Code

Parallel Tuning#241

Code

Open

MiloLurati wants to merge 105 commits intoKernelTuner:masterKernelTuner/kernel_tuner:masterfrom MiloLurati:parallelTuningMiloLurati/kernel_tuner:parallelTuningCopy head branch name to clipboard

+871-40

Commits

Commits on Jan 22, 2024

• created remoteRunner class

  MiloLurati

  committed
  e827e75

  Copy full SHA for e827e75

Commits on Jan 23, 2024

• added remote actor class

  

  Milo Lurati

  committed
  deab579

  Copy full SHA for deab579

• update remote runner

  

  Milo Lurati

  committed
  52287b9

  Copy full SHA for 52287b9

• added remote_mode function argument to tune_kernel and related remote runner selection logic

  

  Milo Lurati

  committed
  40cc888

  Copy full SHA for 40cc888

• added parallel tuning test

  

  Milo Lurati

  committed
  b14aaf0

  Copy full SHA for b14aaf0

• added pool of actors

  

  Milo Lurati

  committed
  1a55a5c

  Copy full SHA for 1a55a5c

Commits on Jan 24, 2024

• clean up remote runner and actor

  

  Milo Lurati

  committed
  4fef594

  Copy full SHA for 4fef594

Commits on Jan 30, 2024

• updates on remote code

  

  Milo Lurati

  committed
  3f3b9e6

  Copy full SHA for 3f3b9e6

Commits on Apr 2, 2024

• changed naming from remote to parallel

  

  Milo Lurati

  committed
  fe5da39

  Copy full SHA for fe5da39

Commits on Apr 4, 2024

• added get_num_devices function

  

  Milo Lurati

  committed
  a43dc84

  Copy full SHA for a43dc84

• added ensemble and parallel runner related stuff

  

  Milo Lurati

  committed
  ab3aa24

  Copy full SHA for ab3aa24

• switched to new naming of parallel remote and some clean up

  

  Milo Lurati

  committed
  e8a7228

  Copy full SHA for e8a7228

• added class instances needed down the line in the execution of the ensemble strategy

  

  Milo Lurati

  committed
  3dd748c

  Copy full SHA for 3dd748c

• changed naming due to ensemble implementation, this was the original remote_actor

  

  Milo Lurati

  committed
  e743bec

  Copy full SHA for e743bec

• started ensemble implementation, very basic functionality works

  

  Milo Lurati

  committed
  df949d0

  Copy full SHA for df949d0

• updated tests

  

  Milo Lurati

  committed
  45a1747

  Copy full SHA for 45a1747

Commits on Apr 5, 2024

• clean up in parallel runner

  

  Milo Lurati

  committed
  5fb5927

  Copy full SHA for 5fb5927

• moved to sub directory ray

  

  Milo Lurati

  committed
  a96ef43

  Copy full SHA for a96ef43

• added subdirectory ray with all 3 actor classes

  

  Milo Lurati

  committed
  c831f5f

  Copy full SHA for c831f5f

• itegrated calls to cache manager functions when running in ensemble

  

  Milo Lurati

  committed
  0cc2a6e

  Copy full SHA for 0cc2a6e

• added cache manager logic

  

  Milo Lurati

  committed
  b816f3d

  Copy full SHA for b816f3d

Commits on Apr 8, 2024

• added instances needed for the ensemble down the line of execution

  

  Milo Lurati

  committed
  781839a

  Copy full SHA for 781839a

• added strategy option to get_options function

  

  Milo Lurati

  committed
  9f8d212

  Copy full SHA for 9f8d212

Commits on Apr 10, 2024

• added ignore_reinit_error to ray init

  

  Milo Lurati

  committed
  d08b5d4

  Copy full SHA for d08b5d4

• added ignore_reinit_error to ray init

  

  Milo Lurati

  committed
  903c981

  Copy full SHA for 903c981

• added cache manager to parallel tuning

  

  Milo Lurati

  committed
  1a2219a

  Copy full SHA for 1a2219a

Commits on Apr 11, 2024

• re-assign tuning options to final version from the cache manager at the end of the parallel runner execution

  

  Milo Lurati

  committed
  a476585

  Copy full SHA for a476585

• small bug fix in execute

  MiloLurati

  committed
  6233e09

  Copy full SHA for 6233e09

Commits on Apr 14, 2024

• Merge branch 'KernelTuner:master' into parallelTuning

  MiloLurati

  authored
  cde62ae

  Copy full SHA for cde62ae

• Merge pull request #2 from KernelTuner/simulation-searchspace-improvements

  Show description for 0722629
  MiloLurati

  authored
  0722629

  Copy full SHA for 0722629

Commits on Apr 16, 2024

• updates to run ensemble in simulation mode on CPUs

  MiloLurati

  committed
  14e5f0b

  Copy full SHA for 14e5f0b

Commits on Apr 23, 2024

• fixed problem with ray resources and stalling actors

  MiloLurati

  committed
  a963dac

  Copy full SHA for a963dac

Commits on Apr 25, 2024

• added setup_resources and new impl of costfunc (not yet tested and still have to deal with stop criterion)

  MiloLurati

  committed
  c55b870

  Copy full SHA for c55b870

• added ensemble and memetic to strategy map and import

  MiloLurati

  committed
  d8541a0

  Copy full SHA for d8541a0

• rearranged how parallel runner deals with cache manager and actor's lifecycle

  MiloLurati

  committed
  c755254

  Copy full SHA for c755254

• initial adaptions for memetic and cleaned up logic of ensemble

  MiloLurati

  committed
  a23ef94

  Copy full SHA for a23ef94

• returning tuning_options for memetic logic

  MiloLurati

  committed
  697ead0

  Copy full SHA for 697ead0

• init impl of memetic strategy

  MiloLurati

  committed
  b247ed0

  Copy full SHA for b247ed0

• initial adapion for memetic strategy

  MiloLurati

  committed
  948ab7f

  Copy full SHA for 948ab7f

• removed brute_force from strategy map and import

  MiloLurati

  committed
  9e40d4e

  Copy full SHA for 9e40d4e

• fixes of new costfunc and stop criterion is checked retrospectively

  MiloLurati

  committed
  3cb428d

  Copy full SHA for 3cb428d

Commits on Apr 29, 2024

• fixed bug with tuning options cache manager

  MiloLurati

  committed
  2d13fc3

  Copy full SHA for 2d13fc3

• fixed some bugs for memetic algo functioning

  MiloLurati

  committed
  1a2ba53

  Copy full SHA for 1a2ba53

• removed debug prints

  MiloLurati

  committed
  2aba6f5

  Copy full SHA for 2aba6f5

• fixed problem with single config input and final results data structure

  MiloLurati

  committed
  cd3f212

  Copy full SHA for cd3f212

• added progress prints of memetic algo and kill statement for cache manager

  MiloLurati

  committed
  af9bd5e

  Copy full SHA for af9bd5e

• sort results for retrospective stop criterion check

  MiloLurati

  committed
  d382f05

  Copy full SHA for d382f05

• added comments

  MiloLurati

  committed
  218b8ac

  Copy full SHA for 218b8ac

• updated returning results logic in _evaluate_configs()

  MiloLurati

  committed
  79b7a50

  Copy full SHA for 79b7a50

• added comments

  MiloLurati

  committed
  88f63b4

  Copy full SHA for 88f63b4

Commits on Apr 30, 2024

• updates to run more strategies then devices available

  MiloLurati

  committed
  a2afd1d

  Copy full SHA for a2afd1d

Commits on May 3, 2024

• returning last two lists of candidates for memetic algo

  MiloLurati

  committed
  d950b2d

  Copy full SHA for d950b2d

• returning last two candidates for memetic algo

  MiloLurati

  committed
  980777f

  Copy full SHA for 980777f

• returning last two populations for memetic algo

  MiloLurati

  committed
  95a2f0f

  Copy full SHA for 95a2f0f

• implemented adaptive local search depth logic and fix few issues, works also in simulation

  MiloLurati

  committed
  89c499b

  Copy full SHA for 89c499b

Commits on May 6, 2024

• modifications related to last iteration of memetic algo

  MiloLurati

  committed
  babba0b

  Copy full SHA for babba0b

• updates related to old popuation logic

  MiloLurati

  committed
  e0e1e61

  Copy full SHA for e0e1e61

Commits on May 7, 2024

• unified two actors into one

  MiloLurati

  committed
  6305782

  Copy full SHA for 6305782

• updates related to actors unification and memetic algo development

  MiloLurati

  committed
  0f2b7e4

  Copy full SHA for 0f2b7e4

• added create_actor_on_device and initialize_ray

  MiloLurati

  committed
  63ddedb

  Copy full SHA for 63ddedb

• updates realted to unification of actors, memetic algo, and reutilization of actors for performance

  MiloLurati

  committed
  d7fe9b4

  Copy full SHA for d7fe9b4

• returning 80% of cpus for simulation mode in get_num_devices

  MiloLurati

  committed
  46fcde1

  Copy full SHA for 46fcde1

• updates realted to actor unification and reutilization of actors for performance

  MiloLurati

  committed
  d543848

  Copy full SHA for d543848

• updates on feval counting and distributing

  MiloLurati

  committed
  15df6ea

  Copy full SHA for 15df6ea

Commits on May 8, 2024

• Merge branch 'KernelTuner:master' into parallelTuning

  MiloLurati

  authored
  96d03b8

  Copy full SHA for 96d03b8

• Merge branch 'parallelTuning' of https://github.com/MiloLurati/kernel_tuner into parallelTuning

  MiloLurati

  committed
  18ce214

  Copy full SHA for 18ce214

Commits on May 10, 2024

• added logic for time limit stop

  MiloLurati

  committed
  ec719a2

  Copy full SHA for ec719a2

• debug prints clean up

  MiloLurati

  committed
  6c2a62b

  Copy full SHA for 6c2a62b

• unified parallel tuning and parallel ensemble logic in ParallelRunner

  MiloLurati

  committed
  c7fd2af

  Copy full SHA for c7fd2af

Commits on May 28, 2024

• added self.init_arguments for parallel runner execution

  MiloLurati

  committed
  af532c5

  Copy full SHA for af532c5

• fix about non-pickleable observers and other small adjustments

  MiloLurati

  committed
  82d9886

  Copy full SHA for 82d9886

• now the cache manager deals only with the cache and not with the entire tuning option dict

  MiloLurati

  committed
  c6a2f36

  Copy full SHA for c6a2f36

• fix related to non-pickleable observers

  MiloLurati

  committed
  5fe2e56

  Copy full SHA for 5fe2e56

• update related to new cache manager

  MiloLurati

  committed
  3b3317c

  Copy full SHA for 3b3317c

• added cleanup at the end of the ensemble

  MiloLurati

  committed
  1593806

  Copy full SHA for 1593806

• changes to hyperparameters

  MiloLurati

  committed
  efd5be2

  Copy full SHA for efd5be2

• changes related to non-pickleable observers

  MiloLurati

  committed
  bc66244

  Copy full SHA for bc66244

• Merge branch 'master' into parallelTuning

  MiloLurati

  authored
  c5cfd05

  Copy full SHA for c5cfd05

Commits on May 31, 2024

• updated init_arguments to a dict

  MiloLurati

  committed
  9e9f1af

  Copy full SHA for 9e9f1af

• updates for searchspace split, ensemble related fix, and observer executing on correct device in parallel mode

  MiloLurati

  committed
  3fed66c

  Copy full SHA for 3fed66c

Commits on Jun 5, 2024

• small corections related to stop criterion for memetic

  MiloLurati

  committed
  86a9b67

  Copy full SHA for 86a9b67

Commits on Jun 7, 2024

• added logic to check if all GPUs are of the same type

  MiloLurati

  committed
  de5fc49

  Copy full SHA for de5fc49

• deleted split searchspace function

  MiloLurati

  committed
  1b0adb0

  Copy full SHA for 1b0adb0

• changed place where ray is initialized

  MiloLurati

  committed
  5130286

  Copy full SHA for 5130286

• setting BO to random sampling if needed

  MiloLurati

  committed
  5b9d817

  Copy full SHA for 5b9d817

• Merge branch 'KernelTuner:master' into parallelTuning

  MiloLurati

  authored
  8b1e57f

  Copy full SHA for 8b1e57f

• added num_gpus option

  MiloLurati

  committed
  040a57e

  Copy full SHA for 040a57e

Commits on Jun 10, 2024

• removed debug print

  MiloLurati

  committed
  acaaeb1

  Copy full SHA for acaaeb1

• added check_and_retrive strategy option

  MiloLurati

  committed
  63d9f65

  Copy full SHA for 63d9f65

Commits on Jun 18, 2024

• moved reinitialization of actor observers to execute method, before was in init

  MiloLurati

  committed
  e604510

  Copy full SHA for e604510

• changes related to re-initialization of observers in actor init and device interface

  MiloLurati

  committed
  5933a69

  Copy full SHA for 5933a69

Commits on Jun 21, 2024

• removed unnecesary blocking ray.get

  MiloLurati

  committed
  4e4c47b

  Copy full SHA for 4e4c47b

Commits on Jul 1, 2024

• removed debug prints

  MiloLurati

  committed
  104205d

  Copy full SHA for 104205d

• added greedy ils esemble instead of default

  MiloLurati

  committed
  123fba5

  Copy full SHA for 123fba5

• added check on strategy_options

  MiloLurati

  committed
  d381011

  Copy full SHA for d381011

• removed all memetic algo related stuff

  MiloLurati

  committed
  7e832e3

  Copy full SHA for 7e832e3

• Merge branch 'KernelTuner:master' into parallelTuning

  MiloLurati

  authored
  e976bf8

  Copy full SHA for e976bf8

• added ray to pyproject.toml

  MiloLurati

  committed
  65d32c1

  Copy full SHA for 65d32c1

• Merge branch 'parallelTuning' of https://github.com/MiloLurati/kernel_tuner into parallelTuning

  MiloLurati

  committed
  a841f2a

  Copy full SHA for a841f2a

• updated toml file with ray dashboard

  MiloLurati

  committed
  503df1b

  Copy full SHA for 503df1b

• fix small bug in _evaluate_configs

  MiloLurati

  committed
  c126a01

  Copy full SHA for c126a01

• adapted test for ensemble

  MiloLurati

  committed
  4df1b0d

  Copy full SHA for 4df1b0d

• cleaned up not used imports

  MiloLurati

  committed
  29a507c

  Copy full SHA for 29a507c

• added comments

  MiloLurati

  committed
  7c49a29

  Copy full SHA for 7c49a29

Commits on Jul 4, 2024

• added documentation and related fixes

  MiloLurati

  committed
  eb5db41

  Copy full SHA for eb5db41