31 lines
771 B
Markdown
31 lines
771 B
Markdown
|
---
|
||
|
|
language:
|
||
|
|
- en
|
||
|
|
- zh
|
||
|
|
tags:
|
||
|
|
- causal-lm
|
||
|
|
- language-model
|
||
|
|
- babylm
|
||
|
|
- babylm-2026
|
||
|
|
- multilingual
|
||
|
|
- gpt2
|
||
|
|
- paradigmfinder
|
||
|
|
pipeline_tag: text-generation
|
||
|
|
library_name: transformers
|
||
|
|
---
|
||
|
|
|
||
|
|
# gpt2_parfind_en_zh_equal
|
||
|
|
|
||
|
|
Bilingual GPT-2 model packaged for the BabyLM 2026 multilingual evaluation track.
|
||
|
|
|
||
|
|
Source experiment: `/home/achille.fusco/pr_baby_lm/BabyLM_2026_ENH/04-experiments/model_gpt2_ParFindFast_eng_zho_BD_budget16k_zhchildes_v1.0`
|
||
|
|
|
||
|
|
Hugging Face target repo: `NeTSlab/gpt2_parfind_en_zh_equal`
|
||
|
|
|
||
|
|
Current status on July 6, 2026: `resume_running`
|
||
|
|
|
||
|
|
Notes:
|
||
|
|
- `main` is intended to point to the final 1000M checkpoint (`epoch_9`).
|
||
|
|
- The custom ParadigmFinder tokenizer is bundled with the model files.
|
||
|
|
- Loading from HF requires `trust_remote_code=True`.
|