gslm-scaling-85m-65p3b

Model-only FP32 weights for a final-budget scaling-curve endpoint from Scaling Properties of Speech Language Models (EMNLP 2024). This checkpoint has 85,339,392 parameters and saw exactly 65,324,480,000 tokens at 262,400 tokens per update.

import torch
from slm import TransformerLM

model = TransformerLM.from_pretrained("tiagoCuervo/gslm-scaling-85m-65p3b", device="cuda")
prompt = torch.tensor([[12, 91, 204]], device="cuda")
units = model.generate(prompt, max_new_tokens=200, temperature=0.8, top_k=50)

Evaluation scores are 59.08% sBLIMP phenomenon macro accuracy, 56.54% sBLIMP voice-pair accuracy, 55.53% sStoryCloze, and 74.67% tStoryCloze.

The model predicts consecutive-run-collapsed 25 Hz layer-11 mHuBERT K-means units (K=500; EOS=500; PAD=501). A vocoder is not bundled.

Speech-tokenizer and vocoder artifacts are credited in the SLM third-party notices. See provenance.json for exact hashes and evaluation metadata.