# liteLLM

## Docs

- [Compare LLM Ratings Against Human Norms](https://dcpma.mintlify.app/analysis-vs-human.md): Use the OASIS-LLM Analysis dashboard page to compare model valence and arousal ratings against the Kurdi et al. 2017 human norms with eight statistical views.
- [Cache Buster: Forcing Decoding Variance at Temperature 0](https://dcpma.mintlify.app/cache-buster.md): How OASIS-LLM appends per-sample salts to force decoding-path divergence without breaking KV prefix caching for the image and instruction prefix.
- [RunConfig Reference: All Configuration Fields](https://dcpma.mintlify.app/configuration.md): Every YAML RunConfig field in OASIS-LLM: provider, model, image_set, sampling, and prompt settings, plus which changes invalidate a run's canonical hash.
- [Cost Estimation: From Empirical Model to Token Pricing](https://dcpma.mintlify.app/cost-estimation-saga.md): Why the Phase 4a empirical cost model was replaced with calibrated token-times-pricing, and what n=10,598 historical trials taught us about estimation.
- [Cost and Latency: Planning Your LLM Rating Runs](https://dcpma.mintlify.app/cost-latency.md): Understand the real cost and latency of OASIS-LLM runs, with pilot benchmarks, full-set extrapolations, and a budget calculation formula.
- [How OASIS-LLM Documents Research Discoveries](https://dcpma.mintlify.app/discoveries-methodology.md): The protocol behind OASIS-LLM discovery pages: reproduce the bug, read primary sources, contradict obvious theories, quantify, and ship the smallest fix.
- [Known Divergences from the OASIS Human Protocol](https://dcpma.mintlify.app/discrepancies.md): OASIS-LLM differs from Kurdi et al. 2017 in six ways. This page documents each divergence and what it means for interpreting your results.
- [Experiment Design: Valence and Arousal Ratings](https://dcpma.mintlify.app/experiment-design.md): How OASIS-LLM replicates the Kurdi et al. 2017 procedure with 7-point scales and paper-verbatim prompts, and where it diverges from the original human study.
- [Glossary of OASIS-LLM Terms and Technical Concepts](https://dcpma.mintlify.app/glossary.md): Definitions for key OASIS-LLM terms: valence, arousal, OASIS, ICC, Spearman-Brown prophecy, KV prefix caching, canonical hash, and JSON schema strict mode.
- [Image Sets: Choose and Customize Your Stimulus Pool](https://dcpma.mintlify.app/image-set.md): Use built-in OASIS image subsets or supply a custom list. Covers named sets, category-stratified sampling, and why image_set changes require a new run name.
- [OASIS-LLM: Rate Images with Vision-Language Models](https://dcpma.mintlify.app/introduction.md): OASIS-LLM replays the OASIS affective rating procedure using LLMs and compares model valence and arousal ratings to 822 human raters across 900 OASIS images.
- [Ollama Operations: Debugging the 60-Second Timeout](https://dcpma.mintlify.app/ollama-operations.md): How a 'Stopping…' deadlock, macOS unified-memory pressure, and a confirmed Flash Attention bug stacked to cause 60-second timeouts in local Ollama runs.
- [Get Started with OASIS-LLM](https://dcpma.mintlify.app/quickstart.md): Install OASIS-LLM, configure your provider API key, and launch a 100-trial pilot rating run against a vision-language model in minutes.
- [Reasoning Capture: Fix for Gemma 4 Schema Failures](https://dcpma.mintlify.app/reasoning-capture.md): How required-schema JSON reasoning broke smaller vision models and the prompt-rewrite fix that restored reliable output from Gemma 4 and similar models.
- [Run Lifecycle: From YAML Config to Completed Trials](https://dcpma.mintlify.app/workflow.md): How OASIS-LLM processes a run from a YAML config through trial enqueueing, async worker execution, and final result storage in DuckDB.