-
Folder 100_heer: 26 images found
Folder 100_heer: 2600 steps
Total steps: 2600
Train batch size: 1
Gradient accumulation steps: 1.0
Epoch: 1
Regulatization factor: 1
max_train_steps (2600 / 1 /…
-
I have a process that writes out data in very small chunks, using `xr.Dataset.to_zarr(region=...)`. I then want to assemble the chunks into more reasonably sized chunks.
An order of magnitude on th…
-
For Llama2 prefill of prompt length 2k, SPDA is currently implemented with 8 passes of partial interleaving, matmul -> softmax -> matmul. This module takes 2.5ms, and constitutes 12% of prefill runtim…
-
will allow clipboard interop in wsl mode
`:help g:clipboard`
-
I tested 4 different runs with exactly the same data and the same process:
1- 2260 workers X 1 core : performance 108 blocks/s
2- 1000 workers X 1 core: performance 127 blocks/s
3- 500 workers x …
-
I am seeing some warnings in blockwise_test.cc. Is this something that could be re-factored ? Otherwise I'll shut-up the warning.
```
[53/140] /usr/bin/c++ -DHWY_SHARED_DEFINE -DTOOLCHAIN_MISS_AS…
-
i notice the battery firmware is not encrypted. i wonder what could be done to the battery controller so aftermarket batteries could be connected
-
https://github.com/blockwise-direct-search/bds/blob/edfd73cb678fb1c8336eb745ea8c72f37f91531d/setup.m#L34-L37
-
I'm liking the shuffle groupby-aggregations stuff, especially when we have many grouping columns and large data (I suspect that this would be a sensible default). Unfortunately, it's not working quit…
-
Here,
https://github.com/blockwise-direct-search/bds/blob/main/tests/testbds_parameters
both `"` and `'` are used. Why? They should be unified. This is ONLY AN EXAMPLE. Check all files.