base.json

{
  "inference_settings": [
    {
      "model_settings": {
        "generator_settings": {
          "model_path": "tests/fixtures/models/llama2_tiny",
          "model_type": "causal",
          "transformers_settings": {},
          "resize_token_embeddings": true,
          "embeddings_initialization_strategy": {
            "<RS>": "<s>",
            "</RS>": "</s>",
            "<bot>": "bot",
            "<user>": "user",
            "<system>": "system"
          },
          "adapter_path": "tests/fixtures/models/llama2_tiny_fine_tuned_rag"
        },
        "question_encoder_settings": {
          "model_path": "tests/fixtures/models/tiny-bert-embedder",
          "model_type": "encoder",
          "transformers_settings": {},
          "embeddings_initialization_strategy": {}
        },
        "index_settings": {
          "index_path": "tests/fixtures/datasets/rag/fake_wiki.faiss",
          "passages_path": "tests/fixtures/datasets/rag/fake_wiki"
        },
        "retrieval_settings": {
          "n_docs": 1,
          "max_doc_length": 100,
          "query_encoder_max_length": 512
        }
      },
      "tokenizer_settings": {
        "use_fast": false,
        "tokenizer_path": "tests/fixtures/models/llama2_tiny_fine_tuned_rag"
      },
      "generation_settings": [
        {
          "transformers_settings": {
            "num_beams": 1,
            "max_new_tokens": 10,
            "repetition_penalty": 1.2,
            "stop_strings": "</RS>",
            "do_sample": false
          },
          "custom_settings": {
            "skip_special_tokens": false,
            "remove_prompt": false
          }
        }
      ]
    }
  ],
  "dataset_settings": {
    "sources": [
      {
        "name": "test",
        "records_path": "tests/fixtures/datasets/chat/train_chat_rag.jsonl",
        "sample_rate": 1.0
      }
    ],
    "prompt_template": {
      "role_tag_mapping": {
        "bot": "<bot>",
        "user": "<user>",
        "system": "<system>"
      },
      "prefix_template": "<RS>{role}",
      "suffix_template": "</RS>"
    },
    "dataset_type": "chat",
    "max_tokens_count": 2000,
    "only_answer_loss": true
  },
  "save_path": "test_inference_rag_output"
}