StepLaw/StepLaw-N_1.0B-D_1.0B-LR3.453e-04-BS262144

Wandb Model Name: step2v2_0618_h2048_ffnh8192_numh16_numl16_lr3.453e-04_bs128_ti7629_mlr1e-5

This model is part of the StepLaw-N_1.0B-D_1.0B collection.

Model Specifications

Architecture

• Hidden size (H): 2048
• Feed-forward network size (FFN): 8192
• Attention heads: 16
• Layers: 16
• Parameter count: 1.1B

Training Parameters

• Learning rate (lr): 3.453e-04
• Batch size (bs): 262144
• Training iterations: 7629
• Training tokens (D): 2.0B

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 3.453e-04 and batch size 262144 for 7629 iterations, using a total of 2.0B training tokens.

Usage Example

from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "StepLaw/StepLaw-N_1.0B-D_1.0B-LR3.453e-04-BS262144"
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))