CoolFace
Modelpublic

rrison/cs336_TinyStoriesV2

sourceHugging Facemitupdated 7mo agoView on Hugging Face
0likes
Model Card

Trained a 17M Language Model for CS336 Assignment 1.

  • —beta0:0.9
  • —beta1:0.95
  • —context_length:256
  • —d_ff:1,344
  • —d_model:512
  • —eval_batches:200
  • —log_interval:100
  • —lr:0.001
  • —max_steps:30,000
  • —num_heads:16
  • —num_layers:4
  • —precision:"bf16"
  • —theta:10,000
  • —vocab_size:10,000

train/val loss is around 1.5. https://wandb.ai/rrisoncai-na/cs336-assignment1/runs/vvr52w7e?nw=nwuserrrisoncai