Local LLMs & agentsProject record · Sep 2026

Local LLM Inference & Agents

Event-ending inference with Qwen

A local-Qwen decision pipeline asks whether the real-world event behind a prediction market is likely to end within a chosen horizon.

Python standard libraryQwenvLLMHTTP APIsJSONConcurrent futuresEvaluation
Why it sits here. High relevance to local LLMs and markets: inspectable live-fact pipeline and exploratory comparison; mixed authorship and small-sample limits constrain outcome claims.
04 / Follow a thought through the machine.STUDY IN SPACE

01 / IMPLEMENTATION & CONTRIBUTION

What the work involves

A shared implementation combines exchange facts, companion-market context, local-model calls, and comparison with a rule baseline. Donald contributed configurable private serving; the source also records jd contributions.

Technical depth

The pipeline groups related markets around a real-world event, hides the target moneyline, separates official feed facts from market-price evidence, and supports monotonic multi-horizon outputs. Uncertainty is an explicit downstream risk state rather than forced certainty.

The project family

market-ending

02 / RESULTS

What came out of it

A dated 25-game exploratory comparison reported 22 judged-correct decisions and three unsure. The small sample and market-close labels do not establish general accuracy or economic value.

03 / SUPPORTING EVIDENCE

Follow the source

Implementation notes, project records, and supporting artifacts.

Source context & project scope

Small opportunistic sample with mostly straightforward positives; no robust generalization or economic-value claim.

The documented ground truth uses market close time, which should not be assumed identical to real event end without further validation.

Cross-exchange game identity is not fully unified.

Pipeline, distinctions between facts and prices, target moneyline exclusion, small dated benchmark, and limits.

SOURCE · 2026-09-17

Gathering, formatting, single/multi-horizon inference, and evaluation implementation.

SOURCE · 2026-09-17

Earlier facts-only model/rule comparison.

SOURCE · 2026-09-17

Mixed contributor history; Donald serving configuration and jd prompt revision.

SOURCE · 2026-09-17
CONTINUE IN LOCAL LLM INFERENCE & AGENTS

Semantic research agents