LeelaSabaki icon indicating copy to clipboard operation
LeelaSabaki copied to clipboard

LeelaSabaki v0.3.0 + Leela Zero 0.15 crash on macOS High Sierra (10.13.6 )

Open tangjin opened this issue 5 years ago • 9 comments

Hi, yichuan I want to run the LeelaSabaki on my Mac-pro, but the process LeelaSabaki and LeelaZero crashed about 5 seconds later. Can you help me? Thank you very much !

1、LeelaSabaki v0.3.0 2、Leela Zero 0.15 3、macOS High Sierra (10.13.6 ) 4、MacBook Pro (13-inch, 2018, Four Thunderbolt 3 Ports) 5、/data0/leela-zero/leelasabaki-macos --flat --heatmap /data0/leela-zero/src/leelaz -t 4 -w /data0/leela-zero/best-network --playouts 10 --noponder

Using 4 thread(s). RNG seed: 9153934374649917980 Detecting residual layers...v1...256 channels...40 blocks. Initializing OpenCL. Detected 1 OpenCL platforms. Platform version: OpenCL 1.2 (May 24 2018 22:33:53) Platform profile: FULL_PROFILE Platform name: Apple Platform vendor: Apple Device ID: 0 Device name: Intel(R) Core(TM) i5-8259U CPU @ 2.30GHz Device type: CPU Device vendor: Intel Device driver: 1.1 Device speed: 2300 MHz Device cores: 8 CU Device score: 512 Device ID: 1 Device name: Intel(R) Iris(TM) Plus Graphics 655 Device type: GPU Device vendor: Intel Inc. Device driver: 1.2(Aug 14 2018 19:16:17) Device speed: 1050 MHz Device cores: 48 CU Device score: 612 Selected platform: Apple Selected device: Intel(R) Iris(TM) Plus Graphics 655 with OpenCL 1.2 capability. Loaded existing SGEMM tuning. Wavefront/Warp size: 8 Max workgroup size: 256 Max workgroup dimensions: 256 256 256

bash-3.2# ps aux|grep -i leela root 3572 0.0 0.0 4267768 908 s001 S+ 9:36下午 0:00.00 grep -i leela bash-3.2#

pic1

tangjin avatar Sep 28 '18 13:09 tangjin

Does this problem only occur when using LeelaSabaki, or will Leela Zero crash as well when attached to Sabaki?

yishn avatar Sep 28 '18 14:09 yishn

When I used Sabaki and Leela Zero, I can play 3 steps, then the Leela Zero crashed, but the Sabaki was running. Path: /data0/leela-zero/src/leelaz Arguments: -g -t 4 -w /data0/leela-zero/best-network --playouts 1 --noponder image

tangjin avatar Sep 28 '18 23:09 tangjin

Then it's not a LeelaSabaki problem. There's most likely a problem in your Leela Zero binary. Can you try running Leela Zero in your Terminal to see if it crashes when you play some moves?

yishn avatar Sep 29 '18 11:09 yishn

When I used Leela Zero only, I can play many steps, the Leela Zero didn't crashed. /data0/leela-zero/src/leelaz --noponder -t 4 --playouts 1 -w /data0/leela-zero/weights.txt image

tangjin avatar Sep 29 '18 14:09 tangjin

Ok, then I need more information on the crash: How do you know Leela Zero crashes; does its process simply disappear? Does Sabaki give any errors or is it stuck with a busy cursor? Did you wait for 37 seconds for Leela to respond? Also, is it normal for Leela Zero to answer at A16; this doesn't seem like a normal move?

yishn avatar Sep 29 '18 14:09 yishn

Hi, Yishn

  1. It is stuck with a busy cursor, and the process disappear in my Terminal.
  2. ps ax|grep leela, cannot find the process.
  3. I wait for 120s
  4. Leela Zero answered at A16, because I set the playouts=1 ?

Thank you.

tangjin avatar Sep 29 '18 15:09 tangjin

@tangjin I have exact same issue on High Sierra, sometimes even the whole system halts. Calling it with your arguments: leelaz --noponder -t 4 --playouts 1 -w weights is fine, but normal functionality is not possible.

This problem was described in this issue: https://github.com/gcp/leela-zero/issues/1677 Hopefully it will be fixed in the next version of leela-zero

Jendker avatar Oct 24 '18 18:10 Jendker

@Jendker Thanks a lot.

tangjin avatar Oct 30 '18 02:10 tangjin

I discussed the issue with OpenCL here: https://github.com/gcp/leela-zero/issues/1951. The problem discussed in mentioned issue is already fixed, so the only solution is the driver fix, but as Apple is abandoning OpenCL (OpenCL is deprecated https://developer.apple.com/macos/whats-new/) this will likely not happen... Still, running on CPU works fine, so it's not that all bad.

Jendker avatar Oct 30 '18 07:10 Jendker