_target_: src.models.TransformerConfig
max_blocks: 10  # this is the rollout length when training policy
tokens_per_block: 17 # how much info we can encode
dropout: 0.1
rank: 32 # lora rank
model_name: "PY007/TinyLlama-1.1B-intermediate-step-715k-1.5T"
vocab_size: 32000 # change to llm vocab dim
embed_dim: 2048  # change this to whatever the embedding dimension is in your pretrained llm 2048 for llama. 2560 for stablelm