Adding regression benchmark for the transformers SHA 777f2243f56a7aa8a9ea66d330af986aa57ebd49

raw_results/2023-09-27_09:51:44_777f2243f56a7aa8a9ea66d330af986aa57ebd49/bert_1gpu_training/0/.config/config.yaml

@@ -0,0 +1,75 @@
+backend:
+  name: pytorch
+  version: ${pytorch_version:}
+  _target_: optimum_benchmark.backends.pytorch.backend.PyTorchBackend
+  seed: 42
+  inter_op_num_threads: null
+  intra_op_num_threads: null
+  initial_isolation_check: false
+  continous_isolation_check: false
+  delete_cache: false
+  no_weights: false
+  device_map: null
+  torch_dtype: float16
+  disable_grad: ${is_inference:${benchmark.name}}
+  eval_mode: ${is_inference:${benchmark.name}}
+  amp_autocast: false
+  amp_dtype: null
+  torch_compile: false
+  torch_compile_config: {}
+  bettertransformer: false
+  quantization_scheme: null
+  quantization_config: {}
+  use_ddp: false
+  ddp_config: {}
+  peft_strategy: null
+  peft_config: {}
+benchmark:
+  name: training
+  _target_: optimum_benchmark.benchmarks.training.benchmark.TrainingBenchmark
+  warmup_steps: 40
+  dataset_shapes:
+    dataset_size: 1500
+    sequence_length: 256
+    num_choices: 1
+    feature_size: 80
+    nb_max_frames: 3000
+    audio_sequence_length: 16000
+  training_arguments:
+    skip_memory_metrics: true
+    output_dir: ./trainer_output
+    use_cpu: ${is_cpu:${device}}
+    ddp_find_unused_parameters: false
+    do_train: true
+    do_eval: false
+    do_predict: false
+    report_to: none
+    per_device_train_batch_size: 32
+experiment_name: bert_1gpu_training
+model: bert-base-uncased
+device: cuda
+task: text-classification
+hub_kwargs:
+  revision: main
+  cache_dir: null
+  force_download: false
+  local_files_only: false
+environment:
+  optimum_version: 1.13.1
+  transformers_version: 4.34.0.dev0
+  accelerate_version: 0.23.0
+  diffusers_version: null
+  python_version: 3.10.12
+  system: Linux
+  cpu: ' AMD EPYC 7643 48-Core Processor'
+  cpu_count: 96
+  cpu_ram_mb: 1082028
+  gpus:
+  - Instinct MI210
+  - Instinct MI210
+  - Instinct MI210
+  - Instinct MI210
+  - Instinct MI210
+  - Instinct MI210
+  - Instinct MI210
+  - Instinct MI210

raw_results/2023-09-27_09:51:44_777f2243f56a7aa8a9ea66d330af986aa57ebd49/bert_1gpu_training/0/.config/hydra.yaml

@@ -0,0 +1,174 @@
+hydra:
+  run:
+    dir: runs/${oc.env:COMMIT_DATE_GMT}_${oc.env:COMMIT_SHA}/${experiment_name}
+  sweep:
+    dir: sweeps/${oc.env:COMMIT_DATE_GMT}_${oc.env:COMMIT_SHA}/${experiment_name}
+    subdir: ${hydra.job.num}
+  launcher:
+    _target_: hydra._internal.core_plugins.basic_launcher.BasicLauncher
+  sweeper:
+    _target_: hydra._internal.core_plugins.basic_sweeper.BasicSweeper
+    max_batch_size: null
+    params:
+      +benchmark.training_arguments.per_device_train_batch_size: '32'
+      backend.torch_dtype: float16,float32
+  help:
+    app_name: ${hydra.job.name}
+    header: '${hydra.help.app_name} is powered by Hydra.
+
+      '
+    footer: 'Powered by Hydra (https://hydra.cc)
+
+      Use --hydra-help to view Hydra specific help
+
+      '
+    template: '${hydra.help.header}
+
+      == Configuration groups ==
+
+      Compose your configuration from those groups (group=option)
+
+
+      $APP_CONFIG_GROUPS
+
+
+      == Config ==
+
+      Override anything in the config (foo.bar=value)
+
+
+      $CONFIG
+
+
+      ${hydra.help.footer}
+
+      '
+  hydra_help:
+    template: 'Hydra (${hydra.runtime.version})
+
+      See https://hydra.cc for more info.
+
+
+      == Flags ==
+
+      $FLAGS_HELP
+
+
+      == Configuration groups ==
+
+      Compose your configuration from those groups (For example, append hydra/job_logging=disabled
+      to command line)
+
+
+      $HYDRA_CONFIG_GROUPS
+
+
+      Use ''--cfg hydra'' to Show the Hydra config.
+
+      '
+    hydra_help: ???
+  hydra_logging:
+    version: 1
+    formatters:
+      colorlog:
+        (): colorlog.ColoredFormatter
+        format: '[%(cyan)s%(asctime)s%(reset)s][%(purple)sHYDRA%(reset)s] %(message)s'
+    handlers:
+      console:
+        class: logging.StreamHandler
+        formatter: colorlog
+        stream: ext://sys.stdout
+    root:
+      level: INFO
+      handlers:
+      - console
+    disable_existing_loggers: false
+  job_logging:
+    version: 1
+    formatters:
+      simple:
+        format: '[%(asctime)s][%(name)s][%(levelname)s] - %(message)s'
+      colorlog:
+        (): colorlog.ColoredFormatter
+        format: '[%(cyan)s%(asctime)s%(reset)s][%(blue)s%(name)s%(reset)s][%(log_color)s%(levelname)s%(reset)s]
+          - %(message)s'
+        log_colors:
+          DEBUG: purple
+          INFO: green
+          WARNING: yellow
+          ERROR: red
+          CRITICAL: red
+    handlers:
+      console:
+        class: logging.StreamHandler
+        formatter: colorlog
+        stream: ext://sys.stdout
+      file:
+        class: logging.FileHandler
+        formatter: simple
+        filename: ${hydra.job.name}.log
+    root:
+      level: INFO
+      handlers:
+      - console
+      - file
+    disable_existing_loggers: false
+  env: {}
+  mode: MULTIRUN
+  searchpath: []
+  callbacks: {}
+  output_subdir: .hydra
+  overrides:
+    hydra:
+    - hydra.mode=MULTIRUN
+    task:
+    - +benchmark.training_arguments.per_device_train_batch_size=32
+    - backend.torch_dtype=float16
+  job:
+    name: experiment
+    chdir: true
+    override_dirname: +benchmark.training_arguments.per_device_train_batch_size=32,backend.torch_dtype=float16
+    id: '0'
+    num: 0
+    config_name: bert_1gpu_training
+    env_set: {}
+    env_copy: []
+    config:
+      override_dirname:
+        kv_sep: '='
+        item_sep: ','
+        exclude_keys: []
+  runtime:
+    version: 1.3.2
+    version_base: '1.3'
+    cwd: /home/user/transformers-regression
+    config_sources:
+    - path: hydra.conf
+      schema: pkg
+      provider: hydra
+    - path: optimum_benchmark
+      schema: pkg
+      provider: main
+    - path: hydra_plugins.hydra_colorlog.conf
+      schema: pkg
+      provider: hydra-colorlog
+    - path: /home/user/transformers-regression/configs
+      schema: file
+      provider: command-line
+    - path: ''
+      schema: structured
+      provider: schema
+    output_dir: /home/user/transformers-regression/sweeps/2023-09-27_09:51:44_777f2243f56a7aa8a9ea66d330af986aa57ebd49/bert_1gpu_training/0
+    choices:
+      benchmark: training
+      backend: pytorch
+      hydra/env: default
+      hydra/callbacks: null
+      hydra/job_logging: colorlog
+      hydra/hydra_logging: colorlog
+      hydra/hydra_help: default
+      hydra/help: default
+      hydra/sweeper: basic
+      hydra/launcher: basic
+      hydra/output: default
+  verbose: false

raw_results/2023-09-27_09:51:44_777f2243f56a7aa8a9ea66d330af986aa57ebd49/bert_1gpu_training/0/.config/overrides.yaml

@@ -0,0 +1,2 @@
+- +benchmark.training_arguments.per_device_train_batch_size=32
+- backend.torch_dtype=float16

raw_results/2023-09-27_09:51:44_777f2243f56a7aa8a9ea66d330af986aa57ebd49/bert_1gpu_training/0/experiment.log

@@ -0,0 +1,15 @@
+[2023-09-27 13:11:10,967][experiment][WARNING] - Multiple GPUs detected but CUDA_DEVICE_ORDER is not set. This means that code might allocate resources from the wrong GPUs even if CUDA_VISIBLE_DEVICES is set. Pytorch uses the `FASTEST_FIRST` order by default, which is not guaranteed to be the same as nvidia-smi. `CUDA_DEVICE_ORDER` will be set to `PCI_BUS_ID` to ensure that the GPUs are allocated in the same order as nvidia-smi. 
+[2023-09-27 13:11:13,718][pytorch][INFO] - 	+ Inferred AutoModel class AutoModelForSequenceClassification for task text-classification and model_type bert
+[2023-09-27 13:11:13,719][backend][INFO] - Configuring pytorch backend
+[2023-09-27 13:11:13,721][pytorch][INFO] - 	+ Loading model on device: cuda
+[2023-09-27 13:11:14,461][benchmark][INFO] - Configuring training benchmark
+[2023-09-27 13:11:14,461][training][INFO] - Running training benchmark
+[2023-09-27 13:11:14,461][dataset_generator][INFO] - Using text-classification task generator
+[2023-09-27 13:11:14,510][pytorch][INFO] - 	+ Setting dataset format to `torch`.
+[2023-09-27 13:11:14,510][pytorch][INFO] - 	+ Wrapping training arguments with transformers.TrainingArguments
+[2023-09-27 13:11:14,511][pytorch][INFO] - 	+ Wrapping model with transformers.Trainer
+[2023-09-27 13:11:14,514][pytorch][INFO] - 	+ Starting training
+[2023-09-27 13:11:30,510][pytorch][INFO] - 	+ Training finished successfully
+[2023-09-27 13:11:30,511][training][INFO] - Saving training results
+[2023-09-27 13:11:30,514][backend][INFO] - Cleaning pytorch backend
+[2023-09-27 13:11:30,514][backend][INFO] - 	+ Deleting pretrained model

raw_results/2023-09-27_09:51:44_777f2243f56a7aa8a9ea66d330af986aa57ebd49/bert_1gpu_training/0/hydra_config.yaml

@@ -0,0 +1,75 @@
+backend:
+  name: pytorch
+  version: 2.1.0+rocm5.6
+  _target_: optimum_benchmark.backends.pytorch.backend.PyTorchBackend
+  seed: 42
+  inter_op_num_threads: null
+  intra_op_num_threads: null
+  initial_isolation_check: false
+  continous_isolation_check: false
+  delete_cache: false
+  no_weights: false
+  device_map: null
+  torch_dtype: float16
+  disable_grad: false
+  eval_mode: false
+  amp_autocast: false
+  amp_dtype: null
+  torch_compile: false
+  torch_compile_config: {}
+  bettertransformer: false
+  quantization_scheme: null
+  quantization_config: {}
+  use_ddp: false
+  ddp_config: {}
+  peft_strategy: null
+  peft_config: {}
+benchmark:
+  name: training
+  _target_: optimum_benchmark.benchmarks.training.benchmark.TrainingBenchmark
+  warmup_steps: 40
+  dataset_shapes:
+    dataset_size: 1500
+    sequence_length: 256
+    num_choices: 1
+    feature_size: 80
+    nb_max_frames: 3000
+    audio_sequence_length: 16000
+  training_arguments:
+    skip_memory_metrics: true
+    output_dir: ./trainer_output
+    use_cpu: false
+    ddp_find_unused_parameters: false
+    do_train: true
+    do_eval: false
+    do_predict: false
+    report_to: none
+    per_device_train_batch_size: 32
+experiment_name: bert_1gpu_training
+model: bert-base-uncased
+device: cuda
+task: text-classification
+hub_kwargs:
+  revision: main
+  cache_dir: null
+  force_download: false
+  local_files_only: false
+environment:
+  optimum_version: 1.13.1
+  transformers_version: 4.34.0.dev0
+  accelerate_version: 0.23.0
+  diffusers_version: null
+  python_version: 3.10.12
+  system: Linux
+  cpu: ' AMD EPYC 7643 48-Core Processor'
+  cpu_count: 96
+  cpu_ram_mb: 1082028
+  gpus:
+  - Instinct MI210
+  - Instinct MI210
+  - Instinct MI210
+  - Instinct MI210
+  - Instinct MI210
+  - Instinct MI210
+  - Instinct MI210
+  - Instinct MI210

raw_results/2023-09-27_09:51:44_777f2243f56a7aa8a9ea66d330af986aa57ebd49/bert_1gpu_training/0/training_results.csv

@@ -0,0 +1,2 @@
+warmup.runtime(s),warmup.throughput(samples/s),training.runtime(s),training.throughput(samples/s),overall_training.runtime(s),overall_training.throughput(samples/s)
+4.898396015167236,261.3100280248165,11.003796339035034,293.7168137631514,15.902196407318115,203.24236459012977

raw_results/2023-09-27_09:51:44_777f2243f56a7aa8a9ea66d330af986aa57ebd49/bert_1gpu_training/1/.config/config.yaml

@@ -0,0 +1,75 @@
+backend:
+  name: pytorch
+  version: ${pytorch_version:}
+  _target_: optimum_benchmark.backends.pytorch.backend.PyTorchBackend
+  seed: 42
+  inter_op_num_threads: null
+  intra_op_num_threads: null
+  initial_isolation_check: false
+  continous_isolation_check: false
+  delete_cache: false
+  no_weights: false
+  device_map: null
+  torch_dtype: float32
+  disable_grad: ${is_inference:${benchmark.name}}
+  eval_mode: ${is_inference:${benchmark.name}}
+  amp_autocast: false
+  amp_dtype: null
+  torch_compile: false
+  torch_compile_config: {}
+  bettertransformer: false
+  quantization_scheme: null
+  quantization_config: {}
+  use_ddp: false
+  ddp_config: {}
+  peft_strategy: null
+  peft_config: {}
+benchmark:
+  name: training
+  _target_: optimum_benchmark.benchmarks.training.benchmark.TrainingBenchmark
+  warmup_steps: 40
+  dataset_shapes:
+    dataset_size: 1500
+    sequence_length: 256
+    num_choices: 1
+    feature_size: 80
+    nb_max_frames: 3000
+    audio_sequence_length: 16000
+  training_arguments:
+    skip_memory_metrics: true
+    output_dir: ./trainer_output
+    use_cpu: ${is_cpu:${device}}
+    ddp_find_unused_parameters: false
+    do_train: true
+    do_eval: false
+    do_predict: false
+    report_to: none
+    per_device_train_batch_size: 32
+experiment_name: bert_1gpu_training
+model: bert-base-uncased
+device: cuda
+task: text-classification
+hub_kwargs:
+  revision: main
+  cache_dir: null
+  force_download: false
+  local_files_only: false
+environment:
+  optimum_version: 1.13.1
+  transformers_version: 4.34.0.dev0
+  accelerate_version: 0.23.0
+  diffusers_version: null
+  python_version: 3.10.12
+  system: Linux
+  cpu: ' AMD EPYC 7643 48-Core Processor'
+  cpu_count: 96
+  cpu_ram_mb: 1082028
+  gpus:
+  - Instinct MI210
+  - Instinct MI210
+  - Instinct MI210
+  - Instinct MI210
+  - Instinct MI210
+  - Instinct MI210
+  - Instinct MI210
+  - Instinct MI210

raw_results/2023-09-27_09:51:44_777f2243f56a7aa8a9ea66d330af986aa57ebd49/bert_1gpu_training/1/.config/hydra.yaml

@@ -0,0 +1,174 @@
+hydra:
+  run:
+    dir: runs/${oc.env:COMMIT_DATE_GMT}_${oc.env:COMMIT_SHA}/${experiment_name}
+  sweep:
+    dir: sweeps/${oc.env:COMMIT_DATE_GMT}_${oc.env:COMMIT_SHA}/${experiment_name}
+    subdir: ${hydra.job.num}
+  launcher:
+    _target_: hydra._internal.core_plugins.basic_launcher.BasicLauncher
+  sweeper:
+    _target_: hydra._internal.core_plugins.basic_sweeper.BasicSweeper
+    max_batch_size: null
+    params:
+      +benchmark.training_arguments.per_device_train_batch_size: '32'
+      backend.torch_dtype: float16,float32
+  help:
+    app_name: ${hydra.job.name}
+    header: '${hydra.help.app_name} is powered by Hydra.
+
+      '
+    footer: 'Powered by Hydra (https://hydra.cc)
+
+      Use --hydra-help to view Hydra specific help
+
+      '
+    template: '${hydra.help.header}
+
+      == Configuration groups ==
+
+      Compose your configuration from those groups (group=option)
+
+
+      $APP_CONFIG_GROUPS
+
+
+      == Config ==
+
+      Override anything in the config (foo.bar=value)
+
+
+      $CONFIG
+
+
+      ${hydra.help.footer}
+
+      '
+  hydra_help:
+    template: 'Hydra (${hydra.runtime.version})
+
+      See https://hydra.cc for more info.
+
+
+      == Flags ==
+
+      $FLAGS_HELP
+
+
+      == Configuration groups ==
+
+      Compose your configuration from those groups (For example, append hydra/job_logging=disabled
+      to command line)
+
+
+      $HYDRA_CONFIG_GROUPS
+
+
+      Use ''--cfg hydra'' to Show the Hydra config.
+
+      '
+    hydra_help: ???
+  hydra_logging:
+    version: 1
+    formatters:
+      colorlog:
+        (): colorlog.ColoredFormatter
+        format: '[%(cyan)s%(asctime)s%(reset)s][%(purple)sHYDRA%(reset)s] %(message)s'
+    handlers:
+      console:
+        class: logging.StreamHandler
+        formatter: colorlog
+        stream: ext://sys.stdout
+    root:
+      level: INFO
+      handlers:
+      - console
+    disable_existing_loggers: false
+  job_logging:
+    version: 1
+    formatters:
+      simple:
+        format: '[%(asctime)s][%(name)s][%(levelname)s] - %(message)s'
+      colorlog:
+        (): colorlog.ColoredFormatter
+        format: '[%(cyan)s%(asctime)s%(reset)s][%(blue)s%(name)s%(reset)s][%(log_color)s%(levelname)s%(reset)s]
+          - %(message)s'
+        log_colors:
+          DEBUG: purple
+          INFO: green
+          WARNING: yellow
+          ERROR: red
+          CRITICAL: red
+    handlers:
+      console:
+        class: logging.StreamHandler
+        formatter: colorlog
+        stream: ext://sys.stdout
+      file:
+        class: logging.FileHandler
+        formatter: simple
+        filename: ${hydra.job.name}.log
+    root:
+      level: INFO
+      handlers:
+      - console
+      - file
+    disable_existing_loggers: false
+  env: {}
+  mode: MULTIRUN
+  searchpath: []
+  callbacks: {}
+  output_subdir: .hydra
+  overrides:
+    hydra:
+    - hydra.mode=MULTIRUN
+    task:
+    - +benchmark.training_arguments.per_device_train_batch_size=32
+    - backend.torch_dtype=float32
+  job:
+    name: experiment
+    chdir: true
+    override_dirname: +benchmark.training_arguments.per_device_train_batch_size=32,backend.torch_dtype=float32
+    id: '1'
+    num: 1
+    config_name: bert_1gpu_training
+    env_set: {}
+    env_copy: []
+    config:
+      override_dirname:
+        kv_sep: '='
+        item_sep: ','
+        exclude_keys: []
+  runtime:
+    version: 1.3.2
+    version_base: '1.3'
+    cwd: /home/user/transformers-regression
+    config_sources:
+    - path: hydra.conf
+      schema: pkg
+      provider: hydra
+    - path: optimum_benchmark
+      schema: pkg
+      provider: main
+    - path: hydra_plugins.hydra_colorlog.conf
+      schema: pkg
+      provider: hydra-colorlog
+    - path: /home/user/transformers-regression/configs
+      schema: file
+      provider: command-line
+    - path: ''
+      schema: structured
+      provider: schema
+    output_dir: /home/user/transformers-regression/sweeps/2023-09-27_09:51:44_777f2243f56a7aa8a9ea66d330af986aa57ebd49/bert_1gpu_training/1
+    choices:
+      benchmark: training
+      backend: pytorch
+      hydra/env: default
+      hydra/callbacks: null
+      hydra/job_logging: colorlog
+      hydra/hydra_logging: colorlog
+      hydra/hydra_help: default
+      hydra/help: default
+      hydra/sweeper: basic
+      hydra/launcher: basic
+      hydra/output: default
+  verbose: false

raw_results/2023-09-27_09:51:44_777f2243f56a7aa8a9ea66d330af986aa57ebd49/bert_1gpu_training/1/.config/overrides.yaml

@@ -0,0 +1,2 @@
+- +benchmark.training_arguments.per_device_train_batch_size=32
+- backend.torch_dtype=float32

raw_results/2023-09-27_09:51:44_777f2243f56a7aa8a9ea66d330af986aa57ebd49/bert_1gpu_training/1/experiment.log

@@ -0,0 +1,14 @@
+[2023-09-27 13:11:32,046][pytorch][INFO] - 	+ Inferred AutoModel class AutoModelForSequenceClassification for task text-classification and model_type bert
+[2023-09-27 13:11:32,047][backend][INFO] - Configuring pytorch backend
+[2023-09-27 13:11:32,047][pytorch][INFO] - 	+ Loading model on device: cuda
+[2023-09-27 13:11:32,468][benchmark][INFO] - Configuring training benchmark
+[2023-09-27 13:11:32,468][training][INFO] - Running training benchmark
+[2023-09-27 13:11:32,468][dataset_generator][INFO] - Using text-classification task generator
+[2023-09-27 13:11:32,495][pytorch][INFO] - 	+ Setting dataset format to `torch`.
+[2023-09-27 13:11:32,496][pytorch][INFO] - 	+ Wrapping training arguments with transformers.TrainingArguments
+[2023-09-27 13:11:32,496][pytorch][INFO] - 	+ Wrapping model with transformers.Trainer
+[2023-09-27 13:11:32,500][pytorch][INFO] - 	+ Starting training
+[2023-09-27 13:12:06,121][pytorch][INFO] - 	+ Training finished successfully
+[2023-09-27 13:12:06,122][training][INFO] - Saving training results
+[2023-09-27 13:12:06,124][backend][INFO] - Cleaning pytorch backend
+[2023-09-27 13:12:06,124][backend][INFO] - 	+ Deleting pretrained model

raw_results/2023-09-27_09:51:44_777f2243f56a7aa8a9ea66d330af986aa57ebd49/bert_1gpu_training/1/hydra_config.yaml

@@ -0,0 +1,75 @@
+backend:
+  name: pytorch
+  version: 2.1.0+rocm5.6
+  _target_: optimum_benchmark.backends.pytorch.backend.PyTorchBackend
+  seed: 42
+  inter_op_num_threads: null
+  intra_op_num_threads: null
+  initial_isolation_check: false
+  continous_isolation_check: false
+  delete_cache: false
+  no_weights: false
+  device_map: null
+  torch_dtype: float32
+  disable_grad: false
+  eval_mode: false
+  amp_autocast: false
+  amp_dtype: null
+  torch_compile: false
+  torch_compile_config: {}
+  bettertransformer: false
+  quantization_scheme: null
+  quantization_config: {}
+  use_ddp: false
+  ddp_config: {}
+  peft_strategy: null
+  peft_config: {}
+benchmark:
+  name: training
+  _target_: optimum_benchmark.benchmarks.training.benchmark.TrainingBenchmark
+  warmup_steps: 40
+  dataset_shapes:
+    dataset_size: 1500
+    sequence_length: 256
+    num_choices: 1
+    feature_size: 80
+    nb_max_frames: 3000
+    audio_sequence_length: 16000
+  training_arguments:
+    skip_memory_metrics: true
+    output_dir: ./trainer_output
+    use_cpu: false
+    ddp_find_unused_parameters: false
+    do_train: true
+    do_eval: false
+    do_predict: false
+    report_to: none
+    per_device_train_batch_size: 32
+experiment_name: bert_1gpu_training
+model: bert-base-uncased
+device: cuda
+task: text-classification
+hub_kwargs:
+  revision: main
+  cache_dir: null
+  force_download: false
+  local_files_only: false
+environment:
+  optimum_version: 1.13.1
+  transformers_version: 4.34.0.dev0
+  accelerate_version: 0.23.0
+  diffusers_version: null
+  python_version: 3.10.12
+  system: Linux
+  cpu: ' AMD EPYC 7643 48-Core Processor'
+  cpu_count: 96
+  cpu_ram_mb: 1082028
+  gpus:
+  - Instinct MI210
+  - Instinct MI210
+  - Instinct MI210
+  - Instinct MI210
+  - Instinct MI210
+  - Instinct MI210
+  - Instinct MI210
+  - Instinct MI210

raw_results/2023-09-27_09:51:44_777f2243f56a7aa8a9ea66d330af986aa57ebd49/bert_1gpu_training/1/training_results.csv

@@ -0,0 +1,2 @@
+warmup.runtime(s),warmup.throughput(samples/s),training.runtime(s),training.throughput(samples/s),overall_training.runtime(s),overall_training.throughput(samples/s)
+9.634719133377075,132.85286081311494,23.885178565979004,135.31403966991977,33.51990079879761,96.42033308511269

raw_results/2023-09-27_09:51:44_777f2243f56a7aa8a9ea66d330af986aa57ebd49/bert_1gpu_training/multirun.yaml

@@ -0,0 +1,246 @@
+hydra:
+  run:
+    dir: runs/${oc.env:COMMIT_DATE_GMT}_${oc.env:COMMIT_SHA}/${experiment_name}
+  sweep:
+    dir: sweeps/${oc.env:COMMIT_DATE_GMT}_${oc.env:COMMIT_SHA}/${experiment_name}
+    subdir: ${hydra.job.num}
+  launcher:
+    _target_: hydra._internal.core_plugins.basic_launcher.BasicLauncher
+  sweeper:
+    _target_: hydra._internal.core_plugins.basic_sweeper.BasicSweeper
+    max_batch_size: null
+    params:
+      +benchmark.training_arguments.per_device_train_batch_size: '32'
+      backend.torch_dtype: float16,float32
+  help:
+    app_name: ${hydra.job.name}
+    header: '${hydra.help.app_name} is powered by Hydra.
+
+      '
+    footer: 'Powered by Hydra (https://hydra.cc)
+
+      Use --hydra-help to view Hydra specific help
+
+      '
+    template: '${hydra.help.header}
+
+      == Configuration groups ==
+
+      Compose your configuration from those groups (group=option)
+
+
+      $APP_CONFIG_GROUPS
+
+
+      == Config ==
+
+      Override anything in the config (foo.bar=value)
+
+
+      $CONFIG
+
+
+      ${hydra.help.footer}
+
+      '
+  hydra_help:
+    template: 'Hydra (${hydra.runtime.version})
+
+      See https://hydra.cc for more info.
+
+
+      == Flags ==
+
+      $FLAGS_HELP
+
+
+      == Configuration groups ==
+
+      Compose your configuration from those groups (For example, append hydra/job_logging=disabled
+      to command line)
+
+
+      $HYDRA_CONFIG_GROUPS
+
+
+      Use ''--cfg hydra'' to Show the Hydra config.
+
+      '
+    hydra_help: ???
+  hydra_logging:
+    version: 1
+    formatters:
+      colorlog:
+        (): colorlog.ColoredFormatter
+        format: '[%(cyan)s%(asctime)s%(reset)s][%(purple)sHYDRA%(reset)s] %(message)s'
+    handlers:
+      console:
+        class: logging.StreamHandler
+        formatter: colorlog
+        stream: ext://sys.stdout
+    root:
+      level: INFO
+      handlers:
+      - console
+    disable_existing_loggers: false
+  job_logging:
+    version: 1
+    formatters:
+      simple:
+        format: '[%(asctime)s][%(name)s][%(levelname)s] - %(message)s'
+      colorlog:
+        (): colorlog.ColoredFormatter
+        format: '[%(cyan)s%(asctime)s%(reset)s][%(blue)s%(name)s%(reset)s][%(log_color)s%(levelname)s%(reset)s]
+          - %(message)s'
+        log_colors:
+          DEBUG: purple
+          INFO: green
+          WARNING: yellow
+          ERROR: red
+          CRITICAL: red
+    handlers:
+      console:
+        class: logging.StreamHandler
+        formatter: colorlog
+        stream: ext://sys.stdout
+      file:
+        class: logging.FileHandler
+        formatter: simple
+        filename: ${hydra.job.name}.log
+    root:
+      level: INFO
+      handlers:
+      - console
+      - file
+    disable_existing_loggers: false
+  env: {}
+  mode: MULTIRUN
+  searchpath: []
+  callbacks: {}
+  output_subdir: .hydra
+  overrides:
+    hydra:
+    - hydra.mode=MULTIRUN
+    task: []
+  job:
+    name: experiment
+    chdir: true
+    override_dirname: ''
+    id: ???
+    num: ???
+    config_name: bert_1gpu_training
+    env_set: {}
+    env_copy: []
+    config:
+      override_dirname:
+        kv_sep: '='
+        item_sep: ','
+        exclude_keys: []
+  runtime:
+    version: 1.3.2
+    version_base: '1.3'
+    cwd: /home/user/transformers-regression
+    config_sources:
+    - path: hydra.conf
+      schema: pkg
+      provider: hydra
+    - path: optimum_benchmark
+      schema: pkg
+      provider: main
+    - path: hydra_plugins.hydra_colorlog.conf
+      schema: pkg
+      provider: hydra-colorlog
+    - path: /home/user/transformers-regression/configs
+      schema: file
+      provider: command-line
+    - path: ''
+      schema: structured
+      provider: schema
+    output_dir: ???
+    choices:
+      benchmark: training
+      backend: pytorch
+      hydra/env: default
+      hydra/callbacks: null
+      hydra/job_logging: colorlog
+      hydra/hydra_logging: colorlog
+      hydra/hydra_help: default
+      hydra/help: default
+      hydra/sweeper: basic
+      hydra/launcher: basic
+      hydra/output: default
+  verbose: false
+backend:
+  name: pytorch
+  version: ${pytorch_version:}
+  _target_: optimum_benchmark.backends.pytorch.backend.PyTorchBackend
+  seed: 42
+  inter_op_num_threads: null
+  intra_op_num_threads: null
+  initial_isolation_check: false
+  continous_isolation_check: false
+  delete_cache: false
+  no_weights: false
+  device_map: null
+  torch_dtype: null
+  disable_grad: ${is_inference:${benchmark.name}}
+  eval_mode: ${is_inference:${benchmark.name}}
+  amp_autocast: false
+  amp_dtype: null
+  torch_compile: false
+  torch_compile_config: {}
+  bettertransformer: false
+  quantization_scheme: null
+  quantization_config: {}
+  use_ddp: false
+  ddp_config: {}
+  peft_strategy: null
+  peft_config: {}
+benchmark:
+  name: training
+  _target_: optimum_benchmark.benchmarks.training.benchmark.TrainingBenchmark
+  warmup_steps: 40
+  dataset_shapes:
+    dataset_size: 1500
+    sequence_length: 256
+    num_choices: 1
+    feature_size: 80
+    nb_max_frames: 3000
+    audio_sequence_length: 16000
+  training_arguments:
+    skip_memory_metrics: true
+    output_dir: ./trainer_output
+    use_cpu: ${is_cpu:${device}}
+    ddp_find_unused_parameters: false
+    do_train: true
+    do_eval: false
+    do_predict: false
+    report_to: none
+experiment_name: bert_1gpu_training
+model: bert-base-uncased
+device: cuda
+task: text-classification
+hub_kwargs:
+  revision: main
+  cache_dir: null
+  force_download: false
+  local_files_only: false
+environment:
+  optimum_version: 1.13.1
+  transformers_version: 4.34.0.dev0
+  accelerate_version: 0.23.0
+  diffusers_version: null
+  python_version: 3.10.12
+  system: Linux
+  cpu: ' AMD EPYC 7643 48-Core Processor'
+  cpu_count: 96
+  cpu_ram_mb: 1082028
+  gpus:
+  - Instinct MI210
+  - Instinct MI210
+  - Instinct MI210
+  - Instinct MI210
+  - Instinct MI210
+  - Instinct MI210
+  - Instinct MI210
+  - Instinct MI210

raw_results/2023-09-27_09:51:44_777f2243f56a7aa8a9ea66d330af986aa57ebd49/llama_1gpu_inference/0/.config/config.yaml

@@ -0,0 +1,73 @@
+backend:
+  name: pytorch
+  version: ${pytorch_version:}
+  _target_: optimum_benchmark.backends.pytorch.backend.PyTorchBackend
+  seed: 42
+  inter_op_num_threads: null
+  intra_op_num_threads: null
+  initial_isolation_check: false
+  continous_isolation_check: false
+  delete_cache: false
+  no_weights: false
+  device_map: null
+  torch_dtype: float16
+  disable_grad: ${is_inference:${benchmark.name}}
+  eval_mode: ${is_inference:${benchmark.name}}
+  amp_autocast: false
+  amp_dtype: null
+  torch_compile: false
+  torch_compile_config: {}
+  bettertransformer: false
+  quantization_scheme: null
+  quantization_config: {}
+  use_ddp: false
+  ddp_config: {}
+  peft_strategy: null
+  peft_config: {}
+benchmark:
+  name: inference
+  _target_: optimum_benchmark.benchmarks.inference.benchmark.InferenceBenchmark
+  duration: 15
+  warmup_runs: 10
+  memory: false
+  energy: false
+  input_shapes:
+    batch_size: 1
+    sequence_length: 200
+    num_choices: 1
+    feature_size: 80
+    nb_max_frames: 3000
+    audio_sequence_length: 16000
+  new_tokens: 200
+  can_diffuse: ${can_diffuse:${task}}
+  can_generate: ${can_generate:${task}}
+  forward_kwargs: {}
+  generate_kwargs: {}
+experiment_name: llama_1gpu_inference
+model: meta-llama/Llama-2-7b-chat-hf
+device: cuda
+task: ${infer_task:${model}}
+hub_kwargs:
+  revision: main
+  cache_dir: null
+  force_download: false
+  local_files_only: false
+environment:
+  optimum_version: 1.13.1
+  transformers_version: 4.34.0.dev0
+  accelerate_version: 0.23.0
+  diffusers_version: null
+  python_version: 3.10.12
+  system: Linux
+  cpu: ' AMD EPYC 7643 48-Core Processor'
+  cpu_count: 96
+  cpu_ram_mb: 1082028
+  gpus:
+  - Instinct MI210
+  - Instinct MI210
+  - Instinct MI210
+  - Instinct MI210
+  - Instinct MI210
+  - Instinct MI210
+  - Instinct MI210
+  - Instinct MI210

raw_results/2023-09-27_09:51:44_777f2243f56a7aa8a9ea66d330af986aa57ebd49/llama_1gpu_inference/0/.config/hydra.yaml

@@ -0,0 +1,174 @@
+hydra:
+  run:
+    dir: runs/${oc.env:COMMIT_DATE_GMT}_${oc.env:COMMIT_SHA}/${experiment_name}
+  sweep:
+    dir: sweeps/${oc.env:COMMIT_DATE_GMT}_${oc.env:COMMIT_SHA}/${experiment_name}
+    subdir: ${hydra.job.num}
+  launcher:
+    _target_: hydra._internal.core_plugins.basic_launcher.BasicLauncher
+  sweeper:
+    _target_: hydra._internal.core_plugins.basic_sweeper.BasicSweeper
+    max_batch_size: null
+    params:
+      benchmark.input_shapes.batch_size: 1,16
+      backend.torch_dtype: float16,float32
+  help:
+    app_name: ${hydra.job.name}
+    header: '${hydra.help.app_name} is powered by Hydra.
+
+      '
+    footer: 'Powered by Hydra (https://hydra.cc)
+
+      Use --hydra-help to view Hydra specific help
+
+      '
+    template: '${hydra.help.header}
+
+      == Configuration groups ==
+
+      Compose your configuration from those groups (group=option)
+
+
+      $APP_CONFIG_GROUPS
+
+
+      == Config ==
+
+      Override anything in the config (foo.bar=value)
+
+
+      $CONFIG
+
+
+      ${hydra.help.footer}
+
+      '
+  hydra_help:
+    template: 'Hydra (${hydra.runtime.version})
+
+      See https://hydra.cc for more info.
+
+
+      == Flags ==
+
+      $FLAGS_HELP
+
+
+      == Configuration groups ==
+
+      Compose your configuration from those groups (For example, append hydra/job_logging=disabled
+      to command line)
+
+
+      $HYDRA_CONFIG_GROUPS
+
+
+      Use ''--cfg hydra'' to Show the Hydra config.
+
+      '
+    hydra_help: ???
+  hydra_logging:
+    version: 1
+    formatters:
+      colorlog:
+        (): colorlog.ColoredFormatter
+        format: '[%(cyan)s%(asctime)s%(reset)s][%(purple)sHYDRA%(reset)s] %(message)s'
+    handlers:
+      console:
+        class: logging.StreamHandler
+        formatter: colorlog
+        stream: ext://sys.stdout
+    root:
+      level: INFO
+      handlers:
+      - console
+    disable_existing_loggers: false
+  job_logging:
+    version: 1
+    formatters:
+      simple:
+        format: '[%(asctime)s][%(name)s][%(levelname)s] - %(message)s'
+      colorlog:
+        (): colorlog.ColoredFormatter
+        format: '[%(cyan)s%(asctime)s%(reset)s][%(blue)s%(name)s%(reset)s][%(log_color)s%(levelname)s%(reset)s]
+          - %(message)s'
+        log_colors:
+          DEBUG: purple
+          INFO: green
+          WARNING: yellow
+          ERROR: red
+          CRITICAL: red
+    handlers:
+      console:
+        class: logging.StreamHandler
+        formatter: colorlog
+        stream: ext://sys.stdout
+      file:
+        class: logging.FileHandler
+        formatter: simple
+        filename: ${hydra.job.name}.log
+    root:
+      level: INFO
+      handlers:
+      - console
+      - file
+    disable_existing_loggers: false
+  env: {}
+  mode: MULTIRUN
+  searchpath: []
+  callbacks: {}
+  output_subdir: .hydra
+  overrides:
+    hydra:
+    - hydra.mode=MULTIRUN
+    task:
+    - benchmark.input_shapes.batch_size=1
+    - backend.torch_dtype=float16
+  job:
+    name: experiment
+    chdir: true
+    override_dirname: backend.torch_dtype=float16,benchmark.input_shapes.batch_size=1
+    id: '0'
+    num: 0
+    config_name: llama2_1gpu_inference
+    env_set: {}
+    env_copy: []
+    config:
+      override_dirname:
+        kv_sep: '='
+        item_sep: ','
+        exclude_keys: []
+  runtime:
+    version: 1.3.2
+    version_base: '1.3'
+    cwd: /home/user/transformers-regression
+    config_sources:
+    - path: hydra.conf
+      schema: pkg
+      provider: hydra
+    - path: optimum_benchmark
+      schema: pkg
+      provider: main
+    - path: hydra_plugins.hydra_colorlog.conf
+      schema: pkg
+      provider: hydra-colorlog
+    - path: /home/user/transformers-regression/configs
+      schema: file
+      provider: command-line
+    - path: ''
+      schema: structured
+      provider: schema
+    output_dir: /home/user/transformers-regression/sweeps/2023-09-27_09:51:44_777f2243f56a7aa8a9ea66d330af986aa57ebd49/llama_1gpu_inference/0
+    choices:
+      benchmark: inference
+      backend: pytorch
+      hydra/env: default
+      hydra/callbacks: null
+      hydra/job_logging: colorlog
+      hydra/hydra_logging: colorlog
+      hydra/hydra_help: default
+      hydra/help: default
+      hydra/sweeper: basic
+      hydra/launcher: basic
+      hydra/output: default
+  verbose: false

raw_results/2023-09-27_09:51:44_777f2243f56a7aa8a9ea66d330af986aa57ebd49/llama_1gpu_inference/0/.config/overrides.yaml

@@ -0,0 +1,2 @@
+- benchmark.input_shapes.batch_size=1
+- backend.torch_dtype=float16

raw_results/2023-09-27_09:51:44_777f2243f56a7aa8a9ea66d330af986aa57ebd49/llama_1gpu_inference/0/experiment.log

@@ -0,0 +1,25 @@
+[2023-09-27 13:12:10,550][inference][INFO] - `new_tokens` was set to 200. `max_new_tokens` and `min_new_tokens` will be set to 200.
+[2023-09-27 13:12:10,914][experiment][WARNING] - Multiple GPUs detected but CUDA_DEVICE_ORDER is not set. This means that code might allocate resources from the wrong GPUs even if CUDA_VISIBLE_DEVICES is set. Pytorch uses the `FASTEST_FIRST` order by default, which is not guaranteed to be the same as nvidia-smi. `CUDA_DEVICE_ORDER` will be set to `PCI_BUS_ID` to ensure that the GPUs are allocated in the same order as nvidia-smi. 
+[2023-09-27 13:12:13,068][pytorch][INFO] - 	+ Inferred AutoModel class AutoModelForCausalLM for task text-generation and model_type llama
+[2023-09-27 13:12:13,068][backend][INFO] - Configuring pytorch backend
+[2023-09-27 13:12:13,070][pytorch][INFO] - 	+ Disabling gradients
+[2023-09-27 13:12:13,070][pytorch][INFO] - 	+ Loading model on device: cuda
+[2023-09-27 13:12:18,820][pytorch][INFO] - 	+ Turning on model's eval mode
+[2023-09-27 13:12:18,828][benchmark][INFO] - Configuring inference benchmark
+[2023-09-27 13:12:18,828][inference][INFO] - Running inference benchmark
+[2023-09-27 13:12:18,829][input_generator][INFO] - Using llama model type generator
+[2023-09-27 13:12:18,850][inference][INFO] - 	+ Preparing input for the forward pass
+[2023-09-27 13:12:18,850][inference][INFO] - 	+ Warming up the forward pass
+[2023-09-27 13:12:19,791][inference][INFO] - 	+ Tracking forward pass latency and throughput
+[2023-09-27 13:12:19,792][latency_tracker][INFO] - Tracked Pytorch devices: [0]
+[2023-09-27 13:12:34,866][inference][INFO] - 	+ Forward pass latency: 6.67e-02 (s)
+[2023-09-27 13:12:34,866][inference][INFO] - 	+ Forward pass throughput: 15.00 (samples/s)
+[2023-09-27 13:12:34,867][inference][INFO] - 	+ Preparing input for the generation pass
+[2023-09-27 13:12:34,867][inference][INFO] - 	+ Warming up the generation pass
+[2023-09-27 13:12:40,362][inference][INFO] - 	+ Tracking generation latency and throughput
+[2023-09-27 13:12:40,363][latency_tracker][INFO] - Tracked Pytorch devices: [0]
+[2023-09-27 13:13:00,103][inference][INFO] - 	+ Generation pass latency: 4.93e+00 (s)
+[2023-09-27 13:13:00,103][inference][INFO] - 	+ Generation pass throughput: 40.60 (tokens/s)
+[2023-09-27 13:13:00,103][inference][INFO] - Saving inference results
+[2023-09-27 13:13:00,107][backend][INFO] - Cleaning pytorch backend
+[2023-09-27 13:13:00,107][backend][INFO] - 	+ Deleting pretrained model

raw_results/2023-09-27_09:51:44_777f2243f56a7aa8a9ea66d330af986aa57ebd49/llama_1gpu_inference/0/hydra_config.yaml

@@ -0,0 +1,79 @@
+backend:
+  name: pytorch
+  version: 2.1.0+rocm5.6
+  _target_: optimum_benchmark.backends.pytorch.backend.PyTorchBackend
+  seed: 42
+  inter_op_num_threads: null
+  intra_op_num_threads: null
+  initial_isolation_check: false
+  continous_isolation_check: false
+  delete_cache: false
+  no_weights: false
+  device_map: null
+  torch_dtype: float16
+  disable_grad: true
+  eval_mode: true
+  amp_autocast: false
+  amp_dtype: null
+  torch_compile: false
+  torch_compile_config: {}
+  bettertransformer: false
+  quantization_scheme: null
+  quantization_config: {}
+  use_ddp: false
+  ddp_config: {}
+  peft_strategy: null
+  peft_config: {}
+benchmark:
+  name: inference
+  _target_: optimum_benchmark.benchmarks.inference.benchmark.InferenceBenchmark
+  duration: 15
+  warmup_runs: 10
+  memory: false
+  energy: false
+  input_shapes:
+    batch_size: 1
+    sequence_length: 200
+    num_choices: 1
+    feature_size: 80
+    nb_max_frames: 3000
+    audio_sequence_length: 16000
+  new_tokens: 200
+  can_diffuse: false
+  can_generate: true
+  forward_kwargs: {}
+  generate_kwargs:
+    max_new_tokens: 200
+    min_new_tokens: 200
+    do_sample: false
+    use_cache: true
+    pad_token_id: 0
+    num_beams: 1
+experiment_name: llama_1gpu_inference
+model: meta-llama/Llama-2-7b-chat-hf
+device: cuda
+task: text-generation
+hub_kwargs:
+  revision: main
+  cache_dir: null
+  force_download: false
+  local_files_only: false
+environment:
+  optimum_version: 1.13.1
+  transformers_version: 4.34.0.dev0
+  accelerate_version: 0.23.0
+  diffusers_version: null
+  python_version: 3.10.12
+  system: Linux
+  cpu: ' AMD EPYC 7643 48-Core Processor'
+  cpu_count: 96
+  cpu_ram_mb: 1082028
+  gpus:
+  - Instinct MI210
+  - Instinct MI210
+  - Instinct MI210
+  - Instinct MI210
+  - Instinct MI210
+  - Instinct MI210
+  - Instinct MI210
+  - Instinct MI210

raw_results/2023-09-27_09:51:44_777f2243f56a7aa8a9ea66d330af986aa57ebd49/llama_1gpu_inference/0/inference_results.csv

@@ -0,0 +1,2 @@
+forward.latency(s),forward.throughput(samples/s),generate.latency(s),generate.throughput(tokens/s)
+0.0667,15.0,4.93,40.6

raw_results/2023-09-27_09:51:44_777f2243f56a7aa8a9ea66d330af986aa57ebd49/llama_1gpu_inference/1/.config/config.yaml

@@ -0,0 +1,73 @@
+backend:
+  name: pytorch
+  version: ${pytorch_version:}
+  _target_: optimum_benchmark.backends.pytorch.backend.PyTorchBackend
+  seed: 42
+  inter_op_num_threads: null
+  intra_op_num_threads: null
+  initial_isolation_check: false
+  continous_isolation_check: false
+  delete_cache: false
+  no_weights: false
+  device_map: null
+  torch_dtype: float32
+  disable_grad: ${is_inference:${benchmark.name}}
+  eval_mode: ${is_inference:${benchmark.name}}
+  amp_autocast: false
+  amp_dtype: null
+  torch_compile: false
+  torch_compile_config: {}
+  bettertransformer: false
+  quantization_scheme: null
+  quantization_config: {}
+  use_ddp: false
+  ddp_config: {}
+  peft_strategy: null
+  peft_config: {}
+benchmark:
+  name: inference
+  _target_: optimum_benchmark.benchmarks.inference.benchmark.InferenceBenchmark
+  duration: 15
+  warmup_runs: 10
+  memory: false
+  energy: false
+  input_shapes:
+    batch_size: 1
+    sequence_length: 200
+    num_choices: 1
+    feature_size: 80
+    nb_max_frames: 3000
+    audio_sequence_length: 16000
+  new_tokens: 200
+  can_diffuse: ${can_diffuse:${task}}
+  can_generate: ${can_generate:${task}}
+  forward_kwargs: {}
+  generate_kwargs: {}
+experiment_name: llama_1gpu_inference
+model: meta-llama/Llama-2-7b-chat-hf
+device: cuda
+task: ${infer_task:${model}}
+hub_kwargs:
+  revision: main
+  cache_dir: null
+  force_download: false
+  local_files_only: false
+environment:
+  optimum_version: 1.13.1
+  transformers_version: 4.34.0.dev0
+  accelerate_version: 0.23.0
+  diffusers_version: null
+  python_version: 3.10.12
+  system: Linux
+  cpu: ' AMD EPYC 7643 48-Core Processor'
+  cpu_count: 96
+  cpu_ram_mb: 1082028
+  gpus:
+  - Instinct MI210
+  - Instinct MI210
+  - Instinct MI210
+  - Instinct MI210
+  - Instinct MI210
+  - Instinct MI210
+  - Instinct MI210
+  - Instinct MI210

raw_results/2023-09-27_09:51:44_777f2243f56a7aa8a9ea66d330af986aa57ebd49/llama_1gpu_inference/1/.config/hydra.yaml

@@ -0,0 +1,174 @@
+hydra:
+  run:
+    dir: runs/${oc.env:COMMIT_DATE_GMT}_${oc.env:COMMIT_SHA}/${experiment_name}
+  sweep:
+    dir: sweeps/${oc.env:COMMIT_DATE_GMT}_${oc.env:COMMIT_SHA}/${experiment_name}
+    subdir: ${hydra.job.num}
+  launcher:
+    _target_: hydra._internal.core_plugins.basic_launcher.BasicLauncher
+  sweeper:
+    _target_: hydra._internal.core_plugins.basic_sweeper.BasicSweeper
+    max_batch_size: null
+    params:
+      benchmark.input_shapes.batch_size: 1,16
+      backend.torch_dtype: float16,float32
+  help:
+    app_name: ${hydra.job.name}
+    header: '${hydra.help.app_name} is powered by Hydra.
+
+      '
+    footer: 'Powered by Hydra (https://hydra.cc)
+
+      Use --hydra-help to view Hydra specific help
+
+      '
+    template: '${hydra.help.header}
+
+      == Configuration groups ==
+
+      Compose your configuration from those groups (group=option)
+
+
+      $APP_CONFIG_GROUPS
+
+
+      == Config ==
+
+      Override anything in the config (foo.bar=value)
+
+
+      $CONFIG
+
+
+      ${hydra.help.footer}
+
+      '
+  hydra_help:
+    template: 'Hydra (${hydra.runtime.version})
+
+      See https://hydra.cc for more info.
+
+
+      == Flags ==
+
+      $FLAGS_HELP
+
+
+      == Configuration groups ==
+
+      Compose your configuration from those groups (For example, append hydra/job_logging=disabled
+      to command line)
+
+
+      $HYDRA_CONFIG_GROUPS
+
+
+      Use ''--cfg hydra'' to Show the Hydra config.
+
+      '
+    hydra_help: ???
+  hydra_logging:
+    version: 1
+    formatters:
+      colorlog:
+        (): colorlog.ColoredFormatter
+        format: '[%(cyan)s%(asctime)s%(reset)s][%(purple)sHYDRA%(reset)s] %(message)s'
+    handlers:
+      console:
+        class: logging.StreamHandler
+        formatter: colorlog
+        stream: ext://sys.stdout
+    root:
+      level: INFO
+      handlers:
+      - console
+    disable_existing_loggers: false
+  job_logging:
+    version: 1
+    formatters:
+      simple:
+        format: '[%(asctime)s][%(name)s][%(levelname)s] - %(message)s'
+      colorlog:
+        (): colorlog.ColoredFormatter
+        format: '[%(cyan)s%(asctime)s%(reset)s][%(blue)s%(name)s%(reset)s][%(log_color)s%(levelname)s%(reset)s]
+          - %(message)s'
+        log_colors:
+          DEBUG: purple
+          INFO: green
+          WARNING: yellow
+          ERROR: red
+          CRITICAL: red
+    handlers:
+      console:
+        class: logging.StreamHandler
+        formatter: colorlog
+        stream: ext://sys.stdout
+      file:
+        class: logging.FileHandler
+        formatter: simple
+        filename: ${hydra.job.name}.log
+    root:
+      level: INFO
+      handlers:
+      - console
+      - file
+    disable_existing_loggers: false
+  env: {}
+  mode: MULTIRUN
+  searchpath: []
+  callbacks: {}
+  output_subdir: .hydra
+  overrides:
+    hydra:
+    - hydra.mode=MULTIRUN
+    task:
+    - benchmark.input_shapes.batch_size=1
+    - backend.torch_dtype=float32
+  job:
+    name: experiment
+    chdir: true
+    override_dirname: backend.torch_dtype=float32,benchmark.input_shapes.batch_size=1
+    id: '1'
+    num: 1
+    config_name: llama2_1gpu_inference
+    env_set: {}
+    env_copy: []
+    config:
+      override_dirname:
+        kv_sep: '='
+        item_sep: ','
+        exclude_keys: []
+  runtime:
+    version: 1.3.2
+    version_base: '1.3'
+    cwd: /home/user/transformers-regression
+    config_sources:
+    - path: hydra.conf
+      schema: pkg
+      provider: hydra
+    - path: optimum_benchmark
+      schema: pkg
+      provider: main
+    - path: hydra_plugins.hydra_colorlog.conf
+      schema: pkg
+      provider: hydra-colorlog
+    - path: /home/user/transformers-regression/configs
+      schema: file
+      provider: command-line
+    - path: ''
+      schema: structured
+      provider: schema
+    output_dir: /home/user/transformers-regression/sweeps/2023-09-27_09:51:44_777f2243f56a7aa8a9ea66d330af986aa57ebd49/llama_1gpu_inference/1
+    choices:
+      benchmark: inference
+      backend: pytorch
+      hydra/env: default
+      hydra/callbacks: null
+      hydra/job_logging: colorlog
+      hydra/hydra_logging: colorlog
+      hydra/hydra_help: default
+      hydra/help: default
+      hydra/sweeper: basic
+      hydra/launcher: basic
+      hydra/output: default
+  verbose: false

raw_results/2023-09-27_09:51:44_777f2243f56a7aa8a9ea66d330af986aa57ebd49/llama_1gpu_inference/1/.config/overrides.yaml

@@ -0,0 +1,2 @@
+- benchmark.input_shapes.batch_size=1
+- backend.torch_dtype=float32

raw_results/2023-09-27_09:51:44_777f2243f56a7aa8a9ea66d330af986aa57ebd49/llama_1gpu_inference/1/experiment.log

@@ -0,0 +1,24 @@
+[2023-09-27 13:13:01,453][inference][INFO] - `new_tokens` was set to 200. `max_new_tokens` and `min_new_tokens` will be set to 200.
+[2023-09-27 13:13:02,669][pytorch][INFO] - 	+ Inferred AutoModel class AutoModelForCausalLM for task text-generation and model_type llama
+[2023-09-27 13:13:02,670][backend][INFO] - Configuring pytorch backend
+[2023-09-27 13:13:02,670][pytorch][INFO] - 	+ Disabling gradients
+[2023-09-27 13:13:02,670][pytorch][INFO] - 	+ Loading model on device: cuda
+[2023-09-27 13:13:08,080][pytorch][INFO] - 	+ Turning on model's eval mode
+[2023-09-27 13:13:08,082][benchmark][INFO] - Configuring inference benchmark
+[2023-09-27 13:13:08,082][inference][INFO] - Running inference benchmark
+[2023-09-27 13:13:08,082][input_generator][INFO] - Using llama model type generator
+[2023-09-27 13:13:08,082][inference][INFO] - 	+ Preparing input for the forward pass
+[2023-09-27 13:13:08,083][inference][INFO] - 	+ Warming up the forward pass
+[2023-09-27 13:13:09,246][inference][INFO] - 	+ Tracking forward pass latency and throughput
+[2023-09-27 13:13:09,246][latency_tracker][INFO] - Tracked Pytorch devices: [0]
+[2023-09-27 13:13:24,381][inference][INFO] - 	+ Forward pass latency: 1.14e-01 (s)
+[2023-09-27 13:13:24,381][inference][INFO] - 	+ Forward pass throughput: 8.77 (samples/s)
+[2023-09-27 13:13:24,382][inference][INFO] - 	+ Preparing input for the generation pass
+[2023-09-27 13:13:24,382][inference][INFO] - 	+ Warming up the generation pass
+[2023-09-27 13:13:33,710][inference][INFO] - 	+ Tracking generation latency and throughput
+[2023-09-27 13:13:33,711][latency_tracker][INFO] - Tracked Pytorch devices: [0]
+[2023-09-27 13:13:52,371][inference][INFO] - 	+ Generation pass latency: 9.33e+00 (s)
+[2023-09-27 13:13:52,371][inference][INFO] - 	+ Generation pass throughput: 21.40 (tokens/s)
+[2023-09-27 13:13:52,371][inference][INFO] - Saving inference results
+[2023-09-27 13:13:52,374][backend][INFO] - Cleaning pytorch backend
+[2023-09-27 13:13:52,374][backend][INFO] - 	+ Deleting pretrained model

raw_results/2023-09-27_09:51:44_777f2243f56a7aa8a9ea66d330af986aa57ebd49/llama_1gpu_inference/1/hydra_config.yaml

@@ -0,0 +1,79 @@
+backend:
+  name: pytorch
+  version: 2.1.0+rocm5.6
+  _target_: optimum_benchmark.backends.pytorch.backend.PyTorchBackend
+  seed: 42
+  inter_op_num_threads: null
+  intra_op_num_threads: null
+  initial_isolation_check: false
+  continous_isolation_check: false
+  delete_cache: false
+  no_weights: false
+  device_map: null
+  torch_dtype: float32
+  disable_grad: true
+  eval_mode: true
+  amp_autocast: false
+  amp_dtype: null
+  torch_compile: false
+  torch_compile_config: {}
+  bettertransformer: false
+  quantization_scheme: null
+  quantization_config: {}
+  use_ddp: false
+  ddp_config: {}
+  peft_strategy: null
+  peft_config: {}
+benchmark:
+  name: inference
+  _target_: optimum_benchmark.benchmarks.inference.benchmark.InferenceBenchmark
+  duration: 15
+  warmup_runs: 10
+  memory: false
+  energy: false
+  input_shapes:
+    batch_size: 1
+    sequence_length: 200
+    num_choices: 1
+    feature_size: 80
+    nb_max_frames: 3000
+    audio_sequence_length: 16000
+  new_tokens: 200
+  can_diffuse: false
+  can_generate: true
+  forward_kwargs: {}
+  generate_kwargs:
+    max_new_tokens: 200
+    min_new_tokens: 200
+    do_sample: false
+    use_cache: true
+    pad_token_id: 0
+    num_beams: 1
+experiment_name: llama_1gpu_inference
+model: meta-llama/Llama-2-7b-chat-hf
+device: cuda
+task: text-generation
+hub_kwargs:
+  revision: main
+  cache_dir: null
+  force_download: false
+  local_files_only: false
+environment:
+  optimum_version: 1.13.1
+  transformers_version: 4.34.0.dev0
+  accelerate_version: 0.23.0
+  diffusers_version: null
+  python_version: 3.10.12
+  system: Linux
+  cpu: ' AMD EPYC 7643 48-Core Processor'
+  cpu_count: 96
+  cpu_ram_mb: 1082028
+  gpus:
+  - Instinct MI210
+  - Instinct MI210
+  - Instinct MI210
+  - Instinct MI210
+  - Instinct MI210
+  - Instinct MI210
+  - Instinct MI210
+  - Instinct MI210

raw_results/2023-09-27_09:51:44_777f2243f56a7aa8a9ea66d330af986aa57ebd49/llama_1gpu_inference/1/inference_results.csv

@@ -0,0 +1,2 @@
+forward.latency(s),forward.throughput(samples/s),generate.latency(s),generate.throughput(tokens/s)
+0.114,8.77,9.33,21.4

raw_results/2023-09-27_09:51:44_777f2243f56a7aa8a9ea66d330af986aa57ebd49/llama_1gpu_inference/2/.config/config.yaml

@@ -0,0 +1,73 @@
+backend:
+  name: pytorch
+  version: ${pytorch_version:}
+  _target_: optimum_benchmark.backends.pytorch.backend.PyTorchBackend
+  seed: 42
+  inter_op_num_threads: null
+  intra_op_num_threads: null
+  initial_isolation_check: false
+  continous_isolation_check: false
+  delete_cache: false
+  no_weights: false
+  device_map: null
+  torch_dtype: float16
+  disable_grad: ${is_inference:${benchmark.name}}
+  eval_mode: ${is_inference:${benchmark.name}}
+  amp_autocast: false
+  amp_dtype: null
+  torch_compile: false
+  torch_compile_config: {}
+  bettertransformer: false
+  quantization_scheme: null
+  quantization_config: {}
+  use_ddp: false
+  ddp_config: {}
+  peft_strategy: null
+  peft_config: {}
+benchmark:
+  name: inference
+  _target_: optimum_benchmark.benchmarks.inference.benchmark.InferenceBenchmark
+  duration: 15
+  warmup_runs: 10
+  memory: false
+  energy: false
+  input_shapes:
+    batch_size: 16
+    sequence_length: 200
+    num_choices: 1
+    feature_size: 80
+    nb_max_frames: 3000
+    audio_sequence_length: 16000
+  new_tokens: 200
+  can_diffuse: ${can_diffuse:${task}}
+  can_generate: ${can_generate:${task}}
+  forward_kwargs: {}
+  generate_kwargs: {}
+experiment_name: llama_1gpu_inference
+model: meta-llama/Llama-2-7b-chat-hf
+device: cuda
+task: ${infer_task:${model}}
+hub_kwargs:
+  revision: main
+  cache_dir: null
+  force_download: false
+  local_files_only: false
+environment:
+  optimum_version: 1.13.1
+  transformers_version: 4.34.0.dev0
+  accelerate_version: 0.23.0
+  diffusers_version: null
+  python_version: 3.10.12
+  system: Linux
+  cpu: ' AMD EPYC 7643 48-Core Processor'
+  cpu_count: 96
+  cpu_ram_mb: 1082028
+  gpus:
+  - Instinct MI210
+  - Instinct MI210
+  - Instinct MI210
+  - Instinct MI210
+  - Instinct MI210
+  - Instinct MI210
+  - Instinct MI210
+  - Instinct MI210

raw_results/2023-09-27_09:51:44_777f2243f56a7aa8a9ea66d330af986aa57ebd49/llama_1gpu_inference/2/.config/hydra.yaml

@@ -0,0 +1,174 @@
+hydra:
+  run:
+    dir: runs/${oc.env:COMMIT_DATE_GMT}_${oc.env:COMMIT_SHA}/${experiment_name}
+  sweep:
+    dir: sweeps/${oc.env:COMMIT_DATE_GMT}_${oc.env:COMMIT_SHA}/${experiment_name}
+    subdir: ${hydra.job.num}
+  launcher:
+    _target_: hydra._internal.core_plugins.basic_launcher.BasicLauncher
+  sweeper:
+    _target_: hydra._internal.core_plugins.basic_sweeper.BasicSweeper
+    max_batch_size: null
+    params:
+      benchmark.input_shapes.batch_size: 1,16
+      backend.torch_dtype: float16,float32
+  help:
+    app_name: ${hydra.job.name}
+    header: '${hydra.help.app_name} is powered by Hydra.
+
+      '
+    footer: 'Powered by Hydra (https://hydra.cc)
+
+      Use --hydra-help to view Hydra specific help
+
+      '
+    template: '${hydra.help.header}
+
+      == Configuration groups ==
+
+      Compose your configuration from those groups (group=option)
+
+
+      $APP_CONFIG_GROUPS
+
+
+      == Config ==
+
+      Override anything in the config (foo.bar=value)
+
+
+      $CONFIG
+
+
+      ${hydra.help.footer}
+
+      '
+  hydra_help:
+    template: 'Hydra (${hydra.runtime.version})
+
+      See https://hydra.cc for more info.
+
+
+      == Flags ==
+
+      $FLAGS_HELP
+
+
+      == Configuration groups ==
+
+      Compose your configuration from those groups (For example, append hydra/job_logging=disabled
+      to command line)
+
+
+      $HYDRA_CONFIG_GROUPS
+
+
+      Use ''--cfg hydra'' to Show the Hydra config.
+
+      '
+    hydra_help: ???
+  hydra_logging:
+    version: 1
+    formatters:
+      colorlog:
+        (): colorlog.ColoredFormatter
+        format: '[%(cyan)s%(asctime)s%(reset)s][%(purple)sHYDRA%(reset)s] %(message)s'
+    handlers:
+      console:
+        class: logging.StreamHandler
+        formatter: colorlog
+        stream: ext://sys.stdout
+    root:
+      level: INFO
+      handlers:
+      - console
+    disable_existing_loggers: false
+  job_logging:
+    version: 1
+    formatters:
+      simple:
+        format: '[%(asctime)s][%(name)s][%(levelname)s] - %(message)s'
+      colorlog:
+        (): colorlog.ColoredFormatter
+        format: '[%(cyan)s%(asctime)s%(reset)s][%(blue)s%(name)s%(reset)s][%(log_color)s%(levelname)s%(reset)s]
+          - %(message)s'
+        log_colors:
+          DEBUG: purple
+          INFO: green
+          WARNING: yellow
+          ERROR: red
+          CRITICAL: red
+    handlers:
+      console:
+        class: logging.StreamHandler
+        formatter: colorlog
+        stream: ext://sys.stdout
+      file:
+        class: logging.FileHandler
+        formatter: simple
+        filename: ${hydra.job.name}.log
+    root:
+      level: INFO
+      handlers:
+      - console
+      - file
+    disable_existing_loggers: false
+  env: {}
+  mode: MULTIRUN
+  searchpath: []
+  callbacks: {}
+  output_subdir: .hydra
+  overrides:
+    hydra:
+    - hydra.mode=MULTIRUN
+    task:
+    - benchmark.input_shapes.batch_size=16
+    - backend.torch_dtype=float16
+  job:
+    name: experiment
+    chdir: true
+    override_dirname: backend.torch_dtype=float16,benchmark.input_shapes.batch_size=16
+    id: '2'
+    num: 2
+    config_name: llama2_1gpu_inference
+    env_set: {}
+    env_copy: []
+    config:
+      override_dirname:
+        kv_sep: '='
+        item_sep: ','
+        exclude_keys: []
+  runtime:
+    version: 1.3.2
+    version_base: '1.3'
+    cwd: /home/user/transformers-regression
+    config_sources:
+    - path: hydra.conf
+      schema: pkg
+      provider: hydra
+    - path: optimum_benchmark
+      schema: pkg
+      provider: main
+    - path: hydra_plugins.hydra_colorlog.conf
+      schema: pkg
+      provider: hydra-colorlog
+    - path: /home/user/transformers-regression/configs
+      schema: file
+      provider: command-line
+    - path: ''
+      schema: structured
+      provider: schema
+    output_dir: /home/user/transformers-regression/sweeps/2023-09-27_09:51:44_777f2243f56a7aa8a9ea66d330af986aa57ebd49/llama_1gpu_inference/2
+    choices:
+      benchmark: inference
+      backend: pytorch
+      hydra/env: default
+      hydra/callbacks: null
+      hydra/job_logging: colorlog
+      hydra/hydra_logging: colorlog
+      hydra/hydra_help: default
+      hydra/help: default
+      hydra/sweeper: basic
+      hydra/launcher: basic
+      hydra/output: default
+  verbose: false

raw_results/2023-09-27_09:51:44_777f2243f56a7aa8a9ea66d330af986aa57ebd49/llama_1gpu_inference/2/.config/overrides.yaml

@@ -0,0 +1,2 @@
+- benchmark.input_shapes.batch_size=16
+- backend.torch_dtype=float16

raw_results/2023-09-27_09:51:44_777f2243f56a7aa8a9ea66d330af986aa57ebd49/llama_1gpu_inference/2/experiment.log

@@ -0,0 +1,24 @@
+[2023-09-27 13:13:53,382][inference][INFO] - `new_tokens` was set to 200. `max_new_tokens` and `min_new_tokens` will be set to 200.
+[2023-09-27 13:13:54,493][pytorch][INFO] - 	+ Inferred AutoModel class AutoModelForCausalLM for task text-generation and model_type llama
+[2023-09-27 13:13:54,494][backend][INFO] - Configuring pytorch backend
+[2023-09-27 13:13:54,494][pytorch][INFO] - 	+ Disabling gradients
+[2023-09-27 13:13:54,494][pytorch][INFO] - 	+ Loading model on device: cuda
+[2023-09-27 13:13:59,774][pytorch][INFO] - 	+ Turning on model's eval mode
+[2023-09-27 13:13:59,775][benchmark][INFO] - Configuring inference benchmark
+[2023-09-27 13:13:59,775][inference][INFO] - Running inference benchmark
+[2023-09-27 13:13:59,776][input_generator][INFO] - Using llama model type generator
+[2023-09-27 13:13:59,776][inference][INFO] - 	+ Preparing input for the forward pass
+[2023-09-27 13:13:59,776][inference][INFO] - 	+ Warming up the forward pass
+[2023-09-27 13:14:04,373][inference][INFO] - 	+ Tracking forward pass latency and throughput
+[2023-09-27 13:14:04,374][latency_tracker][INFO] - Tracked Pytorch devices: [0]
+[2023-09-27 13:14:20,112][inference][INFO] - 	+ Forward pass latency: 5.08e-01 (s)
+[2023-09-27 13:14:20,112][inference][INFO] - 	+ Forward pass throughput: 31.50 (samples/s)
+[2023-09-27 13:14:20,113][inference][INFO] - 	+ Preparing input for the generation pass
+[2023-09-27 13:14:20,113][inference][INFO] - 	+ Warming up the generation pass
+[2023-09-27 13:14:29,505][inference][INFO] - 	+ Tracking generation latency and throughput
+[2023-09-27 13:14:29,506][latency_tracker][INFO] - Tracked Pytorch devices: [0]
+[2023-09-27 13:14:47,603][inference][INFO] - 	+ Generation pass latency: 9.05e+00 (s)
+[2023-09-27 13:14:47,603][inference][INFO] - 	+ Generation pass throughput: 354.00 (tokens/s)
+[2023-09-27 13:14:47,603][inference][INFO] - Saving inference results
+[2023-09-27 13:14:47,605][backend][INFO] - Cleaning pytorch backend
+[2023-09-27 13:14:47,605][backend][INFO] - 	+ Deleting pretrained model

raw_results/2023-09-27_09:51:44_777f2243f56a7aa8a9ea66d330af986aa57ebd49/llama_1gpu_inference/2/hydra_config.yaml

@@ -0,0 +1,79 @@
+backend:
+  name: pytorch
+  version: 2.1.0+rocm5.6
+  _target_: optimum_benchmark.backends.pytorch.backend.PyTorchBackend
+  seed: 42
+  inter_op_num_threads: null
+  intra_op_num_threads: null
+  initial_isolation_check: false
+  continous_isolation_check: false
+  delete_cache: false
+  no_weights: false
+  device_map: null
+  torch_dtype: float16
+  disable_grad: true
+  eval_mode: true
+  amp_autocast: false
+  amp_dtype: null
+  torch_compile: false
+  torch_compile_config: {}
+  bettertransformer: false
+  quantization_scheme: null
+  quantization_config: {}
+  use_ddp: false
+  ddp_config: {}
+  peft_strategy: null
+  peft_config: {}
+benchmark:
+  name: inference
+  _target_: optimum_benchmark.benchmarks.inference.benchmark.InferenceBenchmark
+  duration: 15
+  warmup_runs: 10
+  memory: false
+  energy: false
+  input_shapes:
+    batch_size: 16
+    sequence_length: 200
+    num_choices: 1
+    feature_size: 80
+    nb_max_frames: 3000
+    audio_sequence_length: 16000
+  new_tokens: 200
+  can_diffuse: false
+  can_generate: true
+  forward_kwargs: {}
+  generate_kwargs:
+    max_new_tokens: 200
+    min_new_tokens: 200
+    do_sample: false
+    use_cache: true
+    pad_token_id: 0
+    num_beams: 1
+experiment_name: llama_1gpu_inference
+model: meta-llama/Llama-2-7b-chat-hf
+device: cuda
+task: text-generation
+hub_kwargs:
+  revision: main
+  cache_dir: null
+  force_download: false
+  local_files_only: false
+environment:
+  optimum_version: 1.13.1
+  transformers_version: 4.34.0.dev0
+  accelerate_version: 0.23.0
+  diffusers_version: null
+  python_version: 3.10.12
+  system: Linux
+  cpu: ' AMD EPYC 7643 48-Core Processor'
+  cpu_count: 96
+  cpu_ram_mb: 1082028
+  gpus:
+  - Instinct MI210
+  - Instinct MI210
+  - Instinct MI210
+  - Instinct MI210
+  - Instinct MI210
+  - Instinct MI210
+  - Instinct MI210
+  - Instinct MI210

raw_results/2023-09-27_09:51:44_777f2243f56a7aa8a9ea66d330af986aa57ebd49/llama_1gpu_inference/2/inference_results.csv

@@ -0,0 +1,2 @@
+forward.latency(s),forward.throughput(samples/s),generate.latency(s),generate.throughput(tokens/s)
+0.508,31.5,9.05,354.0

raw_results/2023-09-27_09:51:44_777f2243f56a7aa8a9ea66d330af986aa57ebd49/llama_1gpu_inference/3/.config/config.yaml

@@ -0,0 +1,73 @@
+backend:
+  name: pytorch
+  version: ${pytorch_version:}
+  _target_: optimum_benchmark.backends.pytorch.backend.PyTorchBackend
+  seed: 42
+  inter_op_num_threads: null
+  intra_op_num_threads: null
+  initial_isolation_check: false
+  continous_isolation_check: false
+  delete_cache: false
+  no_weights: false
+  device_map: null
+  torch_dtype: float32
+  disable_grad: ${is_inference:${benchmark.name}}
+  eval_mode: ${is_inference:${benchmark.name}}
+  amp_autocast: false
+  amp_dtype: null
+  torch_compile: false
+  torch_compile_config: {}
+  bettertransformer: false
+  quantization_scheme: null
+  quantization_config: {}
+  use_ddp: false
+  ddp_config: {}
+  peft_strategy: null
+  peft_config: {}
+benchmark:
+  name: inference
+  _target_: optimum_benchmark.benchmarks.inference.benchmark.InferenceBenchmark
+  duration: 15
+  warmup_runs: 10
+  memory: false
+  energy: false
+  input_shapes:
+    batch_size: 16
+    sequence_length: 200
+    num_choices: 1
+    feature_size: 80
+    nb_max_frames: 3000
+    audio_sequence_length: 16000
+  new_tokens: 200
+  can_diffuse: ${can_diffuse:${task}}
+  can_generate: ${can_generate:${task}}
+  forward_kwargs: {}
+  generate_kwargs: {}
+experiment_name: llama_1gpu_inference
+model: meta-llama/Llama-2-7b-chat-hf
+device: cuda
+task: ${infer_task:${model}}
+hub_kwargs:
+  revision: main
+  cache_dir: null
+  force_download: false
+  local_files_only: false
+environment:
+  optimum_version: 1.13.1
+  transformers_version: 4.34.0.dev0
+  accelerate_version: 0.23.0
+  diffusers_version: null
+  python_version: 3.10.12
+  system: Linux
+  cpu: ' AMD EPYC 7643 48-Core Processor'
+  cpu_count: 96
+  cpu_ram_mb: 1082028
+  gpus:
+  - Instinct MI210
+  - Instinct MI210
+  - Instinct MI210
+  - Instinct MI210
+  - Instinct MI210
+  - Instinct MI210
+  - Instinct MI210
+  - Instinct MI210

raw_results/2023-09-27_09:51:44_777f2243f56a7aa8a9ea66d330af986aa57ebd49/llama_1gpu_inference/3/.config/hydra.yaml

@@ -0,0 +1,174 @@
+hydra:
+  run:
+    dir: runs/${oc.env:COMMIT_DATE_GMT}_${oc.env:COMMIT_SHA}/${experiment_name}
+  sweep:
+    dir: sweeps/${oc.env:COMMIT_DATE_GMT}_${oc.env:COMMIT_SHA}/${experiment_name}
+    subdir: ${hydra.job.num}
+  launcher:
+    _target_: hydra._internal.core_plugins.basic_launcher.BasicLauncher
+  sweeper:
+    _target_: hydra._internal.core_plugins.basic_sweeper.BasicSweeper
+    max_batch_size: null
+    params:
+      benchmark.input_shapes.batch_size: 1,16
+      backend.torch_dtype: float16,float32
+  help:
+    app_name: ${hydra.job.name}
+    header: '${hydra.help.app_name} is powered by Hydra.
+
+      '
+    footer: 'Powered by Hydra (https://hydra.cc)
+
+      Use --hydra-help to view Hydra specific help
+
+      '
+    template: '${hydra.help.header}
+
+      == Configuration groups ==
+
+      Compose your configuration from those groups (group=option)
+
+
+      $APP_CONFIG_GROUPS
+
+
+      == Config ==
+
+      Override anything in the config (foo.bar=value)
+
+
+      $CONFIG
+
+
+      ${hydra.help.footer}
+
+      '
+  hydra_help:
+    template: 'Hydra (${hydra.runtime.version})
+
+      See https://hydra.cc for more info.
+
+
+      == Flags ==
+
+      $FLAGS_HELP
+
+
+      == Configuration groups ==
+
+      Compose your configuration from those groups (For example, append hydra/job_logging=disabled
+      to command line)
+
+
+      $HYDRA_CONFIG_GROUPS
+
+
+      Use ''--cfg hydra'' to Show the Hydra config.
+
+      '
+    hydra_help: ???
+  hydra_logging:
+    version: 1
+    formatters:
+      colorlog:
+        (): colorlog.ColoredFormatter
+        format: '[%(cyan)s%(asctime)s%(reset)s][%(purple)sHYDRA%(reset)s] %(message)s'
+    handlers:
+      console:
+        class: logging.StreamHandler
+        formatter: colorlog
+        stream: ext://sys.stdout
+    root:
+      level: INFO
+      handlers:
+      - console
+    disable_existing_loggers: false
+  job_logging:
+    version: 1
+    formatters:
+      simple:
+        format: '[%(asctime)s][%(name)s][%(levelname)s] - %(message)s'
+      colorlog:
+        (): colorlog.ColoredFormatter
+        format: '[%(cyan)s%(asctime)s%(reset)s][%(blue)s%(name)s%(reset)s][%(log_color)s%(levelname)s%(reset)s]
+          - %(message)s'
+        log_colors:
+          DEBUG: purple
+          INFO: green
+          WARNING: yellow
+          ERROR: red
+          CRITICAL: red
+    handlers:
+      console:
+        class: logging.StreamHandler
+        formatter: colorlog
+        stream: ext://sys.stdout
+      file:
+        class: logging.FileHandler
+        formatter: simple
+        filename: ${hydra.job.name}.log
+    root:
+      level: INFO
+      handlers:
+      - console
+      - file
+    disable_existing_loggers: false
+  env: {}
+  mode: MULTIRUN
+  searchpath: []
+  callbacks: {}
+  output_subdir: .hydra
+  overrides:
+    hydra:
+    - hydra.mode=MULTIRUN
+    task:
+    - benchmark.input_shapes.batch_size=16
+    - backend.torch_dtype=float32
+  job:
+    name: experiment
+    chdir: true
+    override_dirname: backend.torch_dtype=float32,benchmark.input_shapes.batch_size=16
+    id: '3'
+    num: 3
+    config_name: llama2_1gpu_inference
+    env_set: {}
+    env_copy: []
+    config:
+      override_dirname:
+        kv_sep: '='
+        item_sep: ','
+        exclude_keys: []
+  runtime:
+    version: 1.3.2
+    version_base: '1.3'
+    cwd: /home/user/transformers-regression
+    config_sources:
+    - path: hydra.conf
+      schema: pkg
+      provider: hydra
+    - path: optimum_benchmark
+      schema: pkg
+      provider: main
+    - path: hydra_plugins.hydra_colorlog.conf
+      schema: pkg
+      provider: hydra-colorlog
+    - path: /home/user/transformers-regression/configs
+      schema: file
+      provider: command-line
+    - path: ''
+      schema: structured
+      provider: schema
+    output_dir: /home/user/transformers-regression/sweeps/2023-09-27_09:51:44_777f2243f56a7aa8a9ea66d330af986aa57ebd49/llama_1gpu_inference/3
+    choices:
+      benchmark: inference
+      backend: pytorch
+      hydra/env: default
+      hydra/callbacks: null
+      hydra/job_logging: colorlog
+      hydra/hydra_logging: colorlog
+      hydra/hydra_help: default
+      hydra/help: default
+      hydra/sweeper: basic
+      hydra/launcher: basic
+      hydra/output: default
+  verbose: false

raw_results/2023-09-27_09:51:44_777f2243f56a7aa8a9ea66d330af986aa57ebd49/llama_1gpu_inference/3/.config/overrides.yaml

@@ -0,0 +1,2 @@
+- benchmark.input_shapes.batch_size=16
+- backend.torch_dtype=float32

raw_results/2023-09-27_09:51:44_777f2243f56a7aa8a9ea66d330af986aa57ebd49/llama_1gpu_inference/3/experiment.log

@@ -0,0 +1,24 @@
+[2023-09-27 13:14:48,805][inference][INFO] - `new_tokens` was set to 200. `max_new_tokens` and `min_new_tokens` will be set to 200.
+[2023-09-27 13:14:49,959][pytorch][INFO] - 	+ Inferred AutoModel class AutoModelForCausalLM for task text-generation and model_type llama
+[2023-09-27 13:14:49,959][backend][INFO] - Configuring pytorch backend
+[2023-09-27 13:14:49,959][pytorch][INFO] - 	+ Disabling gradients
+[2023-09-27 13:14:49,959][pytorch][INFO] - 	+ Loading model on device: cuda
+[2023-09-27 13:14:55,134][pytorch][INFO] - 	+ Turning on model's eval mode
+[2023-09-27 13:14:55,135][benchmark][INFO] - Configuring inference benchmark
+[2023-09-27 13:14:55,136][inference][INFO] - Running inference benchmark
+[2023-09-27 13:14:55,136][input_generator][INFO] - Using llama model type generator
+[2023-09-27 13:14:55,136][inference][INFO] - 	+ Preparing input for the forward pass
+[2023-09-27 13:14:55,137][inference][INFO] - 	+ Warming up the forward pass
+[2023-09-27 13:15:07,999][inference][INFO] - 	+ Tracking forward pass latency and throughput
+[2023-09-27 13:15:08,000][latency_tracker][INFO] - Tracked Pytorch devices: [0]
+[2023-09-27 13:15:25,011][inference][INFO] - 	+ Forward pass latency: 1.42e+00 (s)
+[2023-09-27 13:15:25,012][inference][INFO] - 	+ Forward pass throughput: 11.30 (samples/s)
+[2023-09-27 13:15:25,013][inference][INFO] - 	+ Preparing input for the generation pass
+[2023-09-27 13:15:25,013][inference][INFO] - 	+ Warming up the generation pass
+[2023-09-27 13:15:39,694][inference][INFO] - 	+ Tracking generation latency and throughput
+[2023-09-27 13:15:39,694][latency_tracker][INFO] - Tracked Pytorch devices: [0]
+[2023-09-27 13:16:08,113][inference][INFO] - 	+ Generation pass latency: 1.42e+01 (s)
+[2023-09-27 13:16:08,114][inference][INFO] - 	+ Generation pass throughput: 225.00 (tokens/s)
+[2023-09-27 13:16:08,114][inference][INFO] - Saving inference results
+[2023-09-27 13:16:08,116][backend][INFO] - Cleaning pytorch backend
+[2023-09-27 13:16:08,116][backend][INFO] - 	+ Deleting pretrained model

raw_results/2023-09-27_09:51:44_777f2243f56a7aa8a9ea66d330af986aa57ebd49/llama_1gpu_inference/3/hydra_config.yaml

@@ -0,0 +1,79 @@
+backend:
+  name: pytorch
+  version: 2.1.0+rocm5.6
+  _target_: optimum_benchmark.backends.pytorch.backend.PyTorchBackend
+  seed: 42
+  inter_op_num_threads: null
+  intra_op_num_threads: null
+  initial_isolation_check: false
+  continous_isolation_check: false
+  delete_cache: false
+  no_weights: false
+  device_map: null
+  torch_dtype: float32
+  disable_grad: true
+  eval_mode: true
+  amp_autocast: false
+  amp_dtype: null
+  torch_compile: false
+  torch_compile_config: {}
+  bettertransformer: false
+  quantization_scheme: null
+  quantization_config: {}
+  use_ddp: false
+  ddp_config: {}
+  peft_strategy: null
+  peft_config: {}
+benchmark:
+  name: inference
+  _target_: optimum_benchmark.benchmarks.inference.benchmark.InferenceBenchmark
+  duration: 15
+  warmup_runs: 10
+  memory: false
+  energy: false
+  input_shapes:
+    batch_size: 16
+    sequence_length: 200
+    num_choices: 1
+    feature_size: 80
+    nb_max_frames: 3000
+    audio_sequence_length: 16000
+  new_tokens: 200
+  can_diffuse: false
+  can_generate: true
+  forward_kwargs: {}
+  generate_kwargs:
+    max_new_tokens: 200
+    min_new_tokens: 200
+    do_sample: false
+    use_cache: true
+    pad_token_id: 0
+    num_beams: 1
+experiment_name: llama_1gpu_inference
+model: meta-llama/Llama-2-7b-chat-hf
+device: cuda
+task: text-generation
+hub_kwargs:
+  revision: main
+  cache_dir: null
+  force_download: false
+  local_files_only: false
+environment:
+  optimum_version: 1.13.1
+  transformers_version: 4.34.0.dev0
+  accelerate_version: 0.23.0
+  diffusers_version: null
+  python_version: 3.10.12
+  system: Linux
+  cpu: ' AMD EPYC 7643 48-Core Processor'
+  cpu_count: 96
+  cpu_ram_mb: 1082028
+  gpus:
+  - Instinct MI210
+  - Instinct MI210
+  - Instinct MI210
+  - Instinct MI210
+  - Instinct MI210
+  - Instinct MI210
+  - Instinct MI210
+  - Instinct MI210

raw_results/2023-09-27_09:51:44_777f2243f56a7aa8a9ea66d330af986aa57ebd49/llama_1gpu_inference/3/inference_results.csv

@@ -0,0 +1,2 @@
+forward.latency(s),forward.throughput(samples/s),generate.latency(s),generate.throughput(tokens/s)
+1.42,11.3,14.2,225.0

raw_results/2023-09-27_09:51:44_777f2243f56a7aa8a9ea66d330af986aa57ebd49/llama_1gpu_inference/multirun.yaml

@@ -0,0 +1,245 @@
+hydra:
+  run:
+    dir: runs/${oc.env:COMMIT_DATE_GMT}_${oc.env:COMMIT_SHA}/${experiment_name}
+  sweep:
+    dir: sweeps/${oc.env:COMMIT_DATE_GMT}_${oc.env:COMMIT_SHA}/${experiment_name}
+    subdir: ${hydra.job.num}
+  launcher:
+    _target_: hydra._internal.core_plugins.basic_launcher.BasicLauncher
+  sweeper:
+    _target_: hydra._internal.core_plugins.basic_sweeper.BasicSweeper
+    max_batch_size: null
+    params:
+      benchmark.input_shapes.batch_size: 1,16
+      backend.torch_dtype: float16,float32
+  help:
+    app_name: ${hydra.job.name}
+    header: '${hydra.help.app_name} is powered by Hydra.
+
+      '
+    footer: 'Powered by Hydra (https://hydra.cc)
+
+      Use --hydra-help to view Hydra specific help
+
+      '
+    template: '${hydra.help.header}
+
+      == Configuration groups ==
+
+      Compose your configuration from those groups (group=option)
+
+
+      $APP_CONFIG_GROUPS
+
+
+      == Config ==
+
+      Override anything in the config (foo.bar=value)
+
+
+      $CONFIG
+
+
+      ${hydra.help.footer}
+
+      '
+  hydra_help:
+    template: 'Hydra (${hydra.runtime.version})
+
+      See https://hydra.cc for more info.
+
+
+      == Flags ==
+
+      $FLAGS_HELP
+
+
+      == Configuration groups ==
+
+      Compose your configuration from those groups (For example, append hydra/job_logging=disabled
+      to command line)
+
+
+      $HYDRA_CONFIG_GROUPS
+
+
+      Use ''--cfg hydra'' to Show the Hydra config.
+
+      '
+    hydra_help: ???
+  hydra_logging:
+    version: 1
+    formatters:
+      colorlog:
+        (): colorlog.ColoredFormatter
+        format: '[%(cyan)s%(asctime)s%(reset)s][%(purple)sHYDRA%(reset)s] %(message)s'
+    handlers:
+      console:
+        class: logging.StreamHandler
+        formatter: colorlog
+        stream: ext://sys.stdout
+    root:
+      level: INFO
+      handlers:
+      - console
+    disable_existing_loggers: false
+  job_logging:
+    version: 1
+    formatters:
+      simple:
+        format: '[%(asctime)s][%(name)s][%(levelname)s] - %(message)s'
+      colorlog:
+        (): colorlog.ColoredFormatter
+        format: '[%(cyan)s%(asctime)s%(reset)s][%(blue)s%(name)s%(reset)s][%(log_color)s%(levelname)s%(reset)s]
+          - %(message)s'
+        log_colors:
+          DEBUG: purple
+          INFO: green
+          WARNING: yellow
+          ERROR: red
+          CRITICAL: red
+    handlers:
+      console:
+        class: logging.StreamHandler
+        formatter: colorlog
+        stream: ext://sys.stdout
+      file:
+        class: logging.FileHandler
+        formatter: simple
+        filename: ${hydra.job.name}.log
+    root:
+      level: INFO
+      handlers:
+      - console
+      - file
+    disable_existing_loggers: false
+  env: {}
+  mode: MULTIRUN
+  searchpath: []
+  callbacks: {}
+  output_subdir: .hydra
+  overrides:
+    hydra:
+    - hydra.mode=MULTIRUN
+    task: []
+  job:
+    name: experiment
+    chdir: true
+    override_dirname: ''
+    id: ???
+    num: ???
+    config_name: llama2_1gpu_inference
+    env_set: {}
+    env_copy: []
+    config:
+      override_dirname:
+        kv_sep: '='
+        item_sep: ','
+        exclude_keys: []
+  runtime:
+    version: 1.3.2
+    version_base: '1.3'
+    cwd: /home/user/transformers-regression
+    config_sources:
+    - path: hydra.conf
+      schema: pkg
+      provider: hydra
+    - path: optimum_benchmark
+      schema: pkg
+      provider: main
+    - path: hydra_plugins.hydra_colorlog.conf
+      schema: pkg
+      provider: hydra-colorlog
+    - path: /home/user/transformers-regression/configs
+      schema: file
+      provider: command-line
+    - path: ''
+      schema: structured
+      provider: schema
+    output_dir: ???
+    choices:
+      benchmark: inference
+      backend: pytorch
+      hydra/env: default
+      hydra/callbacks: null
+      hydra/job_logging: colorlog
+      hydra/hydra_logging: colorlog
+      hydra/hydra_help: default
+      hydra/help: default
+      hydra/sweeper: basic
+      hydra/launcher: basic
+      hydra/output: default
+  verbose: false
+backend:
+  name: pytorch
+  version: ${pytorch_version:}
+  _target_: optimum_benchmark.backends.pytorch.backend.PyTorchBackend
+  seed: 42
+  inter_op_num_threads: null
+  intra_op_num_threads: null
+  initial_isolation_check: false
+  continous_isolation_check: false
+  delete_cache: false
+  no_weights: false
+  device_map: null
+  torch_dtype: null
+  disable_grad: ${is_inference:${benchmark.name}}
+  eval_mode: ${is_inference:${benchmark.name}}
+  amp_autocast: false
+  amp_dtype: null
+  torch_compile: false
+  torch_compile_config: {}
+  bettertransformer: false
+  quantization_scheme: null
+  quantization_config: {}
+  use_ddp: false
+  ddp_config: {}
+  peft_strategy: null
+  peft_config: {}
+benchmark:
+  name: inference
+  _target_: optimum_benchmark.benchmarks.inference.benchmark.InferenceBenchmark
+  duration: 15
+  warmup_runs: 10
+  memory: false
+  energy: false
+  input_shapes:
+    batch_size: 2
+    sequence_length: 200
+    num_choices: 1
+    feature_size: 80
+    nb_max_frames: 3000
+    audio_sequence_length: 16000
+  new_tokens: 200
+  can_diffuse: ${can_diffuse:${task}}
+  can_generate: ${can_generate:${task}}
+  forward_kwargs: {}
+  generate_kwargs: {}
+experiment_name: llama_1gpu_inference
+model: meta-llama/Llama-2-7b-chat-hf
+device: cuda
+task: ${infer_task:${model}}
+hub_kwargs:
+  revision: main
+  cache_dir: null
+  force_download: false
+  local_files_only: false
+environment:
+  optimum_version: 1.13.1
+  transformers_version: 4.34.0.dev0
+  accelerate_version: 0.23.0
+  diffusers_version: null
+  python_version: 3.10.12
+  system: Linux
+  cpu: ' AMD EPYC 7643 48-Core Processor'
+  cpu_count: 96
+  cpu_ram_mb: 1082028
+  gpus:
+  - Instinct MI210
+  - Instinct MI210
+  - Instinct MI210
+  - Instinct MI210
+  - Instinct MI210
+  - Instinct MI210
+  - Instinct MI210
+  - Instinct MI210