EXPLORE / 06
Terms index
Specialized AI and software-development terms with concise explanations
Entity records
4629 records · 85/116
- 3361
Forward Pass
The process of computing outputs from inputs in a neural network
Website pending - 3362
Content Script
JavaScript files injected into web pages to manipulate the DOM and execute specific logic.
Website pending - 3363
Service Worker
A script that runs in the background of a web application, handling asynchronous tasks independently.
Website pending - 3364
Benchmark Overfitting
Phenomenon where models optimize specifically for test sets, degrading real-world performance
Website pending - 3365
Serverless Inference
Model serving architecture that auto-scales on demand without managing underlying infrastructure.
Website pending - 3366
Benchmaxxing
Over-optimizing models specifically for test datasets to boost scores rather than improving general capability.
Website pending - 3367
Policy Ratio
Ratio of probabilities under new and old policies for the same action, central to PPO updates
Website pending - 3368
Logprobs
Natural logarithm of probabilities for each output token, used for loss calculation and policy gradients
Website pending - 3369
Train-Inference Mismatch
Discrepancy between probability distributions used in training and actual sampling distributions in inference
Website pending - 3370
ModernBERT
A new generation bidirectional Transformer architecture optimized for long context and efficiency
Website pending - 3371
MTEB Benchmark
A standardized benchmark suite evaluating text embedding models across various downstream tasks like retrieval and classification.
Website pending - 3372
Matryoshka Representation Learning
A technique allowing users to truncate embedding dimensions to trade off accuracy for computational efficiency within a single model.
Website pending - 3373
Inference Backend
Underlying software engine responsible for executing model computations to generate predictions
Website pending - 3374
CrossEncoder
Neural network structure that processes query and document together to output a relevance score.
Website pending - 3375
Reranking
Technique that refines initial retrieval results by scoring them for optimal final ordering.
Website pending - 3376
Delta Update
An optimization strategy that transmits only changed data parts rather than the full dataset.
Website pending - 3377
Harness
The infrastructure layer responsible for managing model execution, tool invocation, and state maintenance.
Website pending - 3378
Parameter Delta
The difference in model weights between two consecutive optimization steps.
Website pending - 3379
Weight Synchronization
The process of transmitting updated model parameters from the training side to the inference side.
Website pending - 3380
Asynchronous RL
An RL paradigm where trainer and inference engine work in parallel without strict synchronous waiting.
Website pending - 3381
Cascaded Pipeline
An architecture chaining multiple independent models sequentially to complete complex tasks.
Website pending - 3382
Scaffold
Structured code or toolsets used to support and simplify complex processes in software development or model training.
Website pending - 3383
Computer Use Agent
An agent that controls a computer via mouse and keyboard based on visual interface recognition, mimicking human operation.
Website pending - 3384
Rejection Sampling
A data generation technique that filters candidate samples meeting specific criteria
Website pending - 3385
Direct Preference Optimization (DPO)
A method to optimize language models directly using preference data without training a separate reward model
Website pending - 3386
Reasoning Traces
Records of the logical steps a model takes from input to final conclusion, used to explain decision-making processes.
Website pending - 3387
RFC
A document mechanism used to solicit community feedback on technical designs or standards
Website pending - 3388
Gaussian Splatting
A 3D scene representation technique for real-time radiance field rendering
Website pending - 3389
CI
Automated build and test practice for code changes
Website pending - 3390
Operator Fusion
Combining multiple consecutive operations into a single kernel to reduce memory access and launch overhead
Website pending - 3391
Environment Interface
Standardized data formats and control protocols defining how agents interact with external environments.
Website pending - 3392
Resource Discovery
The process of automatically identifying, locating, and retrieving available services or components
Website pending - 3393
OFT
Efficient tuning technique using orthogonal transformations to preserve pretrained feature space structure
Website pending - 3394
PEFT
Techniques updating few parameters to significantly reduce memory needs for LLM training
Website pending - 3395
Privacy-Aware Reinforcement Learning
A machine learning approach that incorporates privacy protection constraints into the reward function to optimize policies.
Website pending - 3396
Multi-hop Reasoning
A reasoning process requiring the connection of multiple dispersed facts or steps to derive a final answer.
Website pending - 3397
Deep Research Agent
An AI system capable of autonomously planning, retrieving, and synthesizing multi-source information to solve complex problems.
Website pending - 3398
Local Models
LLMs running on user-owned hardware without relying on cloud APIs.
Website pending - 3399
Model Parameters
Total number of trainable weights in a neural network, affecting size and complexity.
Website pending - 3400
Cross-Origin Storage
A proposed Web API allowing different origins to share storage resources.
Website pending