Spaces:
Sleeping
Sleeping
Commit History
-
Add debug logging to diagnose GEPA candidate generation issue
- Added INFO level logging to make_reflective_dataset() to confirm if GEPA is calling it - Added logging when evaluate() is called with capture_traces=True (training minibatch) - Added GEPA version check - Added logging to show GEPA's captured output - This will help identify why GEPA isn't generating new candidates5b2a612 -
Fix: GEPA not generating candidates - training minibatch was incorrectly cached
ROOT CAUSE: - When GEPA calls adapter.evaluate(capture_traces=True), it needs trajectories for reflective mutation to generate new candidates - The adapter was returning cached results with trajectories=None for training minibatches because len(batch)=4 >= valset_size=3 was True - GEPA checks 'if not trajectories: skip', so no new candidates were generated FIX: 1. When capture_traces=True, ALWAYS treat as training minibatch (dfeedback) - Never use cache when traces are needed 2. Changed valset detection from '>=' to '==' for exact match 3. Added 'and not capture_traces' check before using cacheb4d947d -
Fix: Always pass reflection_lm_callable to GEPA - was broken when LLEGO disabled
The bug: When use_llego_operators=False, reflection_lm was set to None, which meant GEPA could not generate any new candidate prompts. Fix: Always pass the reflection_lm_callable to GEPA regardless of LLEGO mode. This allows GEPA reflection to work even without LLEGO genetic operators.6c19fd6 -
Fix duplicate seed prompts in Pareto front - add deduplication checks
74d4bea -
Remove hardcoded debug.log file writes - fix file not found errors
c0b1bdf -
Deploy Universal Prompt Optimizer to HF Spaces (clean)
cacd4d0