retrieval augmentation, prompting techniques
Retrieval augmentation supplements prompts with relevant information from knowledge bases.
536 technical terms and definitions
Retrieval augmentation supplements prompts with relevant information from knowledge bases.
General term for enhancing LLM outputs with retrieved external information.
Retrieval heads identify which retrieved documents were actually used.
Specialized attention head for deciding when to retrieve.
Time to retrieve documents.
Evaluate retrieval quality.
Retrieval precision measures fraction of retrieved documents that are relevant.
Retrieval recall measures fraction of relevant documents that are retrieved.
Queries processed per second.
Full RAG system.
Use retrieved documents to inform generation.
Select relevant examples from pool using similarity.
Parts of answer not requiring retrieval.
Alternate between retrieval and generation multiple times.
Architecture designed for retrieval augmentation.
Retrograde wells have peak doping below surface providing better punchthrough resistance and reduced junction capacitance.
Well with higher doping deeper down.
Regular retros on AI projects. What worked, what did not, what to improve. Continuous learning culture.
Plan synthetic routes backward from target.
Retry logic attempts failed requests again with exponential backoff.
Retry on transient failures with exponential backoff. Jitter prevents thundering herd.
Automatically retry failed operations.
Retry failed requests with exponential backoff. Add jitter to avoid thundering herd. Set max retries and timeouts.
Return lines recirculate unused chemicals back to reservoirs.
Return loss measures signal reflection magnitude indicating impedance matching quality.
Analyze returned failed units.
Send defective material back to supplier.
Increase sales over time.
Reduce leakage by raising threshold.
Reverse body bias raises threshold voltage reducing leakage at cost of performance.
Bond to substrate first.
Reverse osmosis purifies water by forcing it through semipermeable membranes removing dissolved solids.
Use negative resist with positive mask (or vice versa).
Invertible transformations saving memory.
High-resolution follow-up of detected defects.
Reward hacking finds unintended ways to maximize reward without intended behavior.
Learn reward function from feedback.
Reward model scores outputs based on human preferences. Train from comparisons. Used in RLHF.
Reward model scores outputs by human preference. Trained on comparison data. Guides RL fine-tuning.
Reward modeling predicts human preferences using learned reward functions.
Learn reward function from human preferences.
Reward shaping modifies reward functions to provide denser feedback without changing optimal policy.
Reweighting adjusts training example importance to balance group representation.
Rework repeats process steps on wafers to correct defects or meet specifications.
Reprocess material to meet specs.
Reprocess failed wafers.
Model high-frequency behavior of devices.
Radio frequency power applied to generate and sustain plasma.
RF probe cards enable high-frequency electrical testing of semiconductor devices by maintaining signal integrity through controlled impedance contacts.
Use RF power to sputter insulating targets.