No description
Find a file
2026-11-01 22:11:02 +01:00
eisheater Fix rebase 2026-11-01 22:11:02 +01:00
pyproject.toml Rebrand and better integrate llmeis (EisHeater now) 2026-11-01 21:40:33 +01:00
README.md Rebrand and better integrate llmeis (EisHeater now) 2026-11-01 21:40:33 +01:00

EisHeater

Identify the equivalent circuit model (ECM) of EIS spectra with a Pydantic AI agent.

For each spectrum in a kisstype dataset tar, EisHeater renders Nyquist and Bode plots, resamples the spectrum to 20 log-spaced points, and hands all three forms to a multimodal LLM. The LLM reasons over several turns, calling tools to fit candidate circuits (eisfit) and to simulate them on the resampled grid (eisgenerator), then returns a structured answer. Each spectrum gets a JSON with the circuit string, a confidence, a short description of what the spectrum is believed to measure, and the parameters of a final eisfit.

Requirements

  • Python >= 3.11
  • Any OpenAI compatible llm endpoint serving a multimodal model
  • kiss python bindings:
    • libkisstypepy
    • libeisfitpy
    • eisgeneratorpy

Setup

pip install -e .

Usage

eisheater <dataset> --model <model-name> --endpoint http://<host>:8080/v1 -o results

options (see --help for all):

flag default meaning
--model (required) model name to request from the server
--endpoint http://localhost:8080/v1 OpenAI-compatible base URL
-o, --outdir results output directory
--limit N 0 process only the first N spectra (0 = all)
--max-tokens 65536 max completion tokens per turn (reasoning models need headroom)
--request-timeout 3600 per-request read timeout in seconds
--concurrency 1 spectra processed in parallel per backend
--resume off reuse the outdir's results and only reprocess the errored and new spectra
--turns off also write a per-spectrum <name>.turns.json with the agent's turn history
--turns-stdout off stream the agent's turns (reasoning, tool calls, returns) to stdout as it runs