TagaLLM
Collection
Tagalog and Taglish research models and dataset • 5 items • Updated
TagaLLM 15.52m-512
This is a scratch-trained TagaLLM research checkpoint for Tagalog and Taglish dialogue. It uses a decoder-only Transformer with 6 layers, width 432, 8 attention heads, a 4096-token vocabulary and a 512-token context. The model includes a causal copy head.
The weights contain 69,156,460 bytes. See the manifest for the archived checkpoint hash and the exported files' SHA-256 hashes. The release includes model parameters, tokenizer, configuration and a local runner. The original corpus is a separate artifact under review. This repository does not grant a license to that corpus.
Install Python 3 and run these commands from this repository folder.
python -m pip install torch safetensors tokenizers
python generate.py "Kumusta ka?"
This checkpoint is intended for research and may produce repetitive or inaccurate text. See the research evidence for the run and evaluation details.
During TagaLLM development, a native Tagalog speaker helped review selected model replies for grammar and natural phrasing, with AI assistance.
Project links
- Downloads last month
- 30
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support