saheb/content-moderation-env

Upload folder using huggingface_hub

Update log_end function: Include score in end log output for better tracking

Update grader score precision: Change rounding from 4 to 2 decimal places

Enhance ContentModerationEnvironment: Initialize grader score at episode start and compute it consistently

Update benchmark display: Add clarification on score accuracy and adjust score formatting

Enhance Content Moderation Environment: Add package exports, improve import handling, and update moderation actions

Reduce Uvicorn timeout from 300 to 65 seconds for improved performance

Refactor Dockerfile: Replace uv with pip for dependency installation and improve caching

Enhance Content Moderation Environment: Update Dockerfile, add client, improve README, and refine environment handling

updating benchmarks

benchmarks and readme

tidy

updating benchmarks

prevent reward hacking

view_thread and view_post each reward only once per post per episode.

thread reading enabled

polish

clamping scores to be strictly between 0 and 1

hf port 7860 not necessary for docker; adding app_port in hf yaml readme

HF_TOKEN → api_key

HF_TOKEN first, API_KEY as fallback

remove HF_TOKEN as fallback

opt in load_dotenv

Another attempt at fixing No API calls were made through our LLM proxy

fix(env): prioritize API_KEY over HF_TOKEN for API key lookup

updating benchmarks

fixing env variables and linking benchmarks.html

reset and step fixes

interactive demo UI and some refactoring

feat: add very hard difficulty tier and interactive demo UI

refactoring, polish, and adding very hard data

refactoring, polish, and adding very hard data

refactor: improve inference robustness, update documentation, and enhance Pydantic schema definitions

docs: add Gemini 3.1 Flash Lite performance metrics to README table

fix: add InternalServerError handling to inference fallback and update benchmarks with gemma-4-31b-it results

feat: update inference to use strict JSON schema and add reward metrics to README results table

refactor: transition moderation environment to active-post focus with failure tracking and update model benchmarks

update content moderation datasets and increase max_steps in environment configuration

feat: enhance inference engine with structured logging, conversation history, multi-stage JSON schema fallbacks, and global timeout management.

feat: add .env support, improve rate-limit handling, and update configuration for flexible LLM provider integration

Add uv.lock for reproducible builds

Update Dockerfile and pyproject.toml to use app instance directly; enhance app initialization in server.py with readiness endpoint

Refactor app initialization to use factory pattern and update baseline script references

Add auto-sync workflow to HF Spaces

Refactor app creation and improve comments for clarity

Fix app instantiation for HF Spaces compatibility

Fix app factory and add health checks

chore: update inference client base URL to localhost for local development

chore: add 1.5s step delay to avoid Groq RateLimitError cascade

Fix HF Load Balancer ping (Add GET / health check endpoint)