provisioning-platform/templates/rag.ncl.j2

55 lines
1.5 KiB
Django/Jinja

# RAG Service Configuration
# Generated by TypeDialog - Edit via configure-rag.sh
# Rust struct: platform/crates/rag/src/config.rs::RagConfig
{
enabled = {{ enabled }},
embeddings = {
provider = "{{ embeddings_provider }}",
model = "{{ embeddings_model }}",
dimension = {{ embeddings_dimension }},
{% if embeddings_openai_api_key %}
openai_api_key = "{{ embeddings_openai_api_key }}",
{% endif %}
batch_size = {{ embeddings_batch_size }},
},
vector_db = {
db_type = "{{ vector_db_db_type }}",
url = "{{ vector_db_url }}",
namespace = "{{ vector_db_namespace }}",
{% if vector_db_username %}
username = "{{ vector_db_username }}",
{% endif %}
{% if vector_db_password %}
password = "{{ vector_db_password }}",
{% endif %}
},
llm = {
provider = "{{ llm_provider }}",
model = "{{ llm_model }}",
{% if llm_api_key %}
api_key = "{{ llm_api_key }}",
{% endif %}
{% if llm_api_endpoint %}
api_endpoint = "{{ llm_api_endpoint }}",
{% endif %}
temperature = {{ llm_temperature }},
max_tokens = {{ llm_max_tokens }},
},
retrieval = {
max_results = {{ retrieval_max_results }},
min_similarity_score = {{ retrieval_min_similarity_score }},
enable_reranking = {{ retrieval_enable_reranking }},
},
ingestion = {
batch_size = {{ ingestion_batch_size }},
chunk_size = {{ ingestion_chunk_size }},
chunk_overlap = {{ ingestion_chunk_overlap }},
enable_async = {{ ingestion_enable_async }},
},
}