Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Test binaries (Windows) for release #1945

Closed
gcp opened this issue Oct 22, 2018 · 52 comments

Comments

Projects
None yet
@gcp
Copy link
Member

commented Oct 22, 2018

https://sjeng.org/ftp/work/leela-zero-0.15.999-win64.zip

Last time I tried to update OpenBLAS (and the time before that) we had issues with bugs (e.g. #1648). I updated again and made a test build for the v0.16 release. Can you guys give it a spin?

This is the 64-bit GPU build.

@gcp gcp added the help wanted label Oct 22, 2018

@gcp gcp changed the title Test binaries for release Test binaries (Windows) for release Oct 22, 2018

@Friday9i

This comment has been minimized.

Copy link

commented Oct 22, 2018

Works well, no issue :-).
And it seems significantly faster on my GTX1080 (and Windows 10) than v15, even without a full-tuner: this v15.999 used ~3.02s/move (on the first 2 games around 180 moves each) vs 5 to 6s/move (for typical resign games) with v15 and ~4.5s/move with the next version I compiled 2 weeks ago on my computer. I then tested after --full-tuner and it gives comparable results (around 3s/move)
Thanks a lot, that seems very nice!
Update: I did not manage to run Gogui with it (while it worked perfectly before), while Sabaki works well. No idea why Gogui doesn't work, I tried several times and parameters are the same...

@l1t1

This comment has been minimized.

Copy link

commented Oct 22, 2018

nice job

@Borthralla

This comment has been minimized.

Copy link

commented Oct 22, 2018

When I run leelaz.exe with the -g option, it stops doing anything after "Setting max tree size to 3736 MiB and cache size to 415 MiB.". Also, when I run it without -g and try and use 'genmove b', it doesn't seem to do anything.
AutoGTP seems to work fine though.

@l1t1

This comment has been minimized.

Copy link

commented Oct 22, 2018

there is already an 2da87ea8da0f54e87b70159e6bb82811b61d1c31091b6e019fbe62aeaa803b9c file in my networks directory, but when I type

D:\tool\leela15>autogtp -k sga
AutoGTP v16
Using 1 thread(s) for GPU(s).
Starting tuning process, please wait...

the autogtp still downloads the 2da87ea8da0f54e87b70159e6bb82811b61d1c31091b6e019fbe62aeaa803b9c.gz file at that directory .

@l1t1

This comment was marked as off-topic.

Copy link

commented Oct 23, 2018

what is the meaning of time_settings 0 1 0 in

Got new job: selfplay
net: 2da87ea8da0f54e87b70159e6bb82811b61d1c31091b6e019fbe62aeaa803b9c.
Engine has started.
time_settings 0 1 0
Thinking time set.
@barrtgt

This comment has been minimized.

Copy link

commented Oct 23, 2018

Doesn't load with Gogui 1.4.9 gui. I used 1 thread. In Sabaki 0.35.1, reported playouts are 10x faster than benchmark in command line. Loading a weight in command line with leelaz.exe and doing a genmove doesn't seem to function requires pressing enter twice or using -g option causes leelaz to stop after setting max tree size . When using -g command, the output will stop after setting max tree size, a genmove can be entered after with an additional enter press and it will work as stated below. Autogtp, validation, benchmark and Sabaki works fine.

Benchmarks

blocks filters single half
40 256 88 95
20 256 164 192
20 224 203 error
15 192 284 322
10 128 605 635
6 128 891 869
5 64 2335 2318
4 32 1849 1702
1 8 2882 2831

Windows 10 64-bit. ATI RX 480. I7-7700K.

Single vs. half precision with net a53 at 10 playouts:
1600 games 785:815 49.06%

Version 0.15 vs. 0.15.999 with net a53 at 10 playouts:
1600 games 806:794 50.37%

@l1t1

This comment was marked as off-topic.

Copy link

commented Oct 23, 2018

gcp's file

Got new job: selfplay
net: 2da87ea8da0f54e87b70159e6bb82811b61d1c31091b6e019fbe62aeaa803b9c.
Engine has started.
time_settings 0 1 0
Thinking time set.
1 (B Q4) 2 (W D16) 3 (B Q16) 4 (W J13) 5 (B D4) 6 (W C14) 7 (B E18) 8 (W D17) 9 (B J17) 10 (W F17) 11 (B C13) 12 (W C11)
 13 (B D14) 14 (W C15) 15 (B O8) 16 (W C6) 17 (B H14) 18 (W R17) 19 (B Q17) 20 (W R16) 21 (B R14) 22 (W R15) 23 (B Q15)
24 (W S14) 25 (B E17) 26 (W E16) 27 (B F18) 28 (W G17) 29 (B G18) 30 (W H17) 31 (B H18) 32 (W J16) 33 (B K17) 34 (W J14)
 35 (B R18) 36 (W S18) 37 (B S15) 38 (W T15) 39 (B S13) 40 (W S16) 41 (B F3) 42 (W R3) 43 (B R4) 44 (W Q3) 45 (B P4) 46
(W O2) 47 (B R13) 48 (W D9) 49 (B G12) 50 (W H15) 51 (B D11) 52 (W D12) 53 (B D10) 54 (W C10) 55 (B E9) 56 (W E8) 57 (B
F9) 58 (W D8) 59 (B E12) 60 (W C12) 61 (B F8) 62 (W N3) 63 (B C5) 64 (W P9) 65 (B N10) 66 (W R9) 67 (B K15) 68 (W K16) 6
9 (B L16) 70 (W L15) 71 (B G15) 72 (W F15) 73 (B S3) 74 (W S2) 75 (B S5) 76 (W T3) 77 (B K14) 78 (W J15) 79 (B M15) 80 (
W L17) 81 (B M16) 82 (W L18) 83 (B K18) 84 (W K19) 85 (B J18) 86 (W N17) 87 (B P18) 88 (W R7) 89 (B N4) 90 (W R19) 91 (B
 O3) 92 (W O18) 93 (B O19) 94 (W Q18) 95 (B O17) 96 (W N18) 97 (B M17) 98 (W M18) 99 (B N14) 100 (W O16) 101 (B P17) 102
 (W P19) 103 (B Q19) 104 (W O14) 105 (B P14) 106 (W N16) 107 (B O13) 108 (W P19) 109 (B D18) 110 (W C18) 111 (B Q19) 112
 (W D19) 113 (B O15) 114 (W E19) 115 (B N15) 116 (W F19) 117 (B N2) 118 (W S4) 119 (B R6) 120 (W S7) 121 (B P2) 122 (W E
5) 123 (B S10) 124 (W O7) 125 (B P8) 126 (W N8) 127 (B P7) 128 (W T5) 129 (B T6) 130 (W G5) 131 (B B6) 132 (W S6) 133 (B
 T4) 134 (W O9) 135 (B N9) 136 (W T5) 137 (B C17) 138 (W B17) 139 (B T4) 140 (W P6) 141 (B N7) 142 (W T5) 143 (B B18) 14
4 (W C16) 145 (B T4) 146 (W M8) 147 (B O6) 148 (W T5) 149 (B N19) 150 (W R5) 151 (B P16) 152 (W P19) 153 (B B19) 154 (W
Q19) 155 (B J19) 156 (W C19) 157 (B L19) 158 (W M19) 159 (B O19) 160 (W K19) 161 (B D6) 162 (W N19) 163 (B F7) 164 (W G3
) 165 (B F2) 166 (W K4) 167 (B K6) 168 (W F4) 169 (B G2) 170 (W H3) 171 (B F6) 172 (W J6) 173 (B H6) 174 (W J7) 175 (B H
5) 176 (W L6) 177 (B K7) 178 (W E3) 179 (B H2) 180 (W N5) 181 (B O5) 182 (W M3) 183 (B M4) 184 (W J2) 185 (B E4) 186 (W
L4) 187 (B K8) 188 (W M5) 189 (B M7) 190 (W E2) 191 (B D3) 192 (W H1) 193 (B D2) 194 (W O4) 195 (B N4) 196 (W M2) 197 (B
 O1) 198 (W K5) 199 (B L19) 200 (W O19) 201 (B G19) 202 (W C17) 203 (B M6) 204 (W H4) 205 (B J5) 206 (W J4) 207 (B K3) 2
08 (W K2) 209 (B resign) Game has ended.
Score: W+Resign
Winner: white
Uploading game: ab139821dbe44336918edafa53d18c2f.sgf for network 2da87ea8da0f54e87b70159e6bb82811b61d1c31091b6e019fbe62a
eaa803b9c
Game data 07aa206cd21f44e2bbb9cb22431052ac92fb21f2a2af3d941c5a2e54dcc4bc43 stored in database
2 game(s) (2 self played and 0 matches) played in 146 minutes = 4382 seconds/game, 39125 ms/move, last game took 7572 se
conds.

7572/209=36.22967 sec/move

my next binary complied at Sep 21 from a group

Got new job: selfplay
net: 2da87ea8da0f54e87b70159e6bb82811b61d1c31091b6e019fbe62aeaa803b9c.
Engine has started.
Infinite thinking time set.
1 (B D4) 2 (W Q16) 3 (B R17) 4 (W R16) 5 (B Q17) 6 (W P16) 7 (B P17) 8 (W N17) 9 (B D16) 10 (W Q4) 11 (B O17) 12 (W O16)
 13 (B N18) 14 (W M17) 15 (B M18) 16 (W Q10) 17 (B R3) 18 (W R4) 19 (B Q3) 20 (W P4) 21 (B O2) 22 (W C6) 23 (B C5) 24 (W
 D6) 25 (B F4) 26 (W C14) 27 (B D10) 28 (W E14) 29 (B C12) 30 (W G9) 31 (B F16) 32 (W F12) 33 (B F11) 34 (W G11) 35 (B G
12) 36 (W E12) 37 (B D8) 38 (W H11) 39 (B F7) 40 (W C8) 41 (B C15) 42 (W B15) 43 (B B16) 44 (W E8) 45 (B B14) 46 (W D9)
47 (B R12) 48 (W P12) 49 (B L17) 50 (W L16) 51 (B K16) 52 (W K17) 53 (B L18) 54 (W L15) 55 (B K15) 56 (W L14) 57 (B S10)
 58 (W R9) 59 (B S9) 60 (W S8) 61 (B S15) 62 (W S16) 63 (B R14) 64 (W Q13) 65 (B K14) 66 (W K13) 67 (B L13) 68 (W R8) 69
 (B R13) 70 (W J13) 71 (B H15) 72 (W J5) 73 (B M14) 74 (W N16) 75 (B L12) 76 (W N12) 77 (B N15) 78 (W M16) 79 (B O13) 80
 (W N13) 81 (B O12) 82 (W O11) 83 (B O14) 84 (W Q14) 85 (B P11) 86 (W R15) 87 (B Q12) 88 (W P13) 89 (B N11) 90 (W O10) 9
1 (B N10) 92 (W M11) 93 (B N9) 94 (W M12) 95 (B T15) 96 (W L11) 97 (B N3) 98 (W M4) 99 (B K12) 100 (W J12) 101 (B J6) 10
2 (W K5) 103 (B K6) 104 (W G5) 105 (B L9) 106 (W B13) 107 (B D14) 108 (W A14) 109 (B D13) 110 (W B17) 111 (B A16) 112 (W
 E17) 113 (B D17) 114 (W E16) 115 (B E15) 116 (W F15) 117 (B D15) 118 (W G16) 119 (B M6) 120 (W E2) 121 (B G4) 122 (W H5
) 123 (B C2) 124 (W J17) 125 (B H16) 126 (W G17) 127 (B G15) 128 (W F14) 129 (B H17) 130 (W H18) 131 (B G18) 132 (W F17)
 133 (B J18) 134 (W P10) 135 (B J3) 136 (W D18) 137 (B C18) 138 (W F18) 139 (B N4) 140 (W K3) 141 (B K2) 142 (W G19) 143
 (B K18) 144 (W B18) 145 (B C17) 146 (W B12) 147 (B C9) 148 (W C11) 149 (B E9) 150 (W C19) 151 (B K11) 152 (W K10) 153 (
B F8) 154 (W F6) 155 (B H8) 156 (W E7) 157 (B D11) 158 (W D12) 159 (B F9) 160 (W B8) 161 (B B9) 162 (W L6) 163 (B L7) 16
4 (W L5) 165 (B N6) 166 (W B5) 167 (B B4) 168 (W D5) 169 (B C4) 170 (W F5) 171 (B K9) 172 (W J10) 173 (B C13) 174 (W B10
) 175 (B A9) 176 (W L3) 177 (B B14) 178 (W T16) 179 (B S14) 180 (W C14) 181 (B B11) 182 (W A11) 183 (B B14) 184 (W T13)
185 (B S12) 186 (W C14) 187 (B H3) 188 (W R11) 189 (B S11) 190 (W L2) 191 (B B14) 192 (W T9) 193 (B J1) 194 (W D7) 195 (
B T10) 196 (W T12) 197 (B Q11) 198 (W R10) 199 (B resign) Game has ended.
Score: W+Resign
Winner: white
Uploading game: 8edca0ede67847edb92018b00b6a9529.sgf for network 2da87ea8da0f54e87b70159e6bb82811b61d1c31091b6e019fbe62a
eaa803b9c
Game data 2e680cc288aaa8ea5c44acdf1356c27f28065dd12d46cbd2302bb02a71129dc8 stored in database
10 game(s) (10 self played and 0 matches) played in 963 minutes = 5783 seconds/game, 39915 ms/move, last game took 7536
seconds.

7536/199=37.86935

leelaz15next.zip

@Vargooo

This comment has been minimized.

Copy link

commented Oct 23, 2018

Doesn't work with gogui-twogtp 1.4.10 on my laptop :
It creates a file named leelaz_opencl_tuning, containing

0;XgemmBatched;192;100;192;16; -DKWG=16 -DKWI=2 -DMDIMA=16 -DMDIMC=16 -DMWG=64 -DNDIMB=8 -DNDIMC=8 -DNWG=64 -DSA=1 -DSB=1 -DSTRM=0 -DSTRN=0 -DVWM=4 -DVWN=4;OpenCL: NVIDIA Corporation GeForce GTX 965M @ 1150MHz
0;XgemmBatched;224;100;224;16; -DKWG=16 -DKWI=8 -DMDIMA=8 -DMDIMC=8 -DMWG=32 -DNDIMB=8 -DNDIMC=8 -DNWG=64 -DSA=1 -DSB=1 -DSTRM=0 -DSTRN=0 -DVWM=4 -DVWN=2;OpenCL: NVIDIA Corporation GeForce GTX 965M @ 1150MHz
0;XgemmBatched;256;100;256;16; -DKWG=16 -DKWI=8 -DMDIMA=16 -DMDIMC=16 -DMWG=64 -DNDIMB=8 -DNDIMC=8 -DNWG=64 -DSA=1 -DSB=1 -DSTRM=0 -DSTRN=0 -DVWM=4 -DVWN=4;OpenCL: NVIDIA Corporation GeForce GTX 965M @ 1150MHz
0;XgemmBatched;64;100;64;16; -DKWG=16 -DKWI=2 -DMDIMA=8 -DMDIMC=8 -DMWG=32 -DNDIMB=8 -DNDIMC=8 -DNWG=32 -DSA=1 -DSB=1 -DSTRM=0 -DSTRN=0 -DVWM=4 -DVWN=4;OpenCL: NVIDIA Corporation GeForce GTX 965M @ 1150MHz
0;XgemmBatched;256;25;256;36; -DKWG=16 -DKWI=8 -DMDIMA=8 -DMDIMC=8 -DMWG=32 -DNDIMB=8 -DNDIMC=8 -DNWG=32 -DSA=1 -DSB=1 -DSTRM=0 -DSTRN=0 -DVWM=4 -DVWN=4;OpenCL: NVIDIA Corporation GeForce GTX 965M @ 1150MHz
0;XgemmBatchedHalf;256;25;256;36; -DKWG=16 -DKWI=2 -DMDIMA=16 -DMDIMC=16 -DMWG=64 -DNDIMB=8 -DNDIMC=8 -DNWG=32 -DSA=1 -DSB=1 -DSTRM=0 -DSTRN=0 -DVWM=4 -DVWN=2;OpenCL: NVIDIA Corporation GeForce GTX 965M @ 1150MHz

and does nothing else. The black cmd window stays open, and the computer doesn't freeze, but nothing happens.

@gcp

This comment has been minimized.

Copy link
Member Author

commented Oct 23, 2018

I did not manage to run Gogui with it (while it worked perfectly before)

Ok, we'll have to debug this then.

@gcp

This comment has been minimized.

Copy link
Member Author

commented Oct 23, 2018

When I run leelaz.exe with the -g option, it stops doing anything after "Setting max tree size to 3736 MiB and cache size to 415 MiB."

This is the expected result. You need to enter GTP commands if you run in GTP mode.

Also, when I run it without -g and try and use 'genmove b', it doesn't seem to do anything.

You don't even get "Thinking at most ..." output? Does running the benchmark work?

@gcp

This comment has been minimized.

Copy link
Member Author

commented Oct 23, 2018

In Sabaki 0.35.1, reported playouts are 10x faster than benchmark in command line.

This is the expected result. We now automatically detect and use symmetric positions in the opening. The benchmark is from a non-symmetric position.

@gcp

This comment was marked as off-topic.

Copy link
Member Author

commented Oct 23, 2018

what is the meaning of time_settings 0 1 0 in

Read the GTP specification.

@gcp

This comment has been minimized.

Copy link
Member Author

commented Oct 23, 2018

Loading a weight in command line with leelaz.exe and doing a genmove doesn't seem to function and using -g option causes leelaz to stop after setting max tree size.

Hmm, this matches @Borthralla 's report. What hardware are you using? (GPU/CPU)

@gcp

This comment has been minimized.

Copy link
Member Author

commented Oct 23, 2018

the autogtp still downloads the 2da87ea8da0f54e87b70159e6bb82811b61d1c31091b6e019fbe62aeaa803b9c.gz file at that directory .

This is the expected result: the new AutoGTP no longer unzips files as Leela can read them directly. (I should probably remove the gzip binary in the package). The unzipped file you have there is from the previous version.

@gcp

This comment has been minimized.

Copy link
Member Author

commented Oct 23, 2018

If you are having the "freeze" problem, can you check if this is caused by having threads > 1? AutoGTP will set Leela to 1 thread, whereas the default is 2.

If you have this problem, can you try copying the OpenBLAS DLL from the previous release and seeing if that makes the problem go away?

@Friday9i

This comment has been minimized.

Copy link

commented Oct 23, 2018

I copied the previous openBLAS.dll and it apparently works as with the one from your package for autopgtp. For Gogui, it doesn't solve the "freeze" problem: leelaz.exe appears in the threads with 0% usage, and nothing happens. There is 1 thread is my Gogui instructions (-t 1) and I tried -t 2 before without success.
Note: I'm using the same command as before "(my folder)\gogui-twogtp.exe -black "leelaz.exe --gtp -w networks\2da87ea8da0f54e87b70159e6bb82811b61d1c31091b6e019fbe62aeaa803b9c.gz --noponder -v 10 -r 5 -t 1" -white "leelaz.exe --gtp -w networks\2da87ea8da0f54e87b70159e6bb82811b61d1c31091b6e019fbe62aeaa803b9c.gz --noponder -v 10 -r 5 -t 1" -auto -alternate -games 200 -sgffile (my folder)
"

@gcp

This comment has been minimized.

Copy link
Member Author

commented Oct 23, 2018

OK, not sure what is going on then. I just tested GoGUI locally and it worked :-/

@gcp

This comment has been minimized.

Copy link
Member Author

commented Oct 23, 2018

Ah, but when you say GoGUI doesn't work, do you mean the GUI itself, or the gogui-twogtp command?

@gcp

This comment has been minimized.

Copy link
Member Author

commented Oct 23, 2018

./gogui-twogtp -black "leelaz --gtp -w autogtp/networks/a53cd929d2b3f2c939d41757291aa70b6992a215d31e386051beb7afecaa2113.gz --noponder -v 10 -r 5 -t 1" -white "leelaz --gtp -w autogtp/networks/2da87ea8da0f54e87b70159e6bb82811b61d1c31091b6e019fbe62aeaa803b9c.gz --noponder -v 10 -r 5 -t 1" -auto -alternate -games 200 -sgffile foo -verbose

This worked for me.

If you add the -verbose switch, where does it halt?

@Friday9i

This comment has been minimized.

Copy link

commented Oct 23, 2018

I use gogui-twogtp. Not at home anymore, I'll try -verbose tonight...

@Borthralla

This comment has been minimized.

Copy link

commented Oct 23, 2018

When I run leelaz.exe with the -g option, it stops doing anything after "Setting max tree size to 3736 MiB and cache size to 415 MiB."

This is the expected result. You need to enter GTP commands if you run in GTP mode.

Also, when I run it without -g and try and use 'genmove b', it doesn't seem to do anything.

You don't even get "Thinking at most ..." output? Does running the benchmark work?

You were right that the -g option was just waiting for the gtp command, but calling 'genmove b' doesn't seem to be working properly. Nothing is printed after the command is entered and my gpu usage is 0%. Seems like its not doing anything.

I tried running the benchmark with a 1080ti/i7-6700k just now and it seems to work:

D:\LZ\leela-zero-0.15.999-win64>leelaz -w network.gz --benchmark
Thinking at most 2678400.0 seconds...
NN eval=0.531805
Playouts: 336, Win: 52.96%, PV: D3 C4 D5 C5 D6 B7 D16 Q4 P17 Q15 R6 O3 R4 R3 S3 R5 S4 Q5 S5 R2
Playouts: 978, Win: 53.01%, PV: D3 C4 D5 C5 D6 B7 D16 Q4 P17 Q15 R6 O3 R4 R3 S3 R5 S4 Q5 S5 R2 Q6 S7 S6 R8
Playouts: 1598, Win: 53.02%, PV: D3 C4 D5 C5 D6 B7 D16 Q4 P17 Q15 R6 O3 R4 R3 S3 R5 S4 Q5 S5 R2 Q6 S7 S6 R8 L3
Playouts: 2217, Win: 53.03%, PV: D3 C4 D5 C5 D6 B7 D16 Q4 P17 Q15 R6 O3 R4 R3 S3 R5 S4 Q5 S5 R2 Q6 S7 S6 R8 L3 M4 L4
Playouts: 2868, Win: 53.04%, PV: D3 C4 D5 C5 D6 B7 D16 Q4 P17 Q15 R6 O3 R4 R3 S3 R5 S4 Q5 S5 R2 Q6 S7 S6 R8 L3 M4 L4

  D3 ->    1243 (V: 53.26%) (N: 31.45%) PV: D3 C4 D5 C5 D6 B7 D16 Q4 P17 Q15 R6 O3 R4 R3 S3 R5 S4 Q5 S5 R2 Q6 S7 S6 R8 L3 M4 L4
  C4 ->    1122 (V: 53.16%) (N: 30.72%) PV: C4 D3 E4 E3 F4 G2 Q4 D16 P17 Q15 F17 C14 D17 C17 C18 E17 D18 E16 E18 B17 F16 G18 F18 L17 J17 O16 C12 E14
 D16 ->     432 (V: 52.61%) (N: 17.14%) PV: D16 D3 E4 E3 F4 Q4 C4 G2 P17 Q15 D10 M17 O16 F17 P14 Q13 K17 M15
  Q4 ->     379 (V: 52.49%) (N: 16.08%) PV: Q4 C4 D16 D5 P17 Q15 M17 R3 R4 Q3 O3 P3 P4 O2 R12 Q17 P16 R14 N3
 D17 ->      23 (V: 53.26%) (N:  0.59%) PV: D17 D3 E4 E3 F4 Q4 C4 G2 P17 C15
14.6 average depth, 34 max depth
2485 non leaf nodes, 1.29 average children
3200 visits, 1105059 nodes, 3199 playouts, 233 n/s
@gcp

This comment has been minimized.

Copy link
Member Author

commented Oct 23, 2018

Nothing is printed after the command is entered and my gpu usage is 0%. It just seems not to be doing anything.

Can you try the DLL replacement trick, or setting threads=1?

@Borthralla

This comment has been minimized.

Copy link

commented Oct 23, 2018

Nothing is printed after the command is entered and my gpu usage is 0%. It just seems not to be doing anything.

Can you try the DLL replacement trick, or setting threads=1?

I tried replacing libopenblas.dll with the same file from the .15 release and it did not fix the problem. Changing the number of threads to 1 also did not solve the problem :(

@gcp

This comment has been minimized.

Copy link
Member Author

commented Oct 23, 2018

Maybe I'm stating the obvious, but does entering twice work? Another thing to check is if you didn't accidentally click in the window and enable text selection (this locks up Windows consoles).

If that doesn't work I sure hope I can reproduce this on my Windows machine...

@Borthralla

This comment has been minimized.

Copy link

commented Oct 23, 2018

Maybe I'm stating the obvious, but does entering twice work? Another thing to check is if you didn't accidentally click in the window and enable text selection (this locks up Windows consoles).

If that doesn't work I sure hope I can reproduce this on my Windows machine...

I tried entering it twice, didn't seem to do anything. I performed the exact same actions with the .15 release and it behaved as expected, so I don't think its an issue with the console.

@jillybob

This comment has been minimized.

Copy link

commented Oct 23, 2018

I'm having an issue running this as a bot on Online Go Server. Was previously using latest official release with no problems on the same command. Lizzie works with the 15.999 binaries but for some reason, it doesn't work with GTP. Here is the output: Maybe @roy7 has input. It seems everything works fine, however it stops after 5 visits. Nothing crashes, it just stops.

C:\Windows\System32>gtp2ogs --botid AquaBotTest --apikey REDACTED -d --boardsize 19 --nopause --maxmaintime 1200 --persist --greeting "Hi you are playing against Leela Zero 40b" --farewell Thankyou --minrank 9d --minperiods 2 --maxperiods 10 --minperiodtime 10 --maxperiodtime 30 --maxhandicap 0 --unrankedonly -- C:\Users\User1\Desktop\leela-zero-0.15-win64\leelaz.exe --gpu 0 -w C:\Users\User1\Desktop\leela-zero-0.15-win64\network.gz --gtp -v 5000 --logfile log.txt -r 1 -t 4
Oct 23 20:00:15 gtp2ogs.js:1822 # Connecting to https://online-go.com:443
Oct 23 20:00:15 gtp2ogs.js:1822 # Connected
Oct 23 20:00:16 gtp2ogs.js:1822 # Bot is user id: 495496
Oct 23 20:00:16 gtp2ogs.js:1822 # active_game: {"id":14966139,"phase":"play","name":"Friendly Match","player_to_move":495496,"width":19,"height":19,"move_number":0,"paused":0,"private":false,"black":{"username":"AquaBotTest","id":495496,"rank":14,"professional":false,"accepted":false,"ratings":{"overall":{"deviation":87.08471387552044,"rating":2547.579838757673,"volatility":0.060217302304635824,"games_played":48}}},"white":{"username":"RoyalZero","id":484523,"rank":25,"professional":false,"accepted":false,"ratings":{"overall":{"deviation":87.89688759173441,"rating":2786.9138771814687,"volatility":0.06001107806109481,"games_played":5443}}},"time_per_move":60}
Oct 23 20:00:16 gtp2ogs.js:1028 [Game 14966139] Connecting to game.
Oct 23 20:00:16 gtp2ogs.js:1028 [Game 14966139] gamedata B RoyalZero [19x19]
Oct 23 20:00:16 gtp2ogs.js:1028 [Game 14966139] makeMove 0 is 0 != 0 ?
Oct 23 20:00:16 gtp2ogs.js:353 [14116] Starting C:\Users\User1\Desktop\leela-zero-0.15-win64\leelaz.exe --gpu 0 -w C:\Users\User1\Desktop\leela-zero-0.15-win64\network.gz --gtp -v 5000 --logfile log.txt -r 1 -t 4
Oct 23 20:00:16 gtp2ogs.js:1028 [Game 14966139] Starting new bot process [14116]
Oct 23 20:00:16 gtp2ogs.js:1028 [Game 14966139] State loading for new bot
Oct 23 20:00:16 gtp2ogs.js:353 [14116] >>> boardsize 19
Oct 23 20:00:16 gtp2ogs.js:353 [14116] >>> clear_board
Oct 23 20:00:16 gtp2ogs.js:353 [14116] >>> komi 7.5
Oct 23 20:00:16 gtp2ogs.js:353 [14116] >>> showboard
Oct 23 20:00:16 gtp2ogs.js:353 [14116] Generating move for game 14966139
Oct 23 20:00:16 gtp2ogs.js:1028 [Game 14966139] genmove b
Oct 23 20:00:16 gtp2ogs.js:353 [14116] >>> time_settings 3680 20 1
Oct 23 20:00:16 gtp2ogs.js:353 [14116] >>> time_left black 3493 0
Oct 23 20:00:16 gtp2ogs.js:353 [14116] >>> time_left white 3680 0
Oct 23 20:00:16 gtp2ogs.js:353 [14116] >>> genmove black
Oct 23 20:00:16 gtp2ogs.js:1028 [Game 14966139] clock: {"game_id":14966139,"current_player":495496,"black_player_id":495496,"white_player_id":484523,"title":"Friendly Match","last_move":1540285032485,"expiration":1540288732485,"black_time":{"thinking_time":3600,"periods":5,"period_time":20},"white_time":{"thinking_time":3600,"periods":5,"period_time":20},"start_mode":true,"now":1540285218987}
Oct 23 20:00:16 !!!!! gtp2ogs.js:361 [14116] stderr: Logging to log.txt.
Using 4 thread(s).
RNG seed: 12419442178578261501

Oct 23 20:00:16 !!!!! gtp2ogs.js:361 [14116] stderr: BLAS Core: Haswell

Oct 23 20:00:18 !!!!! gtp2ogs.js:361 [14116] stderr: Detecting residual layer
Oct 23 20:00:18 !!!!! gtp2ogs.js:361 [14116] stderr: s...v1...256 channels...
Oct 23 20:00:19 !!!!! gtp2ogs.js:361 [14116] stderr: 40 blocks.

Oct 23 20:00:22 !!!!! gtp2ogs.js:361 [14116] stderr: Initializing OpenCL (autodetecting precision).

Oct 23 20:00:22 !!!!! gtp2ogs.js:361 [14116] stderr: Detected 1 OpenCL platforms.
Platform version: OpenCL 1.2 CUDA 9.2.217
Platform profile: FULL_PROFILE
Platform name: NVIDIA CUDA
Platform vendor: NVIDIA Corporation

Oct 23 20:00:22 !!!!! gtp2ogs.js:361 [14116] stderr: Device ID: 0
Device name: GeForce GTX 970
Device type: GPU
Device vendor: NVIDIA Corporation
Device driver: 399.07
Device speed: 1177 MHz
Device cores: 13 CU
Device score: 1112
Selected platform: NVIDIA CUDA
Selected device: GeForce GTX 970
with OpenCL 1.2 capability.

Oct 23 20:00:22 !!!!! gtp2ogs.js:361 [14116] stderr: Half precision compute su
Oct 23 20:00:22 !!!!! gtp2ogs.js:361 [14116] stderr: pport: No.
Detected 1 OpenCL platforms.
Platform version: OpenCL 1.2 CUDA 9.2.217
Platform profile: FULL_PROFILE
Platform name: NVIDIA CUDA
Platform vendor: NVIDIA Corporation
Device ID: 0
Device name: GeForce GTX 970
Device type: GPU
Device vendor: NVIDIA Corporation
Device driver: 399.07
Device speed: 1177 MHz
Device cores: 13 CU
Device score: 1112
Selected platform: NVIDIA CUDA
Selected device: GeForce GTX 970
with OpenCL 1.2 capability.

Oct 23 20:00:22 !!!!! gtp2ogs.js:361 [14116] stderr: Half precision comp
Oct 23 20:00:22 !!!!! gtp2ogs.js:361 [14116] stderr: ute support: No.

Oct 23 20:00:22 !!!!! gtp2ogs.js:361 [14116] stderr:
Started OpenCL SGEMM
Oct 23 20:00:22 !!!!! gtp2ogs.js:361 [14116] stderr: tuner.

Oct 23 20:00:22 !!!!! gtp2ogs.js:361 [14116] stderr: Will try 290 valid
Oct 23 20:00:22 !!!!! gtp2ogs.js:361 [14116] stderr: configurations.

Oct 23 20:00:22 !!!!! gtp2ogs.js:361 [14116] stderr: (1/290) KWG=16 KWI=2 MDIMA=8 MDIMC=8 MWG=16 NDIMB=8 NDIMC=8 NWG=16 SA=1 SB=1 STRM=0 STRN=0 VWM=2 VWN=2 0.1993 ms (591.8 GFLOPS)

Oct 23 20:00:22 !!!!! gtp2ogs.js:361 [14116] stderr: (2/290) KWG=16 KWI=2 MDIMA=8 MDIMC
Oct 23 20:00:22 !!!!! gtp2ogs.js:361 [14116] stderr: =8 MWG=32 NDIMB=8 NDIMC=8 NWG=16 SA=1 SB=1 STRM=0 STRN=0 VWM=2 VWN=2 0.1839 ms (641.4 GFLOPS)

Oct 23 20:00:23 !!!!! gtp2ogs.js:361 [14116] stderr: (4/290) KWG=16 KWI
Oct 23 20:00:23 !!!!! gtp2ogs.js:361 [14116] stderr: =2 MDIMA=8 MDIMC=8 MWG=16 NDIMB=8 NDIMC=8 NWG=32 SA=1 SB=1 STRM=0 STRN=0 VWM=2 VWN=2 0.1752 ms (673.3 GFLOPS)

Oct 23 20:00:23 !!!!! gtp2ogs.js:361 [14116] stderr: (5/290) KWG=16 KWI=2 MDIMA=8 MDIMC=8 MWG=32 NDIMB=8 NDIMC=8 NWG=32 SA=1 SB=1 STRM=0 STRN=0 VWM=2 VWN=2
Oct 23 20:00:23 !!!!! gtp2ogs.js:361 [14116] stderr: 0.1232 ms (957.8 GFLOPS)

Oct 23 20:00:41 !!!!! gtp2ogs.js:361 [14116] stderr: (125/290) KWG=16 KWI=2 MDIMA=8 MDIMC=8 MWG=
Oct 23 20:00:41 !!!!! gtp2ogs.js:361 [14116] stderr: 32 NDIMB=8 NDIMC=8 NWG=32 SA=1 SB=1 STRM=0 STRN=0 VWM=4 VWN=2 0.1201 ms (981.9 GFLOPS)

Oct 23 20:00:42 !!!!! gtp2ogs.js:361 [14116] stderr: (131/290) KWG=32 KWI
Oct 23 20:00:42 !!!!! gtp2ogs.js:361 [14116] stderr: =2 MDIMA=8 MDIMC=8 MWG=32 NDIMB=8 NDIMC=8 NWG=32 SA=1 SB=1 STRM=0 STRN=0 VWM=4 VWN=2 0.1189 ms (992.5 GFLOPS)

Oct 23 20:00:44 !!!!! gtp2ogs.js:361 [14116] stderr: (139/290) KWG=32 KWI=2 MDIMA=16 MDIMC=1
Oct 23 20:00:44 !!!!! gtp2ogs.js:361 [14116] stderr: 6 MWG=64 NDIMB=8 NDIMC=8 NWG=32 SA=1 SB=1 STRM=0 STRN=0 VWM=4 VWN=2 0.1182 ms (997.7 GFLOPS)

Oct 23 20:00:47 !!!!! gtp2ogs.js:361 [14116] stderr: (164/290) KWG=32 KWI=8 MDIMA=8 MDIMC=8 MWG=32 NDIMB=8 NDIMC=8 NWG=32 SA=1 SB=1 STRM=0 STRN=0 VWM=4 VWN=2 0.1152 ms (1023.9 GFLOPS)

Oct 23 20:01:05 !!!!! gtp2ogs.js:361 [14116] stderr: (277/290) KWG=32 K
Oct 23 20:01:05 !!!!! gtp2ogs.js:361 [14116] stderr: WI=8 MDIMA=8 MDIMC=8 MWG=32 NDIMB=8 NDIMC=8 NWG=32 SA=1 SB=1 STRM=0 STRN=0 VWM=4 VWN=4 0.1136 ms (1038.7 GFLOPS)

Oct 23 20:01:07 !!!!! gtp2ogs.js:361 [14116] stderr: Wavefront/Warp size: 32
Oct 23 20:01:07 !!!!! gtp2ogs.js:361 [14116] stderr:
Max workgroup size: 1024
Max workgroup dimensions: 1024 1024 64

Oct 23 20:01:10 !!!!! gtp2ogs.js:361 [14116] stderr:
Started OpenCL SGEMM
Oct 23 20:01:10 !!!!! gtp2ogs.js:361 [14116] stderr: tuner.

Oct 23 20:01:10 !!!!! gtp2ogs.js:361 [14116] stderr: Will try 290 valid
Oct 23 20:01:10 !!!!! gtp2ogs.js:361 [14116] stderr: configurations.

Oct 23 20:01:10 !!!!! gtp2ogs.js:361 [14116] stderr: (1/290) KWG=16 KWI=2 MDIM
Oct 23 20:01:10 !!!!! gtp2ogs.js:361 [14116] stderr: A=8 MDIMC=8 MWG=16 NDIMB=8 NDIMC=8 NWG=16 SA=1 SB=1 STRM=0 STRN=0 VWM=2 VWN=2 0.1697 ms (695.3 GFLOPS)

Oct 23 20:01:10 !!!!! gtp2ogs.js:361 [14116] stderr: (2/290) KWG=16 KWI=2 MDI
Oct 23 20:01:10 !!!!! gtp2ogs.js:361 [14116] stderr: MA=8 MDIMC=8 MWG=32 NDIMB=8 NDIMC=8 NWG=16 SA=1 SB=1 STRM=0 STRN=0 VWM=2 VWN=2 0.1327 ms (888.7 GFLOPS)

Oct 23 20:01:10 !!!!! gtp2ogs.js:361 [14116] stderr: (3/290) KWG=16 KWI=2 MD
Oct 23 20:01:10 !!!!! gtp2ogs.js:361 [14116] stderr: IMA=8 MDIMC=8 MWG=64 NDIMB=8 NDIMC=8 NWG=16 SA=1 SB=1 STRM=0 STRN=0 VWM=2 VWN=2 0.1290 ms (914.7 GFLOPS)

Oct 23 20:01:10 !!!!! gtp2ogs.js:361 [14116] stderr: (5/290) KWG=16 KWI=2 MDIMA=8 MDIMC=8 MWG=32 NDI
Oct 23 20:01:10 !!!!! gtp2ogs.js:361 [14116] stderr: MB=8 NDIMC=8 NWG=32 SA=1 SB=1 STRM=0 STRN=0 VWM=2 VWN=2 0.1146 ms (1028.9 GFLOPS)

Oct 23 20:01:10 !!!!! gtp2ogs.js:361 [14116] stderr: (6/290) KWG=16 KWI=2 MDI
Oct 23 20:01:10 !!!!! gtp2ogs.js:361 [14116] stderr: MA=8 MDIMC=8 MWG=64 NDIMB=8 NDIMC=8 NWG=32 SA=1 SB=1 STRM=0 STRN=0 VWM=2 VWN=2 0.1038 ms (1136.7 GFLOPS)

Oct 23 20:01:10 !!!!! gtp2ogs.js:361 [14116] stderr: (67/290) KWG=16 KWI=8 MDIMA=8 MDIMC=8 MWG=64 NDIMB=8 NDIMC=8
Oct 23 20:01:10 !!!!! gtp2ogs.js:361 [14116] stderr: NWG=32 SA=1 SB=1 STRM=0 STRN=0 VWM=2 VWN=2 0.1017 ms (1159.8 GFLOPS)

Oct 23 20:01:11 !!!!! gtp2ogs.js:361 [14116] stderr: (126/290) KWG=16 KWI=2
Oct 23 20:01:11 !!!!! gtp2ogs.js:361 [14116] stderr: MDIMA=8 MDIMC=8 MWG=64 NDIMB=8 NDIMC=8 NWG=32 SA=1 SB=1 STRM=0 STRN=0 VWM=4 VWN=2 0.0976 ms (1208.6 GFLOPS)

Oct 23 20:01:11 !!!!! gtp2ogs.js:361 [14116] stderr: (136/290) KWG=16 KWI=
Oct 23 20:01:11 !!!!! gtp2ogs.js:361 [14116] stderr: 2 MDIMA=16 MDIMC=16 MWG=64 NDIMB=8 NDIMC=8 NWG=32 SA=1 SB=1 STRM=0 STRN=0 VWM=4 VWN=2 0.0973 ms (1212.9 GFLOPS)

Oct 23 20:01:11 !!!!! gtp2ogs.js:361 [14116] stderr: (159/290) KWG=16 KWI=8 MDIMA=8 MDIMC=8 MWG=64 NDIMB=8 NDIMC=8 NWG=32 SA=1 SB=1 STRM=0 STRN=0 VWM=4
Oct 23 20:01:11 !!!!! gtp2ogs.js:361 [14116] stderr: VWN=2 0.0947 ms (1245.8 GFLOPS)

Oct 23 20:01:12 !!!!! gtp2ogs.js:361 [14116] stderr: (274/290) KWG=16 KWI=8 MDIMA=8 MDIMC=8 MWG=64 NDIMB=8 NDIMC=8 NWG=32 SA=1
Oct 23 20:01:12 !!!!! gtp2ogs.js:361 [14116] stderr: SB=1 STRM=0 STRN=0 VWM=4 VWN=4 0.0938 ms (1257.4 GFLOPS)

Oct 23 20:01:12 !!!!! gtp2ogs.js:361 [14116] stderr: Wavefront/Warp size: 32
Oct 23 20:01:12 !!!!! gtp2ogs.js:361 [14116] stderr:
Max workgroup size: 1024
Max workgroup dimensions: 1024 1024 64

Oct 23 20:01:14 !!!!! gtp2ogs.js:361 [14116] stderr: Using OpenCL half precision
Oct 23 20:01:14 gtp2ogs.js:353 [14116] <<< =

Oct 23 20:01:14 !!!!! gtp2ogs.js:361 [14116] stderr: (at least 5% faster than single).
Setting max tree size to 3736 MiB and cache size to 415 MiB.

Oct 23 20:01:14 gtp2ogs.js:353 [14116] <<< =

Oct 23 20:01:14 gtp2ogs.js:353 [14116] <<< =

Oct 23 20:01:14 !!!!! gtp2ogs.js:361 [14116] stderr:
Passes: 0 Black (X) Prisoners: 0
Black (X) to move White (O) Prisoners: 0

a b c d e f g h j k l m n o p q r s t
19 . . . . . . . . . . . . . . . . . . . 19
18 . . . . . . . . . . . . . . . . . . . 18
17 . . . . . . . . . . . . . . . . . . . 17
16 . . . + . . . . . + . . . . . + . . . 16
15 . . . . . . . . . . . . . . . . . . . 15
14 . . . . . . . . . . . . . . . . . . . 14
13 . . . . . . . . . . . . . . . . . . . 13
12 . . . . . . . . . . . . . . . . . . . 12
11 . . . . . . . . . . . . . . . . . . . 11
10 . . . + . . . . . + . . . . . + . . . 10
9 . . . . . . . . . . . . . . . . . . . 9
8 . . . . . . . . . . . . . . . .
Oct 23 20:01:14 !!!!! gtp2ogs.js:361 [14116] stderr: . . . 8
7 . . . . . . . . . . . . . . . . . . . 7
6 . . . . . . . . . . . . . . . . . . . 6
5 . . . . . . . . . . . . . . . . . . . 5
4 . . . + . . . . . + . . . . . + . . . 4
3 . . . . . . . . . . . . . . . . . . . 3
2 . . . . . . . . . . . . . . . . . . . 2
1 . . . . . . . . . . . . . . . . . . . 1
a b c d e f g h j k l m n o p q r s t

Hash: 9A930BE1616C538E Ko-Hash: A14C933E7669946D

Black time: 01:00:00
White time: 01:00:00

Oct 23 20:01:14 gtp2ogs.js:353 [14116] <<< =

Oct 23 20:01:14 gtp2ogs.js:1028 [Game 14966139] State loaded for new bot
Oct 23 20:01:14 !!!!! gtp2ogs.js:361 [14116] stderr: NN eval=0.462006

Oct 23 20:01:14 !!!!! gtp2ogs.js:361 [14116] stderr:

Oct 23 20:01:14 !!!!! gtp2ogs.js:361 [14116] stderr: D4 -> 1 (V: 46.81%) (N: 19.28%) PV: D4
D16 -> 1 (V: 46.81%) (N: 18.93%) PV: D16
Q16 -> 1 (V: 46.78%) (N: 19.10%) PV: Q16
Q4 -> 1 (V: 46.72%) (N: 19.52%) PV: Q4
1.8 average depth, 2 max depth
1 non leaf nodes, 4.00 average children

5 visits, 1806 nodes

@gcp

This comment has been minimized.

Copy link
Member Author

commented Oct 23, 2018

Huh, why are you getting the board output in GTP mode?

@jillybob

This comment has been minimized.

Copy link

commented Oct 23, 2018

I'm not sure, the board output hasn't appeared in latest official release.

@gcp

This comment has been minimized.

Copy link
Member Author

commented Oct 23, 2018

It's an interesting data point. Maybe some error with GTP::setup_default_parameters being called twice?

Edit: Don't see how that's possible though.

@yishn

This comment has been minimized.

Copy link

commented Oct 23, 2018

Huh, why are you getting the board output in GTP mode?

It seems gtp2ogs is sending showboard:

Oct 23 20:00:16 gtp2ogs.js:353 [14116] >>> showboard

For some most commands (in GTP mode) I need to press enter twice to get any response from Leela, e.g. in particular showboard and genmove. Exempt are name, protocol_version, the new lz- commands. I have no idea why genmove still works in Sabaki with this behavior. 😅

Also funny: If I input showboard\nname\n, I only get the response from showboard; name is completely ignored. But if I input genmove b\nname\n, I get ? syntax error.

@arondes

This comment has been minimized.

Copy link

commented Oct 23, 2018

For 40B network, new leelaz is about 50% faster! It is an impressive improvement.

@Borthralla

This comment has been minimized.

Copy link

commented Oct 23, 2018

Maybe I'm stating the obvious, but does entering twice work? Another thing to check is if you didn't accidentally click in the window and enable text selection (this locks up Windows consoles).
If that doesn't work I sure hope I can reproduce this on my Windows machine...

I tried entering it twice, didn't seem to do anything. I performed the exact same actions with the .15 release and it behaved as expected, so I don't think its an issue with the console.

Turns out I just needed to press enter twice instead of re-typing the command 😅
Seems to work fine when I press enter twice. Thanks @yishn !

@jillybob

This comment has been minimized.

Copy link

commented Oct 23, 2018

I have loaded up the same setup in Sabaki and it works no problem. I'm guessing some change makes it incompatible with GTP2OGS


LZ40b> name 
Leela Zero
LZ40b> version 
= 0.15
○ LZ40b> protocol_version 
= 2
○ LZ40b> list_commands 
= protocol_version
name
version
quit
known_command
list_commands
boardsize
clear_board
komi
play
genmove
showboard
undo
final_score
final_status_list
time_settings
time_left
fixed_handicap
place_free_handicap
set_free_handicap
loadsgf
printsgf
kgs-genmove_cleanup
kgs-time_settings
kgs-game_over
heatmap
lz-analyze
lz-genmove_analyze
lz-memory_report
lz-setoption
○ LZ40b> komi 7.5
= 
○ LZ40b> boardsize 19
= 
○ LZ40b> clear_board 
= 
○ LZ40b> genmove B
= Q4
Using 4 thread(s).
RNG seed: 11063861880803315858
BLAS Core: Haswell
Detecting residual layers...v1...256 channels...40 blocks.
Initializing OpenCL (autodetecting precision).
Detected 1 OpenCL platforms.
Platform version: OpenCL 1.2 CUDA 9.2.217
Platform profile: FULL_PROFILE
Platform name:    NVIDIA CUDA
Platform vendor:  NVIDIA Corporation
Device ID:     0
Device name:   GeForce GTX 970
Device type:   GPU
Device vendor: NVIDIA Corporation
Device driver: 399.07
Device speed:  1177 MHz
Device cores:  13 CU
Device score:  1112
Selected platform: NVIDIA CUDA
Selected device: GeForce GTX 970
with OpenCL 1.2 capability.
Half precision compute support: No.
Detected 1 OpenCL platforms.
Platform version: OpenCL 1.2 CUDA 9.2.217
Platform profile: FULL_PROFILE
Platform name:    NVIDIA CUDA
Platform vendor:  NVIDIA Corporation
Device ID:     0
Device name:   GeForce GTX 970
Device type:   GPU
Device vendor: NVIDIA Corporation
Device driver: 399.07
Device speed:  1177 MHz
Device cores:  13 CU
Device score:  1112
Selected platform: NVIDIA CUDA
Selected device: GeForce GTX 970
with OpenCL 1.2 capability.
Half precision compute support: No.
Loaded existing SGEMM tuning.
Wavefront/Warp size: 32
Max workgroup size: 1024
Max workgroup dimensions: 1024 1024 64 
Started OpenCL SGEMM tuner.
Will try 290 valid configurations.
(1/290) KWG=16 KWI=2 MDIMA=8 MDIMC=8 MWG=16 NDIMB=8 NDIMC=8 NWG=16 SA=1 SB=1 STRM=0 STRN=0 VWM=2 VWN=2 0.1704 ms (692.5 GFLOPS)
(2/290) KWG=16 KWI=2 MDIMA=8 MDIMC=8 MWG=32 NDIMB=8 NDIMC=8 NWG=16 SA=1 SB=1 STRM=0 STRN=0 VWM=2 VWN=2 0.1325 ms (890.2 GFLOPS)
(3/290) KWG=16 KWI=2 MDIMA=8 MDIMC=8 MWG=64 NDIMB=8 NDIMC=8 NWG=16 SA=1 SB=1 STRM=0 STRN=0 VWM=2 VWN=2 0.1314 ms (898.1 GFLOPS)
(5/290) KWG=16 KWI=2 MDIMA=8 MDIMC=8 MWG=32 NDIMB=8 NDIMC=8 NWG=32 SA=1 SB=1 STRM=0 STRN=0 VWM=2 VWN=2 0.1172 ms (1006.9 GFLOPS)
(6/290) KWG=16 KWI=2 MDIMA=8 MDIMC=8 MWG=64 NDIMB=8 NDIMC=8 NWG=32 SA=1 SB=1 STRM=0 STRN=0 VWM=2 VWN=2 0.1044 ms (1130.1 GFLOPS)
(67/290) KWG=16 KWI=8 MDIMA=8 MDIMC=8 MWG=64 NDIMB=8 NDIMC=8 NWG=32 SA=1 SB=1 STRM=0 STRN=0 VWM=2 VWN=2 0.1024 ms (1151.6 GFLOPS)
(126/290) KWG=16 KWI=2 MDIMA=8 MDIMC=8 MWG=64 NDIMB=8 NDIMC=8 NWG=32 SA=1 SB=1 STRM=0 STRN=0 VWM=4 VWN=2 0.0962 ms (1225.6 GFLOPS)
(159/290) KWG=16 KWI=8 MDIMA=8 MDIMC=8 MWG=64 NDIMB=8 NDIMC=8 NWG=32 SA=1 SB=1 STRM=0 STRN=0 VWM=4 VWN=2 0.0941 ms (1253.1 GFLOPS)
(224/290) KWG=16 KWI=8 MDIMA=8 MDIMC=8 MWG=64 NDIMB=8 NDIMC=8 NWG=32 SA=1 SB=1 STRM=0 STRN=0 VWM=2 VWN=4 0.0940 ms (1255.1 GFLOPS)
Wavefront/Warp size: 32
Max workgroup size: 1024
Max workgroup dimensions: 1024 1024 64 
Using OpenCL half precision (at least 5% faster than single).
Setting max tree size to 3736 MiB and cache size to 415 MiB.
Thinking at most 36.0 seconds...
NN eval=0.462006
Playouts: 1255, Win: 46.83%, PV: D16 Q4 Q16 D4 R6 R14 C6 C14 O3 Q6 Q7
31.6s left, stopping early.
  Q4 ->     646 (V: 46.81%) (N: 19.10%) PV: Q4 D16 D4 Q16 R14 R6 C14 C6 F17 D14 D13 E14 C16 C17 C15 D17 J17 D11
 Q16 ->     641 (V: 46.82%) (N: 18.93%) PV: Q16 D4 D16 Q4 C6 C14 R6 R14 O3 Q6 Q7 P6 R4 R3 R5 Q3 L3 Q9
  D4 ->     637 (V: 46.82%) (N: 19.52%) PV: D4 Q16 Q4 D16 C14 C6 R14 R6 F17 D14 D13 E14 C16 C17 C15 D17 J17
 D16 ->     633 (V: 46.82%) (N: 19.28%) PV: D16 Q4 Q16 D4 R6 R14 C6 C14 O3 Q6 Q7 P6 R4 R3 R5 Q3 L3 Q9
  Q3 ->      57 (V: 46.80%) (N:  2.12%) PV: Q3 Q16 D4 D16 Q6 C6 C8 F3 F4 G4 F5 C3 E3 G3
 D17 ->      56 (V: 46.80%) (N:  2.09%) PV: D17 D4 Q16 Q4 D14 R14 R12 O17 O16 N16 O15 R17 P17 N17
 C16 ->      56 (V: 46.80%) (N:  2.09%) PV: C16 Q16 D4 Q4 F16 F3 H3 C6 D6 D7 E6 C3 C5 C7
  D3 ->      56 (V: 46.79%) (N:  2.10%) PV: D3 D16 Q4 Q16 D6 R6 R8 O3 O4 N4 O5 R3 P3 N3
 R16 ->      56 (V: 46.79%) (N:  2.09%) PV: R16 D16 Q4 D4 O16 O3 M3 R6 Q6 Q7 P6 R3 R5 R7
 Q17 ->      56 (V: 46.78%) (N:  2.09%) PV: Q17 Q4 D16 D4 Q14 C14 C12 F17 F16 G16 F15 C17 E17 G17
  R4 ->      56 (V: 46.76%) (N:  2.11%) PV: R4 D4 Q16 D16 O4 O17 M17 R14 Q14 Q13 P14 R17 R15 R13
  C4 ->      55 (V: 46.77%) (N:  2.07%) PV: C4 Q4 D16 Q16 F4 F17 H17 C14 D14 D13 E14 C17 C15 C13
10.6 average depth, 19 max depth
2349 non leaf nodes, 1.28 average children
3006 visits, 1059257 nodes, 3005 playouts, 680 n/s
@Vargooo

This comment has been minimized.

Copy link

commented Oct 24, 2018

It doesn't work with gogui-twogtp V.1.4.10

java.exe -jar lib\gogui-twogtp.jar -black "C:\Users\jm\Desktop\gogui1410\lz015999\leelaz.exe --gtp --weights=C:\Users\jm\Desktop\gogui1410\lz015999\184.gz --noponder" -white "C:\Users\jm\Desktop\gogui1410\lz015999\leelaz.exe --gtp --weights=C:\Users\jm\Desktop\gogui1410\lz015999\elf2.txt --noponder" -time 5 -games 5 -verbose -sgffile 184_elf2 -auto -komi 7.5

this worked with LZ0.15, but not with v 0.15.999

What I get is some ititialization that seems OK, and finally "version", and nothing more, but no freezing, it's like twogtp had finished its task.

twogtp1410

@jillybob

This comment has been minimized.

Copy link

commented Oct 24, 2018

So it doesn't work with GTP2OGS nor gogui. We have similar experiences, it doesn't freeze. it just stops working.

@arondes

This comment has been minimized.

Copy link

commented Oct 24, 2018

This is my test result.

block filter old n/s new n/s improvement
40 256 229 322 40.61%
20 256 385 516 34.03%
20 224 528 586 10.98%
15 192 733 806 9.96%
10 128 1835 1804 -1.69%
6 128 3113 2720 -12.62%
5 64 5836 5959 2.11%
@gcp

This comment has been minimized.

Copy link
Member Author

commented Oct 26, 2018

What I get is some ititialization that seems OK, and finally "version", and nothing more, but no freezing, it's like twogtp had finished its task.

It's strange the responses for version and name are inverset in that output.

This whole thing reeks a bit of a problem with buffering stdio somewhere.

@yishn

This comment has been minimized.

Copy link

commented Oct 26, 2018

What I get is some ititialization that seems OK, and finally "version", and nothing more, but no freezing, it's like twogtp had finished its task.

Running Leela Zero in GTP mode and entering version Enter manually doesn't do anything either, I have to press Enter twice. Maybe the twogtp controller only sends Enter once and is waiting (in vain) for an answer.

@nerai

This comment has been minimized.

Copy link
Contributor

commented Oct 26, 2018

https://sjeng.org/ftp/work/leela-zero-0.15.999-win64.zip

@gcp Sorry if I missed it, what code/branch/tag is this based on?

@Vargooo

This comment has been minimized.

Copy link

commented Oct 27, 2018

Maybe the twogtp controller only sends Enter once and is waiting (in vain) for an answer.

Thank you for your answer. I've tried several things, re-entering Enter :o) or launching several instances, but it always ends up with "version"
It's the same with twogtp V1.4.9, (laptop or desktop, it's always the same)

version

@gcp

This comment has been minimized.

Copy link
Member Author

commented Oct 29, 2018

Sorry if I missed it, what code/branch/tag is this based on?

Current next. I will merge this to master as soon as this issue is fixed.

I can reproduce it on a windows machine. Entering twice makes it work, so I think there is some buffering issue with windows console.

@gcp

This comment has been minimized.

Copy link
Member Author

commented Oct 29, 2018

New binaries:

https://sjeng.org/ftp/work/leela-zero-0.15.9999-win64.zip

These should fix the buffering issue. I also removed gzip and some supporting DLL that are no longer needed.

@barrtgt

This comment has been minimized.

Copy link

commented Oct 29, 2018

Gogui gui works fine now and pressing enter twice is no longer necessary. However, after running leela without the playouts command in command line and a genmove, leela will be stuck with the last two lines:

= R4

NN eval=0.520300

Leela will be loading the cpu and gpu until the window is resized horizontally or minimized and restored, then the program will return to normal showing the rest of the lines and waiting for another command.
The -g option bug still persists.
Already posted this in another thread, but I'll post this minor bug in case it was missed:
There seems to be an issue with saving sgf files in a folder in combination with the binary commands. For instance, if the -k option is made to the left of the binary commands, a folder is created, but no sgf files will be. But if you put the -k option to the right of the binary commands, no folder is created, but sgf files will be created in the root directory.

@Vargooo

This comment has been minimized.

Copy link

commented Oct 30, 2018

V 0.15.9999 works fine now with TWOGTP

Thank you very much for this beautiful project and for your help ! 👍

@jillybob

This comment has been minimized.

Copy link

commented Oct 30, 2018

V 0.15.9999 works with GTP2OGS with some conditions.

  1. It doesn't play a move when time is auto-managed. When using the settings below (I've tried KGS time, it doesn't work) The program searches for nodes but doesn't play a move. I know it's searching because GPU monitor shows increased GPU workloads. A workaround I've managed to do successfully is to limit search by visits. -v 400 for example, it successfully stops itself at 400 visits and continues playing. For some reason Leelaz doesn't stop itself to play a move in auto time-manage mode.

  2. In the previous version, leela would send print-outs for the moves it's considering every 1 second. Now however, shows the NN net evaluation (like you can see here below), hits the visit limit, and then prints everything out in one go, instead of "live" debug outputs. In Sabaki, it prints out "live" evaluations . Minor issue compared to issue 1.

Here is the output


C:\Users\jillybob>gtp2ogs --botid AquaBotTest --apikey REDACTED -d --persist -- C:\Users\jillybob\Desktop\leela-zero-0.15.9999-win64\leelaz.exe --gpu 0 -w C:\Users\jillybob\Desktop\leela-zero-0.15.9999-win64\network.gz --gtp --lagbuffer 200 --logfile log.txt -r 1 -t 4
Oct 30 16:15:38   gtp2ogs.js:1822                #  Connecting to https://online-go.com:443
Oct 30 16:15:39   gtp2ogs.js:1822                #  Connected
Oct 30 16:15:39   gtp2ogs.js:1822                #  Bot is user id: 495496
Oct 30 16:15:40   gtp2ogs.js:1822                #  active_game: {"id":15065382,"phase":"play","name":"Friendly Match","player_to_move":495496,"width":19,"height":19,"move_number":1,"paused":0,"private":false,"black":{"username":"RoyalZero","id":484523,"rank":27,"professional":false,"accepted":false,"ratings":{"overall":{"deviation":90.38723544299218,"rating":2740.468834650913,"volatility":0.06022278643444537,"games_played":5488}}},"white":{"username":"AquaBotTest","id":495496,"rank":21,"professional":false,"accepted":false,"ratings":{"overall":{"deviation":84.7684390612251,"rating":2545.7768553302017,"volatility":0.060205592343747685,"games_played":49}}},"time_per_move":10}
Oct 30 16:15:40   gtp2ogs.js:1028                [Game 15065382]            Connecting to game.
Oct 30 16:15:40   gtp2ogs.js:1028                [Game 15065382]  Move 1    gamedata     W   RoyalZero  [19x19]
Oct 30 16:15:40   gtp2ogs.js:1028                [Game 15065382]  Move 1    makeMove 1 is 1 != 1 ?
Oct 30 16:15:40   gtp2ogs.js:353                 [24508] Starting  C:\Users\jillybob\Desktop\leela-zero-0.15.9999-win64\leelaz.exe --gpu 0 -w C:\Users\jillybob\Desktop\leela-zero-0.15.9999-win64\network.gz --gtp --lagbuffer 200 --logfile log.txt -r 1 -t 4
Oct 30 16:15:40   gtp2ogs.js:1028                [Game 15065382]  Move 1    Starting new bot process [24508]
Oct 30 16:15:40   gtp2ogs.js:1028                [Game 15065382]  Move 1    State loading for new bot
Oct 30 16:15:40   gtp2ogs.js:353                 [24508] >>> boardsize 19
Oct 30 16:15:40   gtp2ogs.js:353                 [24508] >>> clear_board
Oct 30 16:15:40   gtp2ogs.js:353                 [24508] >>> komi 7.5
Oct 30 16:15:40   gtp2ogs.js:353                 [24508] >>> play black d4
Oct 30 16:15:40   gtp2ogs.js:353                 [24508] >>> showboard
Oct 30 16:15:40   gtp2ogs.js:353                 [24508] Generating move for game 15065382
Oct 30 16:15:40   gtp2ogs.js:1028                [Game 15065382]  Move 1    genmove w
Oct 30 16:15:40   gtp2ogs.js:353                 [24508] >>> kgs-time_settings byoyomi 30 3 5
Oct 30 16:15:40   gtp2ogs.js:353                 [24508] >>> time_left black 30 0
Oct 30 16:15:40   gtp2ogs.js:353                 [24508] >>> time_left white 23 0
Oct 30 16:15:40   gtp2ogs.js:353                 [24508] >>> genmove white
Oct 30 16:15:40   gtp2ogs.js:1028                [Game 15065382]  Move 1    clock: {"game_id":15065382,"current_player":495496,"black_player_id":484523,"white_player_id":495496,"title":"Friendly Match","last_move":1540876533695,"expiration":1540876613695,"black_time":{"thinking_time":30,"periods":5,"period_time":10},"white_time":{"thinking_time":30,"periods":5,"period_time":10},"now":1540876539775}
Oct 30 16:15:40 !!!!! gtp2ogs.js:361                 [24508] stderr: Logging to log.txt.
Using 4 thread(s).
RNG seed: 14201665767697305486
Using per-move time margin of 2.00s.

Oct 30 16:15:40 !!!!! gtp2ogs.js:361                 [24508] stderr: BLAS Core: Haswell

Oct 30 16:15:41 !!!!! gtp2ogs.js:361                 [24508] stderr: Detecting resi
Oct 30 16:15:41 !!!!! gtp2ogs.js:361                 [24508] stderr: dual layers...v1...
Oct 30 16:15:41 !!!!! gtp2ogs.js:361                 [24508] stderr: 256 channels...
Oct 30 16:15:43 !!!!! gtp2ogs.js:361                 [24508] stderr: 40 blocks.

Oct 30 16:15:46 !!!!! gtp2ogs.js:361                 [24508] stderr: Initializing Ope
Oct 30 16:15:46 !!!!! gtp2ogs.js:361                 [24508] stderr: nCL (autodetecting precision).

Oct 30 16:15:46 !!!!! gtp2ogs.js:361                 [24508] stderr: Detected 1 OpenC
Oct 30 16:15:46 !!!!! gtp2ogs.js:361                 [24508] stderr: L platforms.
Platform version: OpenCL 1.2 CUDA 10.0.132
Platform profile: FULL_PROFILE
Platform name:    NVIDIA CUDA
Platform vendor:  NVIDIA Corporation
Device ID:     0
Device name:   GeForce GTX 970
Device type:   GPU
Device vendor: NVIDIA Corporation
Device driver: 416.34
Device speed:  1177 MHz
Device cores:  13 CU
Device score:  1112
Selected platform: NVIDIA CUDA
Selected device: GeForce GTX 970
with Op
Oct 30 16:15:46 !!!!! gtp2ogs.js:361                 [24508] stderr: enCL 1.2 capability.

Oct 30 16:15:46 !!!!! gtp2ogs.js:361                 [24508] stderr: Half precision compute support:
Oct 30 16:15:46 !!!!! gtp2ogs.js:361                 [24508] stderr: No.
Detected 1 OpenCL platforms.
Platform version: OpenCL 1.2 CUDA 10.0.132
Platform profile: FULL_PROFILE
Platform name:    NVIDIA CUDA
Platform vendor:  NVIDIA Corporation
Device ID:     0
Device name:   GeForce GTX 970
Device type:   GPU
Device vendor: NVIDIA Corporation
Device driver: 416.34
Device speed:  1177 MHz
Device cores:  13 CU
Device score:  1112
Selected platform: NVIDIA CUDA
Selected device: GeForce GTX 970
with OpenCL 1.2 capability.

Oct 30 16:15:46 !!!!! gtp2ogs.js:361                 [24508] stderr: Half precisio
Oct 30 16:15:46 !!!!! gtp2ogs.js:361                 [24508] stderr: n compute support: No.
Loaded existing SGEMM tuning.

Oct 30 16:15:46 !!!!! gtp2ogs.js:361                 [24508] stderr: Wavefront/War
Oct 30 16:15:46 !!!!! gtp2ogs.js:361                 [24508] stderr: p size: 32
Max workgroup size: 1024
Max workgroup dimensions: 1024 1024 64

Oct 30 16:15:48 !!!!! gtp2ogs.js:361                 [24508] stderr: Loaded existing
Oct 30 16:15:48 !!!!! gtp2ogs.js:361                 [24508] stderr: SGEMM tuning.
Wavefront/Warp size: 32
Max workgroup size: 1024
Max workgroup dimensions: 1024 1024 64

Oct 30 16:15:50 !!!!! gtp2ogs.js:361                 [24508] stderr: Using OpenCL h
Oct 30 16:15:50   gtp2ogs.js:353                 [24508] <<< =


Oct 30 16:15:50 !!!!! gtp2ogs.js:361                 [24508] stderr: alf precision (at least 5% faster than single).
Setting max tree size to 3736 MiB and cache size to 415 MiB.

Oct 30 16:15:50   gtp2ogs.js:353                 [24508] <<< =


Oct 30 16:15:51   gtp2ogs.js:353                 [24508] <<< =


Oct 30 16:15:51   gtp2ogs.js:353                 [24508] <<< =


Oct 30 16:15:51   gtp2ogs.js:353                 [24508] <<< =


Oct 30 16:15:51   gtp2ogs.js:1028                [Game 15065382]  Move 1    State loaded for new bot
Oct 30 16:15:51 !!!!! gtp2ogs.js:361                 [24508] stderr:
Passes: 0            Black (X) Prisoners: 0
White (O) to move    White (O) Prisoners: 0

   a b c d e f g h j k l m n o p q r s t
19 . . . . . . . . . . . . . . . . . . . 19
18 . . . . . . . . . . . . . . . . . . . 18
17 . . . . . . . . . . . . . . . . . . . 17
16 . . . + . . . . . + . . . . . + . . . 16
15 . . . . . . . . . . . . . . . . . . . 15
14 . . . . . . . . . . . . . . . . . . . 14
13 . . . . . . . . . . . . . . . . . . . 13
12 . . . . . . . . . . . . . . . . . . . 12
11 . . . . . . . . . . . . . . . . . . . 11
10 . . . + . . . . . + . . . . . + . . . 10
 9 . . . . . . . . . . . . . . . . . . .  9
 8 . . . . . . . . . . . . . . . . . . .  8
 7 . . . . . . . . . . . . . . . . . . .  7
 6 . . . . . . . . . . . . . . . . . . .  6
 5 . . . . . . . . . . . . . . . . . . .  5
 4 . . .(X). . . . . + . . . . . + . . .  4
 3 . . . . . . .
Oct 30 16:15:51 !!!!! gtp2ogs.js:361                 [24508] stderr:  . . . . . . . . . . . .  3
 2 . . . . . . . . . . . . . . . . . . .  2
 1 . . . . . . . . . . . . . . . . . . .  1
   a b c d e f g h j k l m n o p q r s t

Hash: 68CF0A4D4D44DA46 Ko-Hash: F8DD395FF18CB668

Black time: 01:00:00
White time: 01:00:00


Oct 30 16:15:51   gtp2ogs.js:353                 [24508] <<< =


Oct 30 16:15:51   gtp2ogs.js:353                 [24508] <<< =


Oct 30 16:15:51 !!!!! gtp2ogs.js:361                 [24508] stderr: NN eval=0.53
Oct 30 16:15:51 !!!!! gtp2ogs.js:361                 [24508] stderr: 2775
`
@wonderingabout

This comment has been minimized.

Copy link
Contributor

commented Oct 30, 2018

we're almost there 👍

@gcp

This comment has been minimized.

Copy link
Member Author

commented Oct 30, 2018

Leela will be loading the cpu and gpu until the window is resized horizontally or minimized and restored, then the program will return to normal showing the rest of the lines and waiting for another command.

Isn't it just pondering? After it has reported its move, it will start thinking on the opponents' turn.

In the previous version, leela would send print-outs for the moves it's considering every 1 second. Now however, shows the NN net evaluation (like you can see here below), hits the visit limit, and then prints everything out in one go, instead of "live" debug outputs. In Sabaki, it prints out "live" evaluations . Minor issue compared to issue 1.

Isn't this exactly the same problem? You're just looking at the engine while it's pondering. That would also explain why it doesn't respect the time and just keeps on going until it hits the visit limit.

The only relevant change I believe is that there was a fix to make it print the NN eval at the start of pondering too.

@gcp

This comment has been minimized.

Copy link
Member Author

commented Oct 30, 2018

I tried the command sequence that GTP2OGS uses above and indeed the output you are seeing is just the engine pondering. It will stop pondering to process the time_left commands (causing a dump of the search statistics) and restart immediately after that (small nuance here: the engine wouldn't ponder at the start of the game normally, but GTP2OGS actually makes it kick in by sending those time_left commands). After it is sent the genmove white command you will get the thinking output:

Thinking at most 2.3 seconds...
NN eval=0.532757
1.3s left, stopping early.

 Q16 ->    7569 (V: 53.06%) (N: 82.24%) PV: Q16 R17 Q17 R16 Q15 R15 Q14 S13 Q4 D16 F3 L17 D17 E17 C17 E16 B15 C14 C15 D14 E18 F18 D18 Q12 G17 G14 M15 L15
  Q4 ->     678 (V: 53.11%) (N:  6.68%) PV: Q4 Q16 D16 C17 D17 C16 D15 C15 C13 C14 D14 B13 C12 B12 O17 K17 M17 G17 C11 P17 O16 Q14 K15 H15 C18 B18
 D16 ->     652 (V: 53.11%) (N:  6.48%) PV: D16 Q16 Q4 C17 D17 C16 D15 C15 C13 C14 D14 B13 C12 B12 O17 K17 M17 G17 C11 P17 O16 Q14 K15 H15 C18
16.4 average depth, 36 max depth
7190 non leaf nodes, 1.24 average children
8900 visits, 3074847 nodes, 574 playouts, 557 n/s

= Q16

Your log appears to be truncated before it gets to that part. (The tell-tale sign that this is the real move processing and not pondering is the "Thinking at most x.x seconds" line)

As far as I can tell the engine is behaving correctly here. I don't know what would cause the (1) in your report, as your log stops before the engine ever starts thinking about the move.

Maybe @roy7 can comment if he sees an issue here.

@roy7

This comment has been minimized.

Copy link
Collaborator

commented Oct 30, 2018

@jillybob Are you on Windows? I did have an issue with Windows and pondering long ago with the original Leela, where if commands were all sent too quickly in a batch Leela somehow didn't see some of them during the moment it took to turn pondering on, so it'd ponder forever and never flush the rest of stdin. I eventually gave up trying to work around it (it seemed adding a 150ms delay between commands solved it but that isn't easy to do in nodejs/gtp2ogs) and just moved to Linux instead.

@roy7

This comment has been minimized.

Copy link
Collaborator

commented Oct 30, 2018

I went back to my old emails and the issue I had on Windows before boiled down to this. If I send a time_left and a genmove at the same time (which is what gtp2ogs does), there was a ~150ms window that Leela would never see the genmove command while pondering restarts after the time_left command is processed by Leela. Thus real search time never starts, it just ponders forever, since the bot is waiting for a genmove command but gtp2ogs already sent it and thus it is waiting for the bot to move.

A work around I came up with that worked was forcing a 250ms delay between time_left and genmove in gtp2ogs but that's such a hack and nodejs isn't good about forcing blocking delays since it's so async. This only happened on Windows, not on Linux.

We never did isolate if the issue was nodejs or Leela since I didn't have a way to test/reproduce on a Windows command line and the KGS interface doesn't act this way that I know of.

If this is still a problem in Leela Zero / Windows / pondering / gtp2ogs, maybe someone can actually isolate the problem. :) (If a time_left is done immediately before every genmove in the KGS bridge or in gogui or something, does that also cause problems under Windows, would be one test?)

@gcp

This comment has been minimized.

Copy link
Member Author

commented Oct 31, 2018

Based on this, I'll make the v0.16 release.

@gcp gcp closed this Oct 31, 2018

@roy7

This comment has been minimized.

Copy link
Collaborator

commented Oct 31, 2018

I hope eventually someone will solve the issue, if it's still an actual problem with too many commands coming in too quickly when using Windows + pondering. :)

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
You can’t perform that action at this time.