junrushao1994 commented on code in PR #10986:
URL: https://github.com/apache/tvm/pull/10986#discussion_r849893934
##########
python/tvm/meta_schedule/search_strategy/evolutionary_search.py:
##########
@@ -64,13 +64,13 @@ def __init__(
*,
num_trials_per_iter: int,
max_trials_per_task: int,
- population_size: int,
- init_measured_ratio: float,
- init_min_unmeasured: int,
- genetic_num_iters: int,
- genetic_mutate_prob: float,
- genetic_max_fail_count: int,
- eps_greedy: float,
+ population_size: int = 2048,
+ init_measured_ratio: float = 0.2,
+ init_min_unmeasured: int = 50,
+ genetic_num_iters: int = 4,
+ genetic_mutate_prob: float = 0.85,
+ genetic_max_fail_count: int = 10,
+ eps_greedy: float = 0.05,
Review Comment:
Thanks for sharing! If I read it correctly, while indeed it's less frequent,
there is definitely usecase to tweak parameters. To allow this to happen, we
have an extra parameter called `TuneConfig.search_strategy_config`. For example,
```python
tune_relay(
config=TuneConfig(
strategy="evolutionary",
num_trials_per_iter=64,
max_trials_per_task=20000,
max_trials_global=20000,
search_strategy_config = {
"genetic_num_iters": 10,
},
),
)
```
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
