alphadev
alphadev copied to clipboard
google-deepmind
If my knowledge is correct, AlphaDev can call opcodes and provide parameters(like mov rax… instead of just mov). How would I implement this?
Hi, is there any executable implementation to feel how strong AlphaDev is? We are exploring reinforcement learning for superoptimization. But, we are new to deep learning. Thus, I would appreciate...
``` while node.expanded(): action, node = _select_child(config, node, min_max_stats) sim_env.step(action) history.add_action(action) search_path.append(node) # Inside the search tree we use the environment to obtain the next # observation and reward given...
Hi folks, I think you have a bug on line line. the function body is empty, it raises intendation error ``` def legal_actions(self) -> Sequence[Action]: ^ IndentationError: expected an indented...
We're trying to reproduce your work, I was wondering how you encoded register and memory. Can you tell me what each of dimension stands for? We assume B for batch...
Hi, thanks for your sharing. When I want to use your algorithm, an error occurs. Is there any limit on this test to be executed?
I was curious how well the proposed sorting functions for three, four or five integers would perform on my hardware, so I wrote a quick benchmark. It seems like there...
Doesn't really matter in pseudocode, I guess, but value = 0 on line 829 should fix this
In alphadev.py:1031, isn't this applying the last action twice? The first time being on line 1025. Also, on line 1034 why don't we use legal_actions but instead expand on all...