import json from app.prompts import SYSTEM_INSTRUCTION, build_prompt, training_input from app.schemas import ProgressionRequest def make_request(valid_request) -> ProgressionRequest: return ProgressionRequest(**valid_request) def test_prompt_matches_sft_shape(valid_request): req = make_request(valid_request) prompt = build_prompt(req) assert prompt.count("\n\nINPUT:\n") == 1 assert prompt.count("\n\nOUTPUT:\n") == 1 assert prompt.startswith(SYSTEM_INSTRUCTION.strip() + "\n\nINPUT:\n") assert prompt.endswith("\n\nOUTPUT:\n") assert '"previous_activity":{"title":"Practice the skill"}' in prompt assert '"knowledge_submission":"I completed the task."' in prompt assert "previous_activity_title" not in prompt def test_prompt_contains_compact_json(valid_request): req = make_request(valid_request) prompt = build_prompt(req) body = prompt.split("\n\nINPUT:\n")[1].split("\n\nOUTPUT:\n")[0] assert "\n" not in body assert body == json.dumps( training_input(req), ensure_ascii=False, separators=(",", ":"), ) assert json.loads(body) == training_input(req) def test_prompt_does_not_escape_non_ascii(valid_request): valid_request["knowledge_submission"] = "J'ai terminé la tâche." prompt = build_prompt(make_request(valid_request)) assert "J'ai terminé la tâche." in prompt def test_training_input_structure(valid_request): model_input = training_input(make_request(valid_request)) assert list(model_input) == [ "village_goal", "previous_activity", "knowledge_submission", "wisdom", ] assert model_input["previous_activity"] == { "title": "Practice the skill", } def test_wisdom_field_names_preserved(valid_request): model_input = training_input(make_request(valid_request)) assert list(model_input["wisdom"]) == [ "content", "chapterTitle", "BookTypeName", "BookType", ] assert model_input["wisdom"] == valid_request["wisdom"]