-
I'm going to post a number of observations about training here that led me to a hypothesis about network training that affects Leela Zero, Leela Chess and Minigo.
For Leela Zero, during regular rei…
-
Batch size is 2048. Getting the training data from the q I see
x_train, y_train = train_q.get()
print('X.. ' + str(x_train.shape) + ' Y.. ' + str(y_train.shape))
X.. (2048, 2980) Y.…
-
Endgame in Go often decomposes into independent regions.
Each region can have assigned mean and temperature as defined by CGT (Combinatorial Game Theory).
Playing a region with maximal temperature …
-
### Issue
When `communicator` is set to `none`, the VM won't boot even if boot commands are specified in the configuration file, as seen below:
### Input
```json
...
"communicator": "none",
…
-
Hi all,
there's 500k games now without a network promoting. This means the training window is "full". I increased it to 750k games, and did a last learning rate lowering (0.00001 @ bs=96). If this …
-
### Proposal:
Players hold tech points, rather than a research server. They increase tech points by gaining experience, through breaking down items or creating items in relevant tech-trees, perform…
-
exploration of simulated realities, taxes, learning mechanisms, and proof generation has been a fantastic example of meta-narrative creation and market-driven meaning-making through meta-reflection.
…
-
Dear Yang Zhong,
I implemented the random shift constraint similar to what you did in the MoS2 Demo (random shift within a range of 2 angstroms in the ab plane and 0.5 angstroms along the c-axis). …
-
Hi, when using tiled diffusion with more than one controlnet in series, I was running into errors. The workflow is here:
https://github.com/aravindhv10/ComfyUI-TiledDiffusion/blob/fix_controlnet/coun…
-
There was an active discussion in #14 about the drawbacks of passing initialized objects, such as callbacks, to the `NeuralNet` constructor. Let's have a focused discussion about this issue here.
F…