StepLaw-N_429M-D_22.0B-LR3.91E-03-BS196608

Wandb Model Name: step2v2_0618_h1280_ffnh9472_numh10_numl10_lr3.91E-03_bs96_ti115712_mlr1.00E-05

This model is part of the StepLaw-N_429M-D_22.0B collection.

Model Specifications

Architecture

• Hidden size (H): 1280
• Feed-forward network size (FFN): 9472
• Attention heads: 10
• Layers: 10
• Parameter count: 429M

Training Parameters

• Learning rate (lr): 3.91E-03
• Batch size (bs): 196608
• Training iterations: 115712
• Training tokens (D): 22.7B

Model Description

StepLaw models are trained with various hyperparameter settings to enable research on scaling laws and hyperparameter optimization. This specific model was trained with learning rate 3.91E-03 and batch size 196608 for 115712 iterations, using a total of 22.7B training tokens.

Usage Example

from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "StepLaw/StepLaw-N_429M-D_22.0B-LR3.91E-03-BS196608"
tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True, use_fast=False)
model = AutoModelForCausalLM.from_pretrained(model_name, trust_remote_code=True)

# Generate text
inputs = tokenizer("A long time ago in a galaxy far, far away", return_tensors="pt")
outputs = model.generate(**inputs, max_length=100)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))