diff --git a/aif_gen/generate/engine.py b/aif_gen/generate/engine.py index 5187a701..beab95c2 100644 --- a/aif_gen/generate/engine.py +++ b/aif_gen/generate/engine.py @@ -271,9 +271,11 @@ async def _generate_sample( meta_prompt_nonce = f'{prompt_idx}' async with async_semaphore: + output = None if cache is not None: output = await cache.get(meta_prompt, nonce=meta_prompt_nonce) - else: + + if output is None: # Cache miss or no cache - make API call response = await client.chat.completions.create( model=model_name, messages=[{'role': 'user', 'content': meta_prompt}], @@ -306,9 +308,11 @@ async def _generate_sample( task_prompt = task_prompt1 + task_prompt2 async with async_semaphore: + output = None if cache is not None: output = await cache.get(task_prompt1 + task_prompt2) - else: + + if output is None: # Cache miss or no cache - make API calls futures = [] for response_prompt in [task_prompt1, task_prompt2]: coro = client.chat.completions.create( @@ -346,9 +350,11 @@ async def _generate_sample( else: task_prompt = response_mapper.generate_prompt(task, prompt) async with async_semaphore: + output = None if cache is not None: output = await cache.get(task_prompt) - else: + + if output is None: # Cache miss or no cache - make API call response = await client.chat.completions.create( model=model_name, messages=[{'role': 'user', 'content': task_prompt}], diff --git a/aif_gen/generate/mappers/response_mapper.py b/aif_gen/generate/mappers/response_mapper.py index e67cf37e..c5923520 100644 --- a/aif_gen/generate/mappers/response_mapper.py +++ b/aif_gen/generate/mappers/response_mapper.py @@ -15,8 +15,8 @@ class ResponseMapper(ResponseMapperBase): """ NUM_PREFERENCE_AXES_SAMPLES: int = 3 - PREFERENCE_INCLUSION_PROB_POS: float = 0.5 - PREFERENCE_INCLUSION_PROB_NEG: float = 0.5 + PREFERENCE_INCLUSION_PROB_POS: float = 0.8 + PREFERENCE_INCLUSION_PROB_NEG: float = 0.8 def __init__(self, suffix_context: Optional[str] = None) -> None: self._suffix_context = suffix_context