sayshara/simple-ascii-art-embeddings

Simple ASCII Art txtai Embeddings

A txtai embeddings database built from `PinkPixel/ASCII-Art`, transformed for semantic retrieval.

• —Indexed column: text — the system/user prompt text with System: and User: labels removed.
• —Return column: ascii-art — assistant ASCII art with surrounding Markdown code fences stripped.
• —Embedding model: ibm-granite/granite-embedding-english-r2
• —Dimensions: 768
• —Rows: 1,221

Load and search

import json
from txtai.embeddings import Embeddings

repo_dir = "txtai_ascii_art_embeddings"

metadata = {}
with open(f"{repo_dir}/metadata.jsonl", "r", encoding="utf-8") as handle:
    for line in handle:
        row = json.loads(line)
        metadata[row["id"]] = row

embeddings = Embeddings()
embeddings.load(repo_dir)

for result in embeddings.search("a cute kitten sitting down", limit=3):
    row = metadata[str(result["id"])]
    print(result["score"])
    print(row["text"])
    print(row["ascii-art"])

Files

• —config.json, embeddings, documents: txtai database files.
• —metadata.jsonl: id-aligned prompt and ASCII-art records.
• —embedding_config.json: source/model/build metadata.