hrm-rec-strict-small

HRM-rec (hierarchical weight-shared recurrent LM) for the BabyLM 2026 strict-small track. Final checkpoint on main (chck_100M); intermediate checkpoints (if present) on chck_<N>M branches. Custom model — load with trust_remote_code=True.

from transformers import AutoModelForCausalLM, AutoTokenizer
m = AutoModelForCausalLM.from_pretrained("PatrickHaller/hrm-rec-strict-small", trust_remote_code=True)
t = AutoTokenizer.from_pretrained("PatrickHaller/hrm-rec-strict-small")
Downloads last month
899
Safetensors
Model size
0.1B params
Tensor type
F32
·
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support