ProBas_RAG_Assistant / .env.example
Mohamed284's picture
Deploy ProBas RAG Assistant with enriched prebuilt index
0ca97fd
Raw
History Blame Contribute Delete
1.3 kB
# ProBas RAG Assistant configuration
# Copy this file to .env and fill in the values for your deployment.
OPENAI_API_KEY=your_openai_compatible_api_key_here
OPENAI_BASE_URL=https://chat-ai.academiccloud.de/v1
PROBAS_EMBEDDING_MODEL=qwen3-embedding-4b
PROBAS_MAX_RECORDS=0
PORT=7860
# Index build tuning
PROBAS_EMBED_BATCH_SIZE=12 # texts per embedding request (smaller = fewer timeouts)
PROBAS_EMBED_CONCURRENCY=4 # parallel embedding requests (main speed lever)
PROBAS_EMBED_TIMEOUT_SECONDS=180 # per-request timeout for the embedding model
PROBAS_EMBED_MAX_RETRIES=1 # retries before a batch is split in half
PROBAS_CHECKPOINT_EVERY=5 # save a resume checkpoint every N waves
# Retrieval and answer-quality tuning
PROBAS_BM25_WEIGHT=0.30 # lexical weight in the hybrid score
PROBAS_VECTOR_WEIGHT=0.70 # dense embedding weight (carries cross-lingual queries)
PROBAS_MIN_RELEVANCE=0.42 # below this top cosine, a query is answered conversationally
PROBAS_MAX_CONTEXT_CHARS=5000 # per-record excerpt size fed to the model
PROBAS_EVIDENCE_SNIPPET_CHARS=320 # per-record snippet shown in the UI evidence panel (compact)
# PROBAS_DISABLE_AUTOSTART=1 # skip background index build on import (useful for tests)