saiteja33/MapRoute-EMMA-mappers

MapRoute EMMA mappers

Trained MapRoute concept mappers (<concept>_100.pth, PyTorch checkpoints) for Stable Diffusion v1.4 on the EMMA concepts. Generated images are in saiteja33/MapRoute-EMMA-images.

config	method	contents
llm-qwen3-8b	MapRoute+llm-qwen3-8B [Prompting]	206 mappers: object, celebrity, art, nsfw, copyright
minirag-llama3.2-1b-it	MapRoute+KG+minirag+llama32-1b	206 EMMA mappers + 20 challenge mappers
minirag-qwen3-8b	MapRoute+KG+minirag (Qwen3-8b)	129 mappers: object, celebrity
kg-deterministic	MapRoute+KG	70 mappers: art, copyright
art	earlier art-style run	40 art mappers

Each row is one mapper: config, domain, concept, filename, relpath (path under mappers/), size_bytes, sha256, and mapper (the raw .pth bytes).

import io, torch
from datasets import load_dataset
ds = load_dataset("saiteja33/MapRoute-EMMA-mappers", "llm-qwen3-8b", split="train")
row = ds[0]
state = torch.load(io.BytesIO(row["mapper"]), map_location="cpu")

manifest.jsonl lists every mapper with its sha256.