graders logic redefine

Merge pull request #4 from PushkarDesai-06/UIFix

Enhance Gradio UI integration with error handling and environment management

Align GitHub and Hugging Face main histories

Sync main changes to HF without binary history

Merge pull request #3 from PushkarDesai-06/rewardFix

Add Gradio UI integration and reward normalization logic

Merge branch 'main' of https://github.com/bansalrujul07/MedicalTriage

Pipeline changes

Merge branch 'main' of https://github.com/bansalrujul07/MedicalTriage

Reward logic changes

Merge branch 'main' of https://github.com/bansalrujul07/MedicalTriage

Inference changes

Merge branch 'main' of https://github.com/bansalrujul07/MedicalTriage

Test graders logic checks

Trigger fresh HF Space rebuild

Changes applied

Accept HF_TOKEN for submission inference

Restore canonical grader entrypoints

Fix Spaces build and grader wrappers

Remove redundant files and folders

Trigger fresh Space rebuild

Restore task graders blocks in manifest

Remove duplicate graders and canonicalize task entries

Clamp grader scores with epsilon-safe boundaries

regression tests add

Validation script changes applied

Learning Model changes applied

Implement full robustness hardening for grader pipeline

Harden graders with retries and strict defaults

Revert "Default graders to one episode and keep strict score fallback"

Default graders to one episode and keep strict score fallback

Harden all task graders for strict score range and robust imports

Fix grader scores to be strictly within (0,1)

Enforce OpenAI-only LLM config and harden inference proxy calls

LLMAgent: fail loudly in proxy mode instead of silent fallback

Harden grader: override rulebased when proxy env is injected

Use injected API_KEY only and force one proxy LLM call

Force LLMAgent when validator proxy env vars are present

Critical fix: grader now uses LLMAgent with validator-injected API credentials instead of RuleBasedAgent

Fix API key resolution: prioritize injected API_KEY, fallback to OPENAI_API_KEY and HF_TOKEN

Require injected API_KEY for proxy calls

Use injected API_KEY for LLM proxy calls

Fix inference.py: load .env file and provide sensible defaults for LOCAL_IMAGE_NAME

Ignore local training binaries

Make inference logging fully submission-safe

Align inference.py with strict submission contract

Implement OpenEnv compliance and OpenAI baseline requirements

Make grader fail-fast and prioritize submitted RL checkpoints

Finalize strict grader discovery and stabilize RL checkpoint loading