Spaces:
Sleeping
Sleeping
Commit History
-
fix: escape JSON braces in f-string and use env vars for model in debug script
garg-tejas77588b1 -
feat: hackathon compliance - rename to inference.py, expand openenv.yaml
garg-tejas- Rename baseline_inference.py to inference.py (hackathon requirement) - Update to use required env vars: API_BASE_URL, MODEL_NAME, HF_TOKEN - Expand openenv.yaml with full spec (action/observation spaces, rewards, tasks) - Update debug_model_responses.py to import from inference.py - Update Dockerfile for HF Spaces deployment - Update README with project overviewc6369ba
-
refactor: eliminate technical debt across codebase
garg-tejasMajor improvements: - Add logging to all silent exception handlers with debug-level messages - Consolidate timestamp parsing into parse_timestamp() utility in config.py - Remove duplicate difficulty config - use get_difficulty_config() everywhere - Standardize observation conversion with _to_observation() helper - Move magic numbers to config.py: - LLM_TEMPERATURE, LLM_MAX_TOKENS, HF_ROUTER_URL - OUTPUT_PREVIEW_SHORT, OUTPUT_PREVIEW_LONG - MAX_COMMAND_HISTORY - Remove unused imports (hashlib, timedelta, field, Tuple, etc.) - Simplify debug_model_responses.py by reusing baseline_inference helpers Files changed: - config.py: Added timestamp parsing, LLM config, command history config - grader.py: Use centralized timestamp parsing and difficulty config - server/log_anomaly_environment.py: Use centralized config throughout - baseline_inference.py: Simplify observation handling, use config constants - debug_model_responses.py: Reuse _to_observation from baseline_inference - log_utils…f69e6aa
-
fix: simplify fallback detection - track at parse time
garg-tejas- Add _last_parse_used_fallback flag in ReactAgent.parse_action() - Debug script now just reads this flag instead of guessing - Removed all the complex heuristics that were wrongb1313d0 -
fix: add difficulty-aware anomaly type hints to system prompt
garg-tejas- Easy: only error_spike is valid (model was guessing wrong types) - Medium: error_spike, memory_leak, latency_degradation, service_dropout - Hard: only cascade_failure is valid Also fixed debug script to use new code-block detection instead of checking for old 'Command:' prefix format.e2bb581
-
fix: remove api_key parameter from DebugReactAgent constructor
garg-tejasReactAgent reads API key from environment variables, not constructor args135990f -
refactor: remove dead code, fix type issues, add missing type annotations
garg-tejas- Remove unused imports: Path, BaseObservation (server/log_anomaly_environment.py), json (grader.py) - Remove empty __init__ in InvestigationGrader - Consolidate duplicate pydantic BaseModel import in models.py - Add missing return type annotations across all modules - Fix base_url parameter type in run_baseline_inference (Optional[str]) - Add type annotations to debug_model_responses.py methodsaddd314 -
fix: correct debug script to use LogAnomalyEnvironment directly with proper type conversions
garg-tejas6c8400e -
feat: add debug script to capture raw model responses for parsing analysis
garg-tejas3cb2bed