Spaces:
Sleeping
Sleeping
Commit History
-
fix: full Supabase integration audit and cleanup
RemanenetSpy- Add missing query_temporal() to SupabaseStore (fixes AttributeError on time-range queries via BYODB) - Fix SupabaseStore class structure - proper boundary between SupabaseStore and SupabaseVectorStore - Fix embed text drift: use VectorStore._build_embed_text() in add_events_batch (includes raw_text, matches main store quality) - Fix ON CONFLICT DO NOTHING -> DO UPDATE SET in event_vectors (re-ingestion now updates vectors correctly) - Move asyncio to top-level import in connector.py - Wire eject_all_pools() to FastAPI lifespan shutdown (fixes pool leak on server restart) - Fix duplicate step #6 comment in ingest.py -> #7 - Update README and test_integration.ps1 to use Session Pooler URL with cloud deployment warning - Delete 31 dead one-off scripts from project root (check_*.py, fix_*.py, patch_*.py etc.)754e261
-
feat(kaggle): dynamically set vLLM tensor_parallel_size to use all available GPUs (e.g. Kaggle's T4x2) for massive speedup
RemanenetSpy5bfcabd -
fix(kaggle): print progress on every case and incrementally save results to disk to prevent total data loss if Kaggle hits 12-hour timeout
RemanenetSpy20cebf8 -
fix(kaggle): replace .format() with .replace() to avoid KeyError on JSON braces inside the prompt
RemanenetSpydf608c2 -
fix(kaggle): cast all dataset fields to strings to prevent AttributeError on integer answers
RemanenetSpya122973 -
fix(kaggle): handle missing 'id' and 'category' keys in LongMemEval dataset cases gracefully to prevent KeyError
RemanenetSpy3cfa7ac -
fix(kaggle): forcefully remove torchcodec after pip installs to prevent it being re-installed as a dependency and crashing SentenceTransformers
RemanenetSpy11923ea -
fix(kaggle): forcefully remove broken torchcodec directory to prevent SentenceTransformers from crashing during initialization
RemanenetSpy62b9da3 -
fix(kaggle): unpin vllm to support Kaggle Python 3.12 update while keeping torchcodec uninstall
RemanenetSpy08c70ee
-
fix(kaggle): downgrade vllm to 0.5.1 and remove torchcodec to bypass CUDA missing library error
RemanenetSpye155204 -
chore(benchmark): switch Kaggle notebook to ungated Qwen2.5-7B-Instruct to avoid license blocks
RemanenetSpy3bf9a1a -
fix(kaggle): correct dataset path and filename casing for Kaggle Linux filesystem
RemanenetSpy2040e1d -
fix(benchmark): run ALL cases for both S and M splits, save results JSON + BENCHMARKS.md
RemanenetSpy91236f3 -
update kaggle notebook to run both S and M splits sequentially
RemanenetSpy95b8018 -
feat(benchmark): add kaggle notebook for real smriti algorithm using local Llama 3 8B via vLLM
RemanenetSpy80a537b
-
fix(kaggle): remove hardcoded dataSources metadata to fix import error
RemanenetSpy1ae0e85 -
fix(kaggle): update dataset path to match user's custom kaggle dataset output
RemanenetSpycdb2a83 -
feat: add Kaggle notebook for LongMemEval S+M benchmark evaluation
RemanenetSpyfd13692
-
docs: add LongMemEval evaluation and BENCHMARKS.md leaderboard strategy
RemanenetSpyaa606b6 -
chore: reorganize repo — move benchmark_adapter to benchmark/adapters, cerberos contract to docs, remove stale error.txt and duplicate precision-mem-bench.md from docs
RemanenetSpy1cca1b2 -
feat(benchmark): add isolated benchmark suite with harness, adapters, and results
RemanenetSpy38fb868
-
chore: add smriti benchmark runner and eval service
RemanenetSpy64f8bdd