107
Verifiable Rewards
Algorithmic or rule-based automatic correctness signals for RL feedback
Website pending - 2108
Skill
Pre-defined reusable capability modules or instruction sets for AI agents.
Website pending - 2109
Code Agent
AI systems capable of autonomously writing, debugging, and executing programming tasks.
Website pending - 2110
DAPO
An improved policy optimization algorithm for LLMs designed to enhance training stability and efficiency.
Website pending - 2111
RLVR
Reinforcement learning method that uses rules or external tools to automatically verify output correctness for reward signals.
Website pending - 2112
Forward Pass
The process of computing outputs from inputs in a neural network
Website pending - 2113
Content Script
JavaScript files injected into web pages to manipulate the DOM and execute specific logic.
Website pending - 2114
Service Worker
A script that runs in the background of a web application, handling asynchronous tasks independently.
Website pending - 2115
Benchmark Overfitting
Phenomenon where models optimize specifically for test sets, degrading real-world performance
Website pending - 2116
Serverless Inference
Model serving architecture that auto-scales on demand without managing underlying infrastructure.
Website pending - 2117
Benchmaxxing
Over-optimizing models specifically for test datasets to boost scores rather than improving general capability.
Website pending - 2118
Policy Ratio
Ratio of probabilities under new and old policies for the same action, central to PPO updates
Website pending - 2119
Logprobs
Natural logarithm of probabilities for each output token, used for loss calculation and policy gradients
Website pending - 2120
Train-Inference Mismatch
Discrepancy between probability distributions used in training and actual sampling distributions in inference
Website pending