pytorch / ELF

ELF: a platform for game research with AlphaGoZero/AlphaZero reimplementation
Other
3.37k stars 566 forks source link

the model with windows binary is not the final version #134

Open l1t1 opened 5 years ago

l1t1 commented 5 years ago

https://facebook.ai/developers/tools/elf-opengo says Play a game with ELF OpenGo ELF OpenGo public binary, allowing anyone with a CUDA-enabled GPU to play against the final ELF OpenGo model.

but according to the test of https://github.com/pytorch/ELF/issues/133#issuecomment-463867827

the model in the package is elfv0 in fact.

yuandong-tian commented 5 years ago

@zchen0211 who is in charge of windows executable.

l1t1 commented 5 years ago

it would be better if windows version can specify models. btw i didnt saw --load option in df_console.exe -h

l1t1 commented 5 years ago

https://www.lifein19x19.com/viewtopic.php?p=241852#p241852 can you post the changlog

zchen0211 commented 5 years ago

Hi @l1t1 I have updated with latest version, please download the latest one on our dataset, and let me know with any problems.

l1t1 commented 5 years ago

ok, do you have the changlog of this version and that released at feb 21(https://github.com/pytorch/ELF/issues/138#issuecomment-465817975)?

l1t1 commented 5 years ago

the cpu version now works in cmd console

D:\elfv2\play_opengo_v2\elf_cpu>df_console.exe --mcts_rollout_per_thread 80
genmove b
= B2

genmove w
= L2

but failed in sabaki

○ elfv2_1> name 
connection failed
○ elfv2_1> version 
connection failed
○ elfv2_1> protocol_version 
connection failed
○ elfv2_1> list_commands 
connection failed
○ elfv2_1> komi 6.5
connection failed
○ elfv2_1> komi 6.5
connection failed
l1t1 commented 5 years ago

sorry , i didnt do as readme said, after add --load model-v2.bin --mcts_rollout_per_thread 80 to argument, now it works in sabaki

○ elfv2_1> name 
= DF2
○ elfv2_1> version 
= 1.0
○ elfv2_1> protocol_version 
= 2
○ elfv2_1> list_commands 
= boardsize
clear_board
exit
final_score
genmove
komi
list_commands
name
play
protocol_version
quit
showboard
version
○ elfv2_1> komi 6.5
= 
○ elfv2_1> boardsize 19
= 
○ elfv2_1> clear_board 
= 
○ elfv2_1> genmove B
= D16
l1t1 commented 5 years ago

the gogui command seems also work now

D:\tool\go_gui\gogui-twogtp -black "D:\elfv2\play_opengo_v2\elf_cpu\df_console.exe --load d:/1500000.bin --mcts_rollout_per_thread 80" -white "D:\leela-zero-0.16-win64\leelaz.exe --gtp -w D:\elfv2.gz --noponder -v 3201"  -games 10 -sgffile fb_elfv2 -auto -komi 7.5
l1t1 commented 5 years ago

how to print usage?

D:\elfv2\play_opengo_v2\elf_cpu>df_console.exe -h
Traceback (most recent call last):
  File "df_console.py", line 92, in <module>
  File "df_console.py", line 45, in main
  File "rlpytorch\model_loader.py", line 268, in load_env
  File "elf\options\py_option_spec.py", line 58, in parse
  File "argparse.py", line 1730, in parse_args
  File "argparse.py", line 1762, in parse_known_args
  File "argparse.py", line 1968, in _parse_known_args
  File "argparse.py", line 1908, in consume_optional
  File "argparse.py", line 1836, in take_action
  File "argparse.py", line 1020, in __call__
  File "argparse.py", line 2362, in print_help
  File "argparse.py", line 2346, in format_help
  File "argparse.py", line 282, in format_help
  File "argparse.py", line 213, in format_help
  File "argparse.py", line 213, in <listcomp>
  File "argparse.py", line 213, in format_help
  File "argparse.py", line 213, in <listcomp>
  File "argparse.py", line 519, in _format_action
  File "argparse.py", line 606, in _expand_help
ValueError: unsupported format character 'w' (0x77) at index 106
[5572] Failed to execute script df_console

D:\elfv2\play_opengo_v2\elf_cpu>df_console.exe
list_commands
= boardsize
clear_board
exit
final_score
genmove
komi
list_commands
name
play
protocol_version
quit
showboard
version

quit
[2019-02-25 10:56:08.285] [elf::base::Context-3] [info] Prepare to stop ...
[2019-02-25 10:56:08.285] [elfgames::go::GoGameSelfPlay-0-15] [warning] Invalid move: x = 0 y = 1 move: ab pl
ain
[2019-02-25 10:56:08.299] [elfgames::go::mcts::MCTSActor-21] [error] model version 1 and required version 0 a
istent

D:\elfv2\play_opengo_v2\elf_cpu>df_console.exe  --help
Traceback (most recent call last):
  File "df_console.py", line 92, in <module>
  File "df_console.py", line 45, in main
  File "rlpytorch\model_loader.py", line 268, in load_env
  File "elf\options\py_option_spec.py", line 58, in parse
  File "argparse.py", line 1730, in parse_args
  File "argparse.py", line 1762, in parse_known_args
  File "argparse.py", line 1968, in _parse_known_args
  File "argparse.py", line 1908, in consume_optional
  File "argparse.py", line 1836, in take_action
  File "argparse.py", line 1020, in __call__
  File "argparse.py", line 2362, in print_help
  File "argparse.py", line 2346, in format_help
  File "argparse.py", line 282, in format_help
  File "argparse.py", line 213, in format_help
  File "argparse.py", line 213, in <listcomp>
  File "argparse.py", line 213, in format_help
  File "argparse.py", line 213, in <listcomp>
  File "argparse.py", line 519, in _format_action
  File "argparse.py", line 606, in _expand_help
ValueError: unsupported format character 'w' (0x77) at index 106
[2996] Failed to execute script df_console
zchen0211 commented 5 years ago

Hi @l1t1 let me double check if we have a full-description of all "--" arguments. Great thanks for your suggestion! Working on it right now.