fix: cast lm_head and all params to bfloat16 to resolve dtype mismatch

fix: run 3 tasks with clamped scores in (0,1) for validator

fix: restore sys.path, accept API_KEY+HF_TOKEN, fix default port to 7860

fix: single-episode ultra-safe inference for hackathon validator

fix: remove score from END line, clean up log_step format

fix: use HF_TOKEN only as required by hackathon spec

fix: use API_KEY from LiteLLM proxy, correct defaults for hackathon validator

fix: prioritize API_KEY over HF_TOKEN for LiteLLM proxy

cleanup comments and readme

fix: grader variance + match sample inference.py format

fix: add root HTTP route to fix HuggingFace 404 display

fix: align inference.py initialization strictly with hackathon template

fix: correct inference output format

chore: added hf space openenv tag

chore: remove AI comments, separate tasks by difficulty, and sync README

Restored HF config block

Removed HuggingFace YAML metadata from README

Stylistic markdown de-aiing

Swapped sterile README for hackathon swagger

Perfect score formatting

Fix Dockerfile port for HF Spaces

Fix Dockerfile pathing

Added HF Space config

Final Hackathon Submission