feat: use auto gen seed when using LLM cache #441
Conversation
rdagent/core/utils.py
Outdated
| results = [ | ||
| pool.apply_async(_subprocess_wrapper(f, APIBackend.cache_seed_gen.get_next_seed()), args=args) | ||
| pool.apply_async(_subprocess_wrapper, args=(f, random.randint(0, 10000), args)) # noqa: S311 |
There was a problem hiding this comment.
Please use get_next_seed;
We may use other random seed mechanism if we want to solve the confliction with random.
There was a problem hiding this comment.
Why should we use # noqa: S311?
There was a problem hiding this comment.
S311, i don't think we need this encryption requirements
There was a problem hiding this comment.
Please use get_next_seed; We may use other random seed mechanism if we want to solve the confliction with random.
APIBackend cannot be called here because a circular import occurs
There was a problem hiding this comment.
Import it in the function to avoid circular import
rdagent/core/utils.py
Outdated
| return random.randint(0, 10000) # noqa: S311 | ||
|
|
||
|
|
||
| cache_seed_gen = CacheSeedGen() |
There was a problem hiding this comment.
Please use upper case and add more specific prefix.
For example LLM_CACHE_SEED_GEN
Description
By two configs in RD_AGENT_SETTINGS, you can control the behavior of returning answers to the same question when LLM caching is enabled.
- use_auto_chat_cache_seed_gen: bool = False
- init_chat_cache_seed: int = 42
Behavior
Under setting: use_chat_cache=True, dump_chat_cache=True;
Given one question: Q, ask LLM 2 times;
Responses list is: (R1 ≠ R2 ≠ R3 ≠ R4 ≠ R5)
Motivation and Context
How Has This Been Tested?
pytest test/oai/test_completion.pyunder directory ofRD-Agent.Screenshots of Test Results (if appropriate):
Types of changes
📚 Documentation preview 📚: https://RDAgent--441.org.readthedocs.build/en/441/