Local Ollama TALA code generator
This project trains a local 1.5B-parameter Ollama model to generate TALA code, enabling creation of flowcharts exportable to mermaid and napkin formats for hackathon use; it is currently collecting quality training data and weighing whether the model is feasible with limited snippets versus synthetic data.
What it addresses
Collecting enough quality training data is hard: cloning the official d2 repository yielded only 79 code snippets, leaving the builder unsure how to train with so little data or whether to generate synthetic data instead.
Shared over time
Each entry reflects what was shared on that date. Details may have changed since then.
Local Ollama TALA code generator
This project trains a local 1.5B-parameter Ollama model to generate TALA code, enabling creation of flowcharts exportable to mermaid and napkin formats for hackathon use; it is currently collecting quality training data and weighing whether the model is feasible with limited snippets versus synthetic data.
Problem: Collecting enough quality training data is hard: cloning the official d2 repository yielded only 79 code snippets, leaving the builder unsure how to train with so little data or whether to generate synthetic data instead.
Audience: Hackathon participants who use mermaid/napkin-style flowcharts
Posted by @darkhorse_chenn · Original comment ↗
“i thought like why not we train a local ollama model 1.5 b parameters and make it generate tala code ,so i can develop flowcharts which we can get in mermaid ,napkin which we use for hackathons”