Improve configuration with models details.

This commit is contained in:
ben
2025-01-16 21:07:47 +01:00
parent e2561fe3e0
commit 38ab78f649

View File

@@ -4,8 +4,22 @@ rag_embedding_model: ollama:nomic-embed-text
rag_chunk_size: 8192
rag_chunk_overlap: 409
model: ollama:qwen2.5
temperature: 0
clients:
- type: openai-compatible
name: ollama
api_base: http://localhost:11434/v1
api_key: __LLM_API_KEY__
models:
- name: qwen2.5
supports_function_calling: true
max_input_tokens: 128000
- name: qwen2.5-coder:32b
supports_function_calling: true
max_input_tokens: 128000
- name: nomic-embed-text
type: embedding
max_input_tokens: 200000
max_tokens_per_chunk: 2000
default_chunk_size: 8192
max_batch_size: 100