Predict Future Vehicle Trajectory with LLAMA Model:

To attain the most straightforward representation for the input, we represent object trajectories and maps using sequences of line segments and type.

Model

The model is train on argoverse v1 dataset.
model args:

# model
dim = 96 #5 #in vector space
n_layers = 6
n_heads = 12 #5
n_kv_heads = 12 #5
multiple_of = 32
dropout = 0.0

# data
batch_size = 64  # if gradient_accumulation_steps > 1, this is the micro-batch size
input_length = 244
input_width = 5  # input 3->30 # the Llama 2 tokenizer has 32K tokens #output size

How to use:

Source code to run this model can be found at argo_llama