Lc0 v0.30.0 wurde veröffentlicht

By Lothar Jung Date 2023-07-21 19:53 Edited 2023-07-21 20:02 Upvotes 1

Hier die Bekanntmachung:

We are pleased to announce that Lc0 v0.30.0 is out.
You can get it from <https://github.com/LeelaChessZero/lc0/releases/tag/v0.30.0>.

The main features of this release are:

* Support for networks with attention body and smolgen added to blas, cuda, metal and onnx backends.

* WDL conversion for more realistic WDL score and contempt. Adds an Elo based WDL transformation of the NN value head output. Helps with more accurate play at high level (WDL sharpening), more aggressive play against weaker opponents and draw avoiding openings (contempt), piece odds play.

For details on how it works see <https://lczero.org/blog/2023/07/the-lc0-v0.30.0-wdl-rescale/contempt-implementation/>.

New since rc2 are:

* `WDL_mu` score type is now the default and the `--moves-left-threshold` default was changed from 0 to 0.8.

* Fixes for contempt with infinite search/pondering and for the wdl display when pondering.

Anmerkungen:

WDL = WinDrawLost
Die Unterstützung von aktuellen Netzen für die verschiedene Backends wurde erweitert. Wichtig für iGPUs.

By Lothar Jung Date 2023-07-22 09:22 Upvotes 1

Hier die Einzelheiten:

In this version:

Support for networks with attention body and smolgen added to blas, cuda, metal and onnx backends.
WDL conversion for more realistic WDL score and contempt. Adds an Elo based WDL transformation of the NN value head output. Helps with more accurate play at high level (WDL sharpening), more aggressive play against weaker opponents and draw avoiding openings (contempt), piece odds play. For details on how it works see https://lczero.org/blog/2023/07/the-lc0-v0.30.0-wdl-rescale/contempt-implementation/.
A new score type WDL_mu which follows the new eval convention, where +1.00 means 50% white win chance.
Changed mlh threshold effect to create a smooth transition.
WDL_mu score type is now the default and the --moves-left-threshold default was changed from 0 to 0.8.
Simplified to a single --draw-score parameter, adjusting the draw score from white's perspective: 0 gives standard scoring, -1 gives Armageddon scoring.
Updated describenet for new net architectures.
Added a first-move-bonus option to the legacy time manager, to accompany book-ply-bonus for shallow openings.
Persistent L2 cache optimization for the cuda backend. Use the cache_opt=true backend option to turn it on.
Some performance improvements for the cuda, onnx and blas backends.
Added the threads backend option to onnx, defaults to 0 (let the onnxruntime decide) except for onnx-cpu that defaults to 1.
The onnx-dml package now includes a directml.dll installation script.
Some users experienced memory issues with onnx-dml, so the defaults were changed. This may affect performance, in which case you can use the steps=8 backend option to get the old behavior.
The Python bindings are available as a package, see the README for instructions.
Revised 'simple' time manager.
A new spinlock implementation (selected with --search-spin-backoff) to help with many cpu threads (e.g. 128 threads), obviously for cpu backends only.
Fixes for contempt with infinite search/pondering and for the wdl display when pondering.
Some assorted fixes and code cleanups.

By Volker Göbel Date 2023-07-25 13:00

Lothar Jung schrieb:

Auf Discord/Lc0/event-discuss gibt es die Engineparameter zu den jeweiligen bineries.

Bringen die denn auch etwas, im vergleich zu default?

Lc0 T1-4000000
"Rating": 0,
"Dev": "by The LCZero Authors",
"LogoPath": "img/lc0.png",
"Protocol": "UCI",
"Path": "C:/LCZero/lczero_gpu_v10/CUDA/lc0-dag-cup/lc0.exe",
"NetworkPath": "C:/LCZero/lczero_gpu_v10/CUDA/lc0-dag-cup/t1-768x15x24h-swa-4000000.pb.gz",
"args":"--show-hidden",
"Options": {
"Threads": 2,
"Backend":"demux",
"Cpuct": 2.897,
"CpuctBase": 45669,
"CpuctFactor": 3.973,
"FpuValue": 0.98416,
"PolicyTemperature": 1.4061131420002002,
"SmartPruningFactor": 2,
"SmartPruningMinimumBatches": 300,
"LogLiveStats": "true",
"Taskworkers": 3,
"BackendOptions": "backend=cuda-fp16,(gpu=0),(gpu=1)multi_stream=true,threads=2",
"RootHasOwnCpuctParams": "false",
"MovesLeftSlope": 0.0027,
"MovesLeftThreshold": 0.8,
"Contempt": 0,
"WDLCalibrationElo": 3392,
"WDLContemptAttenuation": 0.6,
"WDLEvalObjectivity": 0,
"WDLDrawRateReference": 0.56,
"RamLimitMb": 0,
"MoveOverheadMs": 1000,
"MinibatchSize": 340,
"ScoreType": "WDL_mu",
"StrictTiming": "true",
"TimeManager": "legacy(book-ply-bonus=0.4)",
"VerboseMoveStats": "true",
"SyzygyPath": "C:/Syzygy",
"LogFile": "T1.txt",
"Ponder": "false",
"UCI_ShowWDL": "true"

lc0_r50_T1-4000000
"Rating": 0,
"Dev": "by The LCZero Authors",
"C:/LCZero/lczero_gpu_v10/CUDA/lc0-dag-New_adjust_042efe25/t1-768x15x24h-swa-4000000.pb.gz",
"args":"--show-hidden",
"Options": {
"Threads": 2,
"Backend":"demux",
"Cpuct": 2.7317680215626656,
"CpuctBase": 45669,
"CpuctFactor": 3.973,
"CpuctUtilityStdevPrior": 0.464189126225121,
"CpuctUtilityStdevScale": 0.5001351117120049,
"FpuValue": 0.6913510449351836,
"PolicyTemperature": 1.4061131420002002,
"SmartPruningFactor": 2,
"SmartPruningMinimumBatches": 300,
"LogLiveStats": "true",
"Taskworkers": 3,
"BackendOptions": "backend=cuda-fp16,(gpu=0),(gpu=1),multi_stream=true,threads=2",
"RootHasOwnCpuctParams": "false",
"MovesLeftSlope": 0.0027,
"MovesLeftThreshold": 0.8,
"Contempt": 0,
"WDLCalibrationElo": 3392,
"WDLContemptAttenuation": 0.6,
"WDLEvalObjectivity": 0,
"WDLDrawRateReference": 0.56,
"RamLimitMb": 0,
"MoveOverheadMs": 1000,
"MinibatchSize": 340,
"ScoreType": "WDL_mu",
"StrictTiming": "true",
"TimeManager": "legacy(book-ply-bonus=0.4)",
"VerboseMoveStats": "true",
"SyzygyPath": "C:/Syzygy",
"LogFile": "lc0_r50.txt",
"Ponder": "false",
"UCI_ShowWDL": "true"

By Lothar Jung Date 2023-07-25 13:06 Edited 2023-07-25 13:42

Gerade laufen Tests bei den 0.31.0 dev bineries.
Ich kann nicht über mehr berichten, als das was auf Discord thematisiert wird.