Spaces:
Sleeping
Few tasks: @inference.py README CLAUDE All have mention of -e MEDCHAIN_TASK=hospital_network_crisis But we want to remove this as it is not required, remove it from @server as well, we don't need to support it. Like remove this as well env_vars={"MEDCHAIN_TASK": task_name}
PLease define MAX_STEPS_PER_TASK as per task not globally like 150 (give sufficient room though as error may occur in between)
Add expontential backoff if frequent 429 errors to slow down request however be careful not to increase the time too much as all tasks need to be finish under 20 minutes. If there are frequent 429 errors then downgrade to smaller model (use SMALL_MODEL and hardcode it to openai/gpt-oss-20b:groq). log warning when downgrading and when getting 429 error.
Put log.debug for duration for each task take to complete and overall duration of script
Look and tell what librabries/dependencies are currently not pinned that may cause issue in reproducibility. Docker base image (pin it to ghcr.io/meta-pytorch/openenv-base@sha256:3e478c17bcdee6969218c03d0c1986bdb0db9abd4e763a9180b9014b3dde211a)
tool handling check with other models