Commits

Branch selector

 next

User selector

Ttl

Datepicker

All time

Commit History

Commits on Apr 2, 2019

• Choose move based on normal distribution LCB.

  Show description for fd23877
  

  Ttl

  authored and

  gcp

  committed
  fd23877

  Copy full SHA for fd23877

• Enable batching for self-play.

  Show description for 9beb22e
  

  Ttl

  authored and

  gcp

  committed
  9beb22e

  Copy full SHA for 9beb22e

• Autogtp: Tune for batchsize 1

  Show description for ac6c4ab
  

  Ttl

  authored and

  gcp

  committed
  ac6c4ab

  Copy full SHA for ac6c4ab

• Add O(sqrt(log(n))) scaling to tree search.

  Show description for e0d64c3
  

  Ttl

  authored and

  gcp

  committed
  e0d64c3

  Copy full SHA for e0d64c3

• Optimized out and out_in kernels.

  Show description for c881ab6
  

  Ttl

  authored and

  gcp

  committed
  c881ab6

  Copy full SHA for c881ab6

Commits on Oct 15, 2018

• Report tuner errors to stderr.

  Show description for 7f5073e
  

  Ttl

  authored and

  gcp

  committed
  7f5073e

  Copy full SHA for 7f5073e

Commits on Sep 14, 2018

• Fix vectorized Winograd transform.

  Show description for 0a0d134
  

  Ttl

  authored and

  gcp

  committed
  0a0d134

  Copy full SHA for 0a0d134

Commits on Sep 13, 2018

• Winograd filter transform and CPU in transform optimization.

  Show description for 7e13bf0
  

  Ttl

  authored and

  gcp

  committed
  7e13bf0

  Copy full SHA for 7e13bf0

• Make Winograd matrices global.

  Show description for dd95cab
  

  Ttl

  authored and

  gcp

  committed
  dd95cab

  Copy full SHA for dd95cab

Commits on Aug 16, 2018

• Script for quantizing weights.

  Show description for f85a685
  

  Ttl

  authored and

  gcp

  committed
  f85a685

  Copy full SHA for f85a685

Commits on Aug 14, 2018

• Fix OpenCL buffer sizes.

  Show description for 07c908e
  

  Ttl

  authored and

  gcp

  committed
  07c908e

  Copy full SHA for 07c908e

Commits on Aug 9, 2018

• OpenCL tuner fixes.

  Show description for d2ad525
  

  Ttl

  authored and

  gcp

  committed
  d2ad525

  Copy full SHA for d2ad525

Commits on Aug 6, 2018

• Use L2-norm in self check.

  Show description for 488de43
  

  Ttl

  authored and

  gcp

  committed
  488de43

  Copy full SHA for 488de43

Commits on Aug 3, 2018

• Half precision compute support.

  Show description for 5de99db
  

  Ttl

  authored and

  gcp

  committed
  5de99db

  Copy full SHA for 5de99db

Commits on Jul 28, 2018

• Remove separate half precision kernel.

  Show description for 6f8c873
  

  Ttl

  authored and

  gcp

  committed
  6f8c873

  Copy full SHA for 6f8c873

Commits on Jul 24, 2018

• Increase error budget in tuner.

  Show description for 8c65b6c
  

  Ttl

  authored and

  gcp

  committed
  8c65b6c

  Copy full SHA for 8c65b6c

• Use Winograd F(4x4, 3x3).

  Show description for 7cfbb72
  

  Ttl

  authored and

  gcp

  committed
  7cfbb72

  Copy full SHA for 7cfbb72

Commits on Jun 11, 2018

• Script for converting minigo weights.

  Show description for 54e130e
  

  Ttl

  authored and

  gcp

  committed
  54e130e

  Copy full SHA for 54e130e

Commits on Jun 6, 2018

• Non-pruning (simple) time management.

  Show description for d362ee8
  

  Ttl

  authored and

  gcp

  committed
  d362ee8

  Copy full SHA for d362ee8

Commits on Apr 11, 2018

• Fix dropped bias when widening.

  Show description for 5b2460c
  

  Ttl

  authored and

  gcp

  committed
  5b2460c

  Copy full SHA for 5b2460c

Commits on Mar 30, 2018

• net2net: option to increase the number of input planes.

  Show description for 872d4bb
  

  Ttl

  authored and

  gcp

  committed
  872d4bb

  Copy full SHA for 872d4bb

Commits on Mar 26, 2018

• Batch norm beta support.

  Show description for 5a2aca1
  

  Ttl

  authored and

  gcp

  committed
  5a2aca1

  Copy full SHA for 5a2aca1

Commits on Mar 23, 2018

• Implement stochastic weight averaging.

  Show description for 94f48bb
  

  Ttl

  authored and

  gcp

  committed
  94f48bb

  Copy full SHA for 94f48bb

Commits on Feb 9, 2018

• Add restrict, clean xgemm.

  Show description for 7330d4d
  

  Ttl

  authored and

  gcp

  committed
  7330d4d

  Copy full SHA for 7330d4d

• Fix debug data chunking.

  Show description for ff5a78d
  

  Ttl

  authored and

  gcp

  committed
  ff5a78d

  Copy full SHA for ff5a78d

Commits on Feb 5, 2018

• Remove unnecessary size check, release lock earlier.

  

  Ttl

  authored and

  gcp

  committed
  bd3081c

  Copy full SHA for bd3081c

• More efficient residual connection.

  Show description for ab0d091
  

  Ttl

  authored and

  gcp

  committed
  ab0d091

  Copy full SHA for ab0d091

• Move output head convolutions to GPU.

  

  Ttl

  authored and

  gcp

  committed
  f4a36c1

  Copy full SHA for f4a36c1

• Use pinned memory for device -> host transfer.

  

  Ttl

  authored and

  gcp

  committed
  beb4fe1

  Copy full SHA for beb4fe1

• Fix ignored biases.

  Show description for ebbf2ea
  

  Ttl

  authored and

  gcp

  committed
  ebbf2ea

  Copy full SHA for ebbf2ea

• Faster network parsing and error checking.

  Show description for ec0de09
  

  Ttl

  authored and

  gcp

  committed
  ec0de09

  Copy full SHA for ec0de09

Commits on Jan 24, 2018

• Add net2net.py for enlarging networks.

  Show description for 6f93c0e
  

  Ttl

  authored and

  gcp

  committed
  6f93c0e

  Copy full SHA for 6f93c0e

Commits on Jan 23, 2018

• Check that filters/blocks match in net_to_model.

  Show description for a3a2e0f
  

  Ttl

  authored and

  gcp

  committed
  a3a2e0f

  Copy full SHA for a3a2e0f

• Remove filter size restriction in BatchNorm.

  Show description for dfba8bc
  

  Ttl

  authored and

  gcp

  committed
  dfba8bc

  Copy full SHA for dfba8bc

• Fix V matrix size in Winograd CPU convolution.

  Show description for 420c9f2
  

  Ttl

  authored and

  gcp

  committed
  420c9f2

  Copy full SHA for 420c9f2

Pagination

Previous

Next