multiprocessing: False
root: ${root:}
use_cache: False
crop_larger_margin: True # Increase size of crop by 10px to include more context
useful_lists_path: ${root:}/src/useful_lists
load_models:
glip: False
midas: True
blip: False
xvlm: True
codex: True
codellama: False
gpt4: True
gpt4_guess: True
gpt4_qa: True
gpt4_general: True
dino: True
sam: True
gpt4vlm: True
gpt4codex: True
o3codex: False
bert: True
use_local_models: True # If True, will use local models instead of downloading them;
# Not available for all models, but will use local models if available.
llm:
n_votes: 1 # NOTE: Due to changes in OpenAI API, this now makes N individual requests instead of a single request with N messages.
qa_prompt: ${root:}/src/prompts/llm/qa_prompt.txt
guess_prompt: ${root:}/src/prompts/llm/guess_prompt.txt
temperature: 0.0
model: gpt-4o-mini
path: None
name: gpt4
api: responses # Options: completions, responses
codex:
model: gpt-4.1
temperature: 0.0
path: None
name: gpt4codex
prompt: ${root:}/src/prompts/codex/codex.prompt
task_prompt: ${root:}/src/prompts/codex/task.prompt
query_prompt: ${root:}/src/prompts/codex/query.prompt
api: responses # Options: completions, responses
object_segmentation:
model: facebook/sam-vit-base
path: ${pretrained_models_path:}/sam
name: sam
object_captioning:
model: gpt-4o-mini
path: None
half_precision: True
name: gpt4vlm
temperature: 0.0
object_detection:
model: IDEA-Research/grounding-dino-base
path: ${pretrained_models_path:}/dino
detection_threshold: 0.39
name: dino
object_recognition:
model: xvlm
path: ${pretrained_models_path:}/xvlm
detection_threshold: 0.45
name: xvlm
depth_estimation:
model: intel/dpt-hybrid-midas
path: ${pretrained_models_path:}/midas
name: midas
text_encoder:
model: bert-base-uncased
path: ${pretrained_models_path:}/bert
name: bert
