humbleworth

/

tld-embedding

like 1

Follow

HumbleWorth 2

Hi HumbleWorth team,

Thanks a lot for releasing the TLD embedding model and the domain-MLM checkpoint.
We’re integrating your models in a domain appraisal pipeline
(Kaggle/Colab + GPU), and we’d like to load humbleworth/tld-embedding as a
drop-in SentenceTransformers model when possible.

Currently, the repo provides model.safetensors and tokenizer files, but it lacks
modules.json and config_sentence_transformers.json (404 on the Hub).
Also, the config has model_type: "dual_task_tld", which suggests a custom
architecture. Could you please provide one of the following?

1. A SentenceTransformers export (i.e., modules.json + config_sentence_transformers.json)
   so we can load via SentenceTransformer(<local_dir>), or
2. The minimal custom class / loader code for "dual_task_tld" in Transformers,
   and the expected MASK token / special tokens behavior to replicate your inference.

Repo links:

• TLD embeddings: https://huggingface.co/humbleworth/tld-embedding

Many thanks for your great work!