Update dataset

audace/pytorch+cuda+float16+bnb-4bit/TigerResearch/tigerbot-13b-base/.hydra/config.yaml

@@ -0,0 +1,77 @@
+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
+  delete_cache: false
+  no_weights: true
+  device_map: null
+  torch_dtype: float16
+  eval_mode: ${is_inference:${benchmark.name}}
+  disable_grad: ${is_inference:${benchmark.name}}
+  amp_autocast: false
+  amp_dtype: null
+  torch_compile: false
+  torch_compile_config: {}
+  to_bettertransformer: false
+  use_flash_attention_2: false
+  quantization_scheme: bnb
+  quantization_config:
+    load_in_4bit: true
+  data_parallel: false
+  deepspeed_inference: false
+  deepspeed_inference_config: {}
+  peft_strategy: null
+  peft_config: {}
+benchmark:
+  name: inference
+  _target_: optimum_benchmark.benchmarks.inference.benchmark.InferenceBenchmark
+  duration: 10
+  warmup_runs: 10
+  memory: true
+  energy: true
+  input_shapes:
+    batch_size: 1
+    sequence_length: 256
+    num_choices: 1
+    feature_size: 80
+    nb_max_frames: 3000
+    audio_sequence_length: 16000
+  new_tokens: 256
+  can_diffuse: ${can_diffuse:${task}}
+  can_generate: ${can_generate:${task}}
+  forward_kwargs: {}
+  generate_kwargs: {}
+launcher:
+  name: process
+  _target_: optimum_benchmark.launchers.process.launcher.ProcessLauncher
+  device_isolation: true
+  start_method: spawn
+experiment_name: pytorch+cuda+float16+bnb-4bit
+device: cuda
+model: TigerResearch/tigerbot-13b-base
+task: ${infer_task:${model}}
+hub_kwargs:
+  revision: main
+  cache_dir: null
+  force_download: false
+  local_files_only: false
+  trust_remote_code: true
+environment:
+  optimum_version: 1.16.1
+  optimum_commit: null
+  transformers_version: 4.36.2
+  transformers_commit: null
+  accelerate_version: 0.26.1
+  accelerate_commit: null
+  diffusers_version: null
+  diffusers_commit: null
+  python_version: 3.10.12
+  system: Linux
+  cpu: ' AMD Ryzen 9 7950X 16-Core Processor'
+  cpu_count: 32
+  cpu_ram_mb: 134796
+  gpus:
+  - NVIDIA GeForce RTX 4090

audace/pytorch+cuda+float16+bnb-4bit/TigerResearch/tigerbot-13b-base/.hydra/hydra.yaml

@@ -0,0 +1,176 @@
+hydra:
+  run:
+    dir: dataset/${oc.env:HOSTNAME}/${experiment_name}/${model}
+  sweep:
+    dir: multirun/${now:%Y-%m-%d}/${now:%H-%M-%S}
+    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: null
+  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: RUN
+  searchpath: []
+  callbacks: {}
+  output_subdir: .hydra
+  overrides:
+    hydra:
+    - hydra.mode=RUN
+    task:
+    - model=TigerResearch/tigerbot-13b-base
+  job:
+    name: cli
+    chdir: true
+    override_dirname: model=TigerResearch/tigerbot-13b-base
+    id: ???
+    num: ???
+    config_name: pytorch+cuda+float16+bnb-4bit
+    env_set:
+      COUNTRY_ISO_CODE: FRA
+      OVERRIDE_BENCHMARKS: '0'
+      CUDA_VISIBLE_DEVICES: '0'
+      CUDA_DEVICE_ORDER: PCI_BUS_ID
+    env_copy: []
+    config:
+      override_dirname:
+        kv_sep: '='
+        item_sep: ','
+        exclude_keys: []
+  runtime:
+    version: 1.3.2
+    version_base: '1.3'
+    cwd: /workspace/llm-perf
+    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: /workspace/llm-perf/configs
+      schema: file
+      provider: command-line
+    - path: ''
+      schema: structured
+      provider: schema
+    output_dir: /workspace/llm-perf/dataset/audace/pytorch+cuda+float16+bnb-4bit/TigerResearch/tigerbot-13b-base
+    choices:
+      launcher: process
+      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

audace/pytorch+cuda+float16+bnb-4bit/TigerResearch/tigerbot-13b-base/.hydra/overrides.yaml

@@ -0,0 +1 @@
+- model=TigerResearch/tigerbot-13b-base

audace/pytorch+cuda+float16+bnb-4bit/TigerResearch/tigerbot-13b-base/cli.log

@@ -0,0 +1,19 @@
+[2024-01-13 09:15:00,813][launcher][INFO] - Configuring process launcher
+[2024-01-13 09:15:00,814][process][INFO] - Setting multiprocessing start method to spawn.
+[2024-01-13 09:15:00,815][process][INFO] - 	+ Launched worker process with PID 440606.
+[2024-01-13 09:15:00,815][isolation][INFO] - 	+ Launched device(s) isolation process 440607.
+[2024-01-13 09:15:03,008][numexpr.utils][INFO] - Note: NumExpr detected 32 cores but "NUMEXPR_MAX_THREADS" not set, so enforcing safe limit of 8.
+[2024-01-13 09:15:03,009][numexpr.utils][INFO] - NumExpr defaulting to 8 threads.
+[2024-01-13 09:15:03,165][datasets][INFO] - PyTorch version 2.1.2+cu118 available.
+[2024-01-13 09:15:04,972][backend][INFO] - Configuring pytorch backend
+[2024-01-13 09:15:04,973][pytorch][INFO] - 	+ Inferred AutoModel class AutoModelForCausalLM for task text-generation and model_type llama
+[2024-01-13 09:15:04,973][pytorch][INFO] - 	+ Disabling gradients
+[2024-01-13 09:15:04,973][pytorch][INFO] - 	+ Processing quantization config
+[2024-01-13 09:15:04,973][pytorch][INFO] - 	+ Processing BitsAndBytes config
+[2024-01-13 09:15:04,973][pytorch][INFO] - 	+ Loading model with no weights
+[2024-01-13 09:15:04,973][pytorch][INFO] - 	+ Creating no weights model directory
+[2024-01-13 09:15:04,973][pytorch][INFO] - 	+ Saving pretrained config
+[2024-01-13 09:15:04,974][pytorch][INFO] - 	+ Creating no weights model
+[2024-01-13 09:15:04,992][pytorch][INFO] - 	+ Saving no weights model
+[2024-01-13 09:15:04,993][pytorch][INFO] - 	+ Loading no weights model
+[2024-01-13 09:15:04,993][pytorch][INFO] - 	+ Loading quantized model

audace/pytorch+cuda+float16+bnb-4bit/TigerResearch/tigerbot-13b-base/hydra_config.yaml

@@ -0,0 +1,86 @@
+launcher:
+  name: process
+  _target_: optimum_benchmark.launchers.process.launcher.ProcessLauncher
+  device_isolation: true
+  start_method: spawn
+backend:
+  name: pytorch
+  version: 2.1.2+cu118
+  _target_: optimum_benchmark.backends.pytorch.backend.PyTorchBackend
+  seed: 42
+  inter_op_num_threads: null
+  intra_op_num_threads: null
+  delete_cache: false
+  no_weights: true
+  device_map: null
+  torch_dtype: float16
+  eval_mode: true
+  disable_grad: true
+  amp_autocast: false
+  amp_dtype: null
+  torch_compile: false
+  torch_compile_config: {}
+  to_bettertransformer: false
+  use_flash_attention_2: false
+  quantization_scheme: bnb
+  quantization_config:
+    llm_int8_threshold: 0.0
+    load_in_4bit: true
+  data_parallel: false
+  deepspeed_inference: false
+  deepspeed_inference_config: {}
+  peft_strategy: null
+  peft_config: {}
+benchmark:
+  name: inference
+  _target_: optimum_benchmark.benchmarks.inference.benchmark.InferenceBenchmark
+  duration: 10
+  warmup_runs: 10
+  memory: true
+  energy: true
+  input_shapes:
+    batch_size: 1
+    sequence_length: 256
+    num_choices: 1
+    feature_size: 80
+    nb_max_frames: 3000
+    audio_sequence_length: 16000
+  new_tokens: 256
+  can_diffuse: false
+  can_generate: true
+  forward_kwargs: {}
+  generate_kwargs:
+    num_return_sequences: 1
+    max_new_tokens: 256
+    min_new_tokens: 256
+    do_sample: false
+    use_cache: true
+    pad_token_id: 0
+    temperature: 1.0
+    num_beams: 1
+experiment_name: pytorch+cuda+float16+bnb-4bit
+device: cuda
+model: TigerResearch/tigerbot-13b-base
+task: text-generation
+hub_kwargs:
+  revision: main
+  cache_dir: null
+  force_download: false
+  local_files_only: false
+  trust_remote_code: true
+environment:
+  optimum_version: 1.16.1
+  optimum_commit: null
+  transformers_version: 4.36.2
+  transformers_commit: null
+  accelerate_version: 0.26.1
+  accelerate_commit: null
+  diffusers_version: null
+  diffusers_commit: null
+  python_version: 3.10.12
+  system: Linux
+  cpu: ' AMD Ryzen 9 7950X 16-Core Processor'
+  cpu_count: 32
+  cpu_ram_mb: 134796
+  gpus:
+  - NVIDIA GeForce RTX 4090

audace/pytorch+cuda+float16/AI-Sweden-Models/gpt-sw3-1.3b/.hydra/config.yaml

@@ -0,0 +1,77 @@
+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
+  continuous_isolation: true
+  isolation_check_interval: null
+  delete_cache: false
+  no_weights: true
+  device_map: null
+  torch_dtype: float16
+  eval_mode: ${is_inference:${benchmark.name}}
+  disable_grad: ${is_inference:${benchmark.name}}
+  amp_autocast: false
+  amp_dtype: null
+  torch_compile: false
+  torch_compile_config: {}
+  to_bettertransformer: false
+  use_flash_attention_2: false
+  quantization_scheme: null
+  quantization_config: {}
+  data_parallel: false
+  deepspeed_inference: false
+  deepspeed_inference_config: {}
+  peft_strategy: null
+  peft_config: {}
+benchmark:
+  name: inference
+  _target_: optimum_benchmark.benchmarks.inference.benchmark.InferenceBenchmark
+  duration: 10
+  warmup_runs: 10
+  memory: true
+  energy: true
+  input_shapes:
+    batch_size: 1
+    sequence_length: 256
+    num_choices: 1
+    feature_size: 80
+    nb_max_frames: 3000
+    audio_sequence_length: 16000
+  new_tokens: 256
+  can_diffuse: ${can_diffuse:${task}}
+  can_generate: ${can_generate:${task}}
+  forward_kwargs: {}
+  generate_kwargs: {}
+launcher:
+  name: process
+  _target_: optimum_benchmark.launchers.process.launcher.ProcessLauncher
+  start_method: spawn
+experiment_name: pytorch+cuda+float16
+device: cuda
+model: AI-Sweden-Models/gpt-sw3-1.3b
+task: ${infer_task:${model}}
+hub_kwargs:
+  revision: main
+  cache_dir: null
+  force_download: false
+  local_files_only: false
+  trust_remote_code: true
+environment:
+  optimum_version: 1.16.1
+  optimum_commit: null
+  transformers_version: 4.36.2
+  transformers_commit: null
+  accelerate_version: 0.25.0
+  accelerate_commit: null
+  diffusers_version: null
+  diffusers_commit: null
+  python_version: 3.10.12
+  system: Linux
+  cpu: ' AMD Ryzen 9 7950X 16-Core Processor'
+  cpu_count: 32
+  cpu_ram_mb: 134796
+  gpus:
+  - NVIDIA GeForce RTX 4090

audace/pytorch+cuda+float16/AI-Sweden-Models/gpt-sw3-1.3b/.hydra/hydra.yaml

@@ -0,0 +1,176 @@
+hydra:
+  run:
+    dir: dataset/${oc.env:HOSTNAME}/${experiment_name}/${model}
+  sweep:
+    dir: multirun/${now:%Y-%m-%d}/${now:%H-%M-%S}
+    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: null
+  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: RUN
+  searchpath: []
+  callbacks: {}
+  output_subdir: .hydra
+  overrides:
+    hydra:
+    - hydra.mode=RUN
+    task:
+    - model=AI-Sweden-Models/gpt-sw3-1.3b
+  job:
+    name: cli
+    chdir: true
+    override_dirname: model=AI-Sweden-Models/gpt-sw3-1.3b
+    id: ???
+    num: ???
+    config_name: pytorch+cuda+float16
+    env_set:
+      COUNTRY_ISO_CODE: FRA
+      OVERRIDE_BENCHMARKS: '0'
+      CUDA_VISIBLE_DEVICES: '0'
+      CUDA_DEVICE_ORDER: PCI_BUS_ID
+    env_copy: []
+    config:
+      override_dirname:
+        kv_sep: '='
+        item_sep: ','
+        exclude_keys: []
+  runtime:
+    version: 1.3.2
+    version_base: '1.3'
+    cwd: /workspace/llm-perf
+    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: /workspace/llm-perf/configs
+      schema: file
+      provider: command-line
+    - path: ''
+      schema: structured
+      provider: schema
+    output_dir: /workspace/llm-perf/dataset/audace/pytorch+cuda+float16/AI-Sweden-Models/gpt-sw3-1.3b
+    choices:
+      launcher: process
+      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

audace/pytorch+cuda+float16/AI-Sweden-Models/gpt-sw3-1.3b/.hydra/overrides.yaml

@@ -0,0 +1 @@
+- model=AI-Sweden-Models/gpt-sw3-1.3b

audace/pytorch+cuda+float16/AI-Sweden-Models/gpt-sw3-1.3b/cli.log

@@ -0,0 +1,76 @@
+[2024-01-10 02:39:11,179][launcher][INFO] - Configuring process launcher
+[2024-01-10 02:39:11,179][process][INFO] - Setting multiprocessing start method to spawn.
+[2024-01-10 02:39:11,180][process][INFO] - 	+ Launched experiment in process with PID 73076.
+[2024-01-10 02:39:13,201][numexpr.utils][INFO] - Note: NumExpr detected 32 cores but "NUMEXPR_MAX_THREADS" not set, so enforcing safe limit of 8.
+[2024-01-10 02:39:13,201][numexpr.utils][INFO] - NumExpr defaulting to 8 threads.
+[2024-01-10 02:39:13,359][datasets][INFO] - PyTorch version 2.1.2+cu118 available.
+[2024-01-10 02:39:14,738][backend][WARNING] - Could not find the model's generation config
+[2024-01-10 02:39:14,738][backend][INFO] - Configuring pytorch backend
+[2024-01-10 02:39:14,738][backend][INFO] - 	+ Running continuous isolation check
+[2024-01-10 02:39:14,739][backend][INFO] - 	+ Started isolation process with PID 73302
+[2024-01-10 02:39:14,739][pytorch][INFO] - 	+ Inferred AutoModel class AutoModelForCausalLM for task text-generation and model_type gpt2
+[2024-01-10 02:39:14,739][pytorch][INFO] - 	+ Disabling gradients
+[2024-01-10 02:39:14,739][pytorch][INFO] - 	+ Loading model with no weights
+[2024-01-10 02:39:14,739][pytorch][INFO] - 	+ Creating no weights model directory
+[2024-01-10 02:39:14,739][pytorch][INFO] - 	+ Saving pretrained config to /tmp/tmp8n0oqyqc/no_weights
+[2024-01-10 02:39:14,740][pytorch][INFO] - 	+ Creating no weights model to /tmp/tmp8n0oqyqc/no_weights
+[2024-01-10 02:39:14,760][pytorch][INFO] - 	+ Saving no weights model to /tmp/tmp8n0oqyqc/no_weights
+[2024-01-10 02:39:14,760][pytorch][INFO] - 	+ Loading no weights model
+[2024-01-10 02:39:14,760][pytorch][INFO] - 	+ Loading model directly on device: cuda
+[2024-01-10 02:39:14,926][pytorch][INFO] - 	+ Randomizing model weights
+[2024-01-10 02:39:14,928][pytorch][INFO] - 	+ Tying model weights after randomization
+[2024-01-10 02:39:14,929][pytorch][INFO] - 	+ Turning on model's eval mode
+[2024-01-10 02:39:14,991][benchmark][INFO] - Configuring inference benchmark
+[2024-01-10 02:39:14,991][inference][INFO] - Running inference benchmark
+[2024-01-10 02:39:14,991][inference][INFO] - 	+ Updating input shapes with model shapes
+[2024-01-10 02:39:14,991][inference][INFO] - 	+ Preparing backend for inference
+[2024-01-10 02:39:14,991][inference][INFO] - 	+ Creating input generator
+[2024-01-10 02:39:14,991][input-generator][INFO] - Using text-generation task generator
+[2024-01-10 02:39:14,991][inference][INFO] - 	+ Preparing input for the forward pass
+[2024-01-10 02:39:14,991][backend][INFO] - 	+ Moving inputs tensors to device cuda
+[2024-01-10 02:39:14,992][inference][INFO] - 	+ Tracking forward pass peak memory
+[2024-01-10 02:39:14,992][memory][INFO] - Tracking CUDA devices: [0]
+[2024-01-10 02:39:14,992][memory][INFO] - Tracking Pytorch CUDA devices: [0]
+[2024-01-10 02:39:15,220][inference][INFO] - 	+ Forward pass max memory used: 3715 (MB)
+[2024-01-10 02:39:15,220][inference][INFO] - 	+ Forward pass max memory reserved: 2856 (MB)
+[2024-01-10 02:39:15,220][inference][INFO] - 	+ Forward pass max memory allocated: 2831 (MB)
+[2024-01-10 02:39:15,220][inference][INFO] - 	+ Preparing input for the generation pass
+[2024-01-10 02:39:15,220][backend][INFO] - 	+ Moving inputs tensors to device cuda
+[2024-01-10 02:39:15,220][inference][INFO] - 	+ Tracking generation pass peak memory
+[2024-01-10 02:39:15,220][memory][INFO] - Tracking CUDA devices: [0]
+[2024-01-10 02:39:15,221][memory][INFO] - Tracking Pytorch CUDA devices: [0]
+[2024-01-10 02:39:16,154][isolation][INFO] - Continuously checking only permitted process(es) {73076, 73302} are running on isolated device(s) {0}
+[2024-01-10 02:39:17,090][inference][INFO] - 	+ Generation pass max memory used: 3903 (MB)
+[2024-01-10 02:39:17,090][inference][INFO] - 	+ Generation pass max memory reserved: 3045 (MB)
+[2024-01-10 02:39:17,090][inference][INFO] - 	+ Generation pass max memory allocated: 2999 (MB)
+[2024-01-10 02:39:17,090][inference][INFO] - 	+ Preparing input for the forward pass
+[2024-01-10 02:39:17,090][backend][INFO] - 	+ Moving inputs tensors to device cuda
+[2024-01-10 02:39:17,090][inference][INFO] - 	+ Warming up the forward pass
+[2024-01-10 02:39:17,177][inference][INFO] - 	+ Tracking forward pass latency and throughput
+[2024-01-10 02:39:27,205][inference][INFO] - 	+ Forward pass latency: 8.82e-03 (s)
+[2024-01-10 02:39:27,205][inference][INFO] - 	+ Forward pass throughput: 113.00 (samples/s)
+[2024-01-10 02:39:27,206][inference][INFO] - 	+ Preparing input for the generation pass
+[2024-01-10 02:39:27,206][backend][INFO] - 	+ Moving inputs tensors to device cuda
+[2024-01-10 02:39:27,206][inference][INFO] - 	+ Warming up the generation pass
+[2024-01-10 02:39:28,896][inference][INFO] - 	+ Tracking generation latency and throughput
+[2024-01-10 02:39:39,027][inference][INFO] - 	+ Generation pass latency: 1.69e+00 (s)
+[2024-01-10 02:39:39,027][inference][INFO] - 	+ Generation pass throughput: 151.00 (tokens/s)
+[2024-01-10 02:39:39,027][inference][INFO] - 	+ Preparing input for the forward pass
+[2024-01-10 02:39:39,027][backend][INFO] - 	+ Moving inputs tensors to device cuda
+[2024-01-10 02:39:39,027][inference][INFO] - 	+ Tracking forward pass energy consumption
+[2024-01-10 02:39:53,473][inference][INFO] - 	+ Forward pass energy consumption: 7.84e-07 (kWh/sample)
+[2024-01-10 02:39:53,474][inference][INFO] - 	+ Forward pass carbon emissions: 5.28e-08 (kgCO2eq/sample)
+[2024-01-10 02:39:53,474][inference][INFO] - 	+ Full details in the CodeCarbon report: /workspace/llm-perf/dataset/audace/pytorch+cuda+float16/AI-Sweden-Models/gpt-sw3-1.3b/forward_codecarbon.csv
+[2024-01-10 02:39:53,474][inference][INFO] - 	+ Preparing input for the generation pass
+[2024-01-10 02:39:53,474][backend][INFO] - 	+ Moving inputs tensors to device cuda
+[2024-01-10 02:39:53,474][inference][INFO] - 	+ Tracking generation pass energy consumption
+[2024-01-10 02:40:09,533][inference][INFO] - 	+ Generation pass energy consumption: 3.69e-07 (kWh/token)
+[2024-01-10 02:40:09,533][inference][INFO] - 	+ Generation pass carbon emissions: 2.48e-08 (kgCO2eq/token)
+[2024-01-10 02:40:09,533][inference][INFO] - 	+ Full details in the CodeCarbon report: /workspace/llm-perf/dataset/audace/pytorch+cuda+float16/AI-Sweden-Models/gpt-sw3-1.3b/generate_codecarbon.csv
+[2024-01-10 02:40:09,533][inference][INFO] - Saving results
+[2024-01-10 02:40:09,535][backend][INFO] - Cleaning pytorch backend
+[2024-01-10 02:40:09,535][backend][INFO] - 	+ Terminating isolation process
+[2024-01-10 02:40:09,552][backend][INFO] - 	+ Deleting pretrained model
+[2024-01-10 02:40:09,718][pytorch][INFO] - 	+ Emptying CUDA cache
+[2024-01-10 02:40:09,822][pytorch][INFO] - 	+ Cleaning temporary directory
+[2024-01-10 02:40:10,361][process][INFO] - 	+ Process exited successfully, closing it.

audace/pytorch+cuda+float16/AI-Sweden-Models/gpt-sw3-1.3b/forward_codecarbon.csv

@@ -0,0 +1,2 @@
+timestamp,project_name,run_id,duration,emissions,emissions_rate,cpu_power,gpu_power,ram_power,cpu_energy,gpu_energy,ram_energy,energy_consumed,country_name,country_iso_code,region,cloud_provider,cloud_region,os,python_version,codecarbon_version,cpu_count,cpu_model,gpu_count,gpu_model,longitude,latitude,ram_total_size,tracking_mode,on_cloud,pue
+2024-01-10T02:39:53,codecarbon,f07070fd-42c0-4923-8ea5-f915461a070d,10.036018133163452,6.121168747844664e-05,6.099200566026888e-06,42.5,0.0,0.4853367805480957,0.00011847845878865984,0.0007885197974819846,1.3217834748168874e-06,0.0009083200397454613,France,FRA,île-de-france,,,Linux-5.15.0-91-generic-x86_64-with-glibc2.35,3.10.12,2.3.2,32,AMD Ryzen 9 7950X 16-Core Processor,1,1 x NVIDIA GeForce RTX 4090,2.4075,48.8323,125.53921508789062,process,N,1.0

audace/pytorch+cuda+float16/AI-Sweden-Models/gpt-sw3-1.3b/generate_codecarbon.csv

@@ -0,0 +1,2 @@
+timestamp,project_name,run_id,duration,emissions,emissions_rate,cpu_power,gpu_power,ram_power,cpu_energy,gpu_energy,ram_energy,energy_consumed,country_name,country_iso_code,region,cloud_provider,cloud_region,os,python_version,codecarbon_version,cpu_count,cpu_model,gpu_count,gpu_model,longitude,latitude,ram_total_size,tracking_mode,on_cloud,pue
+2024-01-10T02:40:09,codecarbon,e8ac54f0-ce40-45ee-9dad-429e8e6697a0,11.604247808456421,4.450521381776943e-05,3.8352519312227135e-06,42.5,180.57111186485835,0.4855942726135254,0.00013699179159270395,0.0005218854175080279,1.535520054475607e-06,0.0006604127291552073,France,FRA,île-de-france,,,Linux-5.15.0-91-generic-x86_64-with-glibc2.35,3.10.12,2.3.2,32,AMD Ryzen 9 7950X 16-Core Processor,1,1 x NVIDIA GeForce RTX 4090,2.4075,48.8323,125.53921508789062,process,N,1.0

audace/pytorch+cuda+float16/AI-Sweden-Models/gpt-sw3-1.3b/hydra_config.yaml

@@ -0,0 +1,85 @@
+launcher:
+  name: process
+  _target_: optimum_benchmark.launchers.process.launcher.ProcessLauncher
+  start_method: spawn
+backend:
+  name: pytorch
+  version: 2.1.2+cu118
+  _target_: optimum_benchmark.backends.pytorch.backend.PyTorchBackend
+  seed: 42
+  inter_op_num_threads: null
+  intra_op_num_threads: null
+  continuous_isolation: true
+  isolation_check_interval: 1.0
+  delete_cache: false
+  no_weights: true
+  device_map: null
+  torch_dtype: float16
+  eval_mode: true
+  disable_grad: true
+  amp_autocast: false
+  amp_dtype: null
+  torch_compile: false
+  torch_compile_config: {}
+  to_bettertransformer: false
+  use_flash_attention_2: false
+  quantization_scheme: null
+  quantization_config: {}
+  data_parallel: false
+  deepspeed_inference: false
+  deepspeed_inference_config: {}
+  peft_strategy: null
+  peft_config: {}
+benchmark:
+  name: inference
+  _target_: optimum_benchmark.benchmarks.inference.benchmark.InferenceBenchmark
+  duration: 10
+  warmup_runs: 10
+  memory: true
+  energy: true
+  input_shapes:
+    batch_size: 1
+    sequence_length: 256
+    num_choices: 1
+    feature_size: 80
+    nb_max_frames: 3000
+    audio_sequence_length: 16000
+  new_tokens: 256
+  can_diffuse: false
+  can_generate: true
+  forward_kwargs: {}
+  generate_kwargs:
+    num_return_sequences: 1
+    max_new_tokens: 256
+    min_new_tokens: 256
+    do_sample: false
+    use_cache: true
+    pad_token_id: 0
+    temperature: 1.0
+    num_beams: 1
+experiment_name: pytorch+cuda+float16
+device: cuda
+model: AI-Sweden-Models/gpt-sw3-1.3b
+task: text-generation
+hub_kwargs:
+  revision: main
+  cache_dir: null
+  force_download: false
+  local_files_only: false
+  trust_remote_code: true
+environment:
+  optimum_version: 1.16.1
+  optimum_commit: null
+  transformers_version: 4.36.2
+  transformers_commit: null
+  accelerate_version: 0.25.0
+  accelerate_commit: null
+  diffusers_version: null
+  diffusers_commit: null
+  python_version: 3.10.12
+  system: Linux
+  cpu: ' AMD Ryzen 9 7950X 16-Core Processor'
+  cpu_count: 32
+  cpu_ram_mb: 134796
+  gpus:
+  - NVIDIA GeForce RTX 4090

audace/pytorch+cuda+float16/AI-Sweden-Models/gpt-sw3-1.3b/inference_results.csv

@@ -0,0 +1,2 @@
+forward.latency(s),forward.throughput(samples/s),forward.peak_memory(MB),forward.max_memory_used(MB),forward.max_memory_allocated(MB),forward.max_memory_reserved(MB),forward.energy_consumption(kWh/sample),forward.carbon_emissions(kgCO2eq/sample),generate.latency(s),generate.throughput(tokens/s),decode.latency(s),decode.throughput(tokens/s),generate.peak_memory(MB),generate.max_memory_used(MB),generate.max_memory_allocated(MB),generate.max_memory_reserved(MB),generate.energy_consumption(kWh/token),generate.carbon_emissions(kgCO2eq/token)
+0.00882,113.0,3715,3715,2831,2856,7.84e-07,5.28e-08,1.69,151.0,1.68,152.0,3903,3903,2999,3045,3.69e-07,2.48e-08

audace/pytorch+cuda+float16/BEE-spoke-data/Mixtral-GQA-400m-v2/.hydra/config.yaml

@@ -0,0 +1,76 @@
+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
+  delete_cache: false
+  no_weights: true
+  device_map: null
+  torch_dtype: float16
+  eval_mode: ${is_inference:${benchmark.name}}
+  disable_grad: ${is_inference:${benchmark.name}}
+  amp_autocast: false
+  amp_dtype: null
+  torch_compile: false
+  torch_compile_config: {}
+  to_bettertransformer: false
+  use_flash_attention_2: false
+  quantization_scheme: null
+  quantization_config: {}
+  data_parallel: false
+  deepspeed_inference: false
+  deepspeed_inference_config: {}
+  peft_strategy: null
+  peft_config: {}
+benchmark:
+  name: inference
+  _target_: optimum_benchmark.benchmarks.inference.benchmark.InferenceBenchmark
+  duration: 10
+  warmup_runs: 10
+  memory: true
+  energy: true
+  input_shapes:
+    batch_size: 1
+    sequence_length: 256
+    num_choices: 1
+    feature_size: 80
+    nb_max_frames: 3000
+    audio_sequence_length: 16000
+  new_tokens: 256
+  can_diffuse: ${can_diffuse:${task}}
+  can_generate: ${can_generate:${task}}
+  forward_kwargs: {}
+  generate_kwargs: {}
+launcher:
+  name: process
+  _target_: optimum_benchmark.launchers.process.launcher.ProcessLauncher
+  device_isolation: true
+  start_method: spawn
+experiment_name: pytorch+cuda+float16
+device: cuda
+model: BEE-spoke-data/Mixtral-GQA-400m-v2
+task: ${infer_task:${model}}
+hub_kwargs:
+  revision: main
+  cache_dir: null
+  force_download: false
+  local_files_only: false
+  trust_remote_code: true
+environment:
+  optimum_version: 1.16.1
+  optimum_commit: null
+  transformers_version: 4.36.2
+  transformers_commit: null
+  accelerate_version: 0.26.1
+  accelerate_commit: null
+  diffusers_version: null
+  diffusers_commit: null
+  python_version: 3.10.12
+  system: Linux
+  cpu: ' AMD Ryzen 9 7950X 16-Core Processor'
+  cpu_count: 32
+  cpu_ram_mb: 134796
+  gpus:
+  - NVIDIA GeForce RTX 4090

audace/pytorch+cuda+float16/BEE-spoke-data/Mixtral-GQA-400m-v2/.hydra/hydra.yaml

@@ -0,0 +1,176 @@
+hydra:
+  run:
+    dir: dataset/${oc.env:HOSTNAME}/${experiment_name}/${model}
+  sweep:
+    dir: multirun/${now:%Y-%m-%d}/${now:%H-%M-%S}
+    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: null
+  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: RUN
+  searchpath: []
+  callbacks: {}
+  output_subdir: .hydra
+  overrides:
+    hydra:
+    - hydra.mode=RUN
+    task:
+    - model=BEE-spoke-data/Mixtral-GQA-400m-v2
+  job:
+    name: cli
+    chdir: true
+    override_dirname: model=BEE-spoke-data/Mixtral-GQA-400m-v2
+    id: ???
+    num: ???
+    config_name: pytorch+cuda+float16
+    env_set:
+      COUNTRY_ISO_CODE: FRA
+      OVERRIDE_BENCHMARKS: '0'
+      CUDA_VISIBLE_DEVICES: '0'
+      CUDA_DEVICE_ORDER: PCI_BUS_ID
+    env_copy: []
+    config:
+      override_dirname:
+        kv_sep: '='
+        item_sep: ','
+        exclude_keys: []
+  runtime:
+    version: 1.3.2
+    version_base: '1.3'
+    cwd: /workspace/llm-perf
+    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: /workspace/llm-perf/configs
+      schema: file
+      provider: command-line
+    - path: ''
+      schema: structured
+      provider: schema
+    output_dir: /workspace/llm-perf/dataset/audace/pytorch+cuda+float16/BEE-spoke-data/Mixtral-GQA-400m-v2
+    choices:
+      launcher: process
+      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

audace/pytorch+cuda+float16/BEE-spoke-data/Mixtral-GQA-400m-v2/.hydra/overrides.yaml

@@ -0,0 +1 @@
+- model=BEE-spoke-data/Mixtral-GQA-400m-v2

audace/pytorch+cuda+float16/BEE-spoke-data/Mixtral-GQA-400m-v2/cli.log

@@ -0,0 +1,72 @@
+[2024-01-13 09:46:54,246][launcher][INFO] - Configuring process launcher
+[2024-01-13 09:46:54,246][process][INFO] - Setting multiprocessing start method to spawn.
+[2024-01-13 09:46:54,248][process][INFO] - 	+ Launched worker process with PID 505734.
+[2024-01-13 09:46:54,248][isolation][INFO] - 	+ Launched device(s) isolation process 505735.
+[2024-01-13 09:46:56,106][numexpr.utils][INFO] - Note: NumExpr detected 32 cores but "NUMEXPR_MAX_THREADS" not set, so enforcing safe limit of 8.
+[2024-01-13 09:46:56,106][numexpr.utils][INFO] - NumExpr defaulting to 8 threads.
+[2024-01-13 09:46:56,245][datasets][INFO] - PyTorch version 2.1.2+cu118 available.
+[2024-01-13 09:46:57,449][backend][INFO] - Configuring pytorch backend
+[2024-01-13 09:46:57,449][pytorch][INFO] - 	+ Inferred AutoModel class AutoModelForCausalLM for task text-generation and model_type mixtral
+[2024-01-13 09:46:57,449][pytorch][INFO] - 	+ Disabling gradients
+[2024-01-13 09:46:57,449][pytorch][INFO] - 	+ Loading model with no weights
+[2024-01-13 09:46:57,449][pytorch][INFO] - 	+ Creating no weights model directory
+[2024-01-13 09:46:57,450][pytorch][INFO] - 	+ Saving pretrained config
+[2024-01-13 09:46:57,450][pytorch][INFO] - 	+ Creating no weights model
+[2024-01-13 09:46:57,468][pytorch][INFO] - 	+ Saving no weights model
+[2024-01-13 09:46:57,468][pytorch][INFO] - 	+ Loading no weights model
+[2024-01-13 09:46:57,468][pytorch][INFO] - 	+ Loading model directly on device: cuda
+[2024-01-13 09:46:57,644][pytorch][INFO] - 	+ Randomizing model weights
+[2024-01-13 09:46:57,646][pytorch][INFO] - 	+ Tying model weights after randomization
+[2024-01-13 09:46:57,646][pytorch][INFO] - 	+ Turning on model's eval mode
+[2024-01-13 09:46:57,704][benchmark][INFO] - Configuring inference benchmark
+[2024-01-13 09:46:57,704][inference][INFO] - Running inference benchmark
+[2024-01-13 09:46:57,704][inference][INFO] - 	+ Updating input shapes with model shapes
+[2024-01-13 09:46:57,705][inference][INFO] - 	+ Preparing backend for inference
+[2024-01-13 09:46:57,705][inference][INFO] - 	+ Creating input generator
+[2024-01-13 09:46:57,705][input-generator][INFO] - Using text-generation task generator
+[2024-01-13 09:46:57,705][inference][INFO] - 	+ Preparing input for the forward pass
+[2024-01-13 09:46:57,705][backend][INFO] - 	+ Moving inputs tensors to device cuda
+[2024-01-13 09:46:57,705][inference][INFO] - 	+ Tracking forward pass peak memory
+[2024-01-13 09:46:57,705][memory][INFO] - Tracking CUDA devices: [0]
+[2024-01-13 09:46:57,705][memory][INFO] - Tracking Pytorch CUDA devices: [0]
+[2024-01-13 09:46:58,032][inference][INFO] - 	+ Forward pass max memory used: 4967 (MB)
+[2024-01-13 09:46:58,032][inference][INFO] - 	+ Forward pass max memory reserved: 4108 (MB)
+[2024-01-13 09:46:58,032][inference][INFO] - 	+ Forward pass max memory allocated: 4080 (MB)
+[2024-01-13 09:46:58,032][inference][INFO] - 	+ Preparing input for the generation pass
+[2024-01-13 09:46:58,032][backend][INFO] - 	+ Moving inputs tensors to device cuda
+[2024-01-13 09:46:58,032][inference][INFO] - 	+ Tracking generation pass peak memory
+[2024-01-13 09:46:58,032][memory][INFO] - Tracking CUDA devices: [0]
+[2024-01-13 09:46:58,032][memory][INFO] - Tracking Pytorch CUDA devices: [0]
+[2024-01-13 09:46:59,801][inference][INFO] - 	+ Generation pass max memory used: 4971 (MB)
+[2024-01-13 09:46:59,801][inference][INFO] - 	+ Generation pass max memory reserved: 4112 (MB)
+[2024-01-13 09:46:59,801][inference][INFO] - 	+ Generation pass max memory allocated: 4083 (MB)
+[2024-01-13 09:46:59,801][inference][INFO] - 	+ Preparing input for the forward pass
+[2024-01-13 09:46:59,801][backend][INFO] - 	+ Moving inputs tensors to device cuda
+[2024-01-13 09:46:59,801][inference][INFO] - 	+ Warming up the forward pass
+[2024-01-13 09:46:59,942][inference][INFO] - 	+ Tracking forward pass latency and throughput
+[2024-01-13 09:47:09,967][inference][INFO] - 	+ Forward pass latency: 1.43e-02 (s)
+[2024-01-13 09:47:09,967][inference][INFO] - 	+ Forward pass throughput: 69.90 (samples/s)
+[2024-01-13 09:47:09,967][inference][INFO] - 	+ Preparing input for the generation pass
+[2024-01-13 09:47:09,967][backend][INFO] - 	+ Moving inputs tensors to device cuda
+[2024-01-13 09:47:09,967][inference][INFO] - 	+ Warming up the generation pass
+[2024-01-13 09:47:11,576][inference][INFO] - 	+ Tracking generation latency and throughput
+[2024-01-13 09:47:22,842][inference][INFO] - 	+ Generation pass latency: 1.61e+00 (s)
+[2024-01-13 09:47:22,842][inference][INFO] - 	+ Generation pass throughput: 159.00 (tokens/s)
+[2024-01-13 09:47:22,842][inference][INFO] - 	+ Preparing input for the forward pass
+[2024-01-13 09:47:22,842][backend][INFO] - 	+ Moving inputs tensors to device cuda
+[2024-01-13 09:47:22,842][inference][INFO] - 	+ Tracking forward pass energy consumption
+[2024-01-13 09:47:37,301][inference][INFO] - 	+ Forward pass energy consumption: 8.17e-07 (kWh/sample)
+[2024-01-13 09:47:37,302][inference][INFO] - 	+ Forward pass carbon emissions: 5.51e-08 (kgCO2eq/sample)
+[2024-01-13 09:47:37,302][inference][INFO] - 	+ Full details in the CodeCarbon report: /workspace/llm-perf/dataset/audace/pytorch+cuda+float16/BEE-spoke-data/Mixtral-GQA-400m-v2/forward_codecarbon.csv
+[2024-01-13 09:47:37,302][inference][INFO] - 	+ Preparing input for the generation pass
+[2024-01-13 09:47:37,302][backend][INFO] - 	+ Moving inputs tensors to device cuda
+[2024-01-13 09:47:37,302][inference][INFO] - 	+ Tracking generation pass energy consumption
+[2024-01-13 09:47:51,822][inference][INFO] - 	+ Generation pass energy consumption: 2.62e-07 (kWh/token)
+[2024-01-13 09:47:51,822][inference][INFO] - 	+ Generation pass carbon emissions: 1.77e-08 (kgCO2eq/token)
+[2024-01-13 09:47:51,823][inference][INFO] - 	+ Full details in the CodeCarbon report: /workspace/llm-perf/dataset/audace/pytorch+cuda+float16/BEE-spoke-data/Mixtral-GQA-400m-v2/generate_codecarbon.csv
+[2024-01-13 09:47:51,823][inference][INFO] - Saving results
+[2024-01-13 09:47:51,825][backend][INFO] - Cleaning pytorch backend
+[2024-01-13 09:47:51,825][backend][INFO] - 	+ Deleting pretrained model
+[2024-01-13 09:47:51,913][pytorch][INFO] - 	+ Emptying CUDA cache
+[2024-01-13 09:47:51,995][pytorch][INFO] - 	+ Cleaning temporary directory
+[2024-01-13 09:47:52,520][isolation][INFO] - 	+ Closing device(s) isolation process...

audace/pytorch+cuda+float16/BEE-spoke-data/Mixtral-GQA-400m-v2/forward_codecarbon.csv

@@ -0,0 +1,2 @@
+timestamp,project_name,run_id,duration,emissions,emissions_rate,cpu_power,gpu_power,ram_power,cpu_energy,gpu_energy,ram_energy,energy_consumed,country_name,country_iso_code,region,cloud_provider,cloud_region,os,python_version,codecarbon_version,cpu_count,cpu_model,gpu_count,gpu_model,longitude,latitude,ram_total_size,tracking_mode,on_cloud,pue
+2024-01-13T09:47:37,codecarbon,ec0895b8-f8d9-4554-8fce-4936a9d61ae6,10.03585147857666,3.8655154139889615e-05,3.851706476765427e-06,42.5,0.0,0.3769354820251465,0.00011847589181529153,0.0004541011966139319,1.0266233958191153e-06,0.0005736037118250425,France,FRA,île-de-france,,,Linux-5.15.0-91-generic-x86_64-with-glibc2.35,3.10.12,2.3.2,32,AMD Ryzen 9 7950X 16-Core Processor,1,1 x NVIDIA GeForce RTX 4090,2.4075,48.8323,125.53921508789062,process,N,1.0

audace/pytorch+cuda+float16/BEE-spoke-data/Mixtral-GQA-400m-v2/generate_codecarbon.csv

@@ -0,0 +1,2 @@
+timestamp,project_name,run_id,duration,emissions,emissions_rate,cpu_power,gpu_power,ram_power,cpu_energy,gpu_energy,ram_energy,energy_consumed,country_name,country_iso_code,region,cloud_provider,cloud_region,os,python_version,codecarbon_version,cpu_count,cpu_model,gpu_count,gpu_model,longitude,latitude,ram_total_size,tracking_mode,on_cloud,pue
+2024-01-13T09:47:51,codecarbon,b2ecfb96-40fa-4d69-9444-cf636a2aacf3,10.084577083587646,2.71611266938231e-05,2.6933332423059206e-06,42.5,0.0,0.37764215469360357,0.00011905126257075206,0.00028296467081578847,1.0279409838868029e-06,0.0004030438743704273,France,FRA,île-de-france,,,Linux-5.15.0-91-generic-x86_64-with-glibc2.35,3.10.12,2.3.2,32,AMD Ryzen 9 7950X 16-Core Processor,1,1 x NVIDIA GeForce RTX 4090,2.4075,48.8323,125.53921508789062,process,N,1.0

audace/pytorch+cuda+float16/BEE-spoke-data/Mixtral-GQA-400m-v2/hydra_config.yaml

@@ -0,0 +1,84 @@
+launcher:
+  name: process
+  _target_: optimum_benchmark.launchers.process.launcher.ProcessLauncher
+  device_isolation: true
+  start_method: spawn
+backend:
+  name: pytorch
+  version: 2.1.2+cu118
+  _target_: optimum_benchmark.backends.pytorch.backend.PyTorchBackend
+  seed: 42
+  inter_op_num_threads: null
+  intra_op_num_threads: null
+  delete_cache: false
+  no_weights: true
+  device_map: null
+  torch_dtype: float16
+  eval_mode: true
+  disable_grad: true
+  amp_autocast: false
+  amp_dtype: null
+  torch_compile: false
+  torch_compile_config: {}
+  to_bettertransformer: false
+  use_flash_attention_2: false
+  quantization_scheme: null
+  quantization_config: {}
+  data_parallel: false
+  deepspeed_inference: false
+  deepspeed_inference_config: {}
+  peft_strategy: null
+  peft_config: {}
+benchmark:
+  name: inference
+  _target_: optimum_benchmark.benchmarks.inference.benchmark.InferenceBenchmark
+  duration: 10
+  warmup_runs: 10
+  memory: true
+  energy: true
+  input_shapes:
+    batch_size: 1
+    sequence_length: 256
+    num_choices: 1
+    feature_size: 80
+    nb_max_frames: 3000
+    audio_sequence_length: 16000
+  new_tokens: 256
+  can_diffuse: false
+  can_generate: true
+  forward_kwargs: {}
+  generate_kwargs:
+    num_return_sequences: 1
+    max_new_tokens: 256
+    min_new_tokens: 256
+    do_sample: false
+    use_cache: true
+    pad_token_id: 0
+    temperature: 1.0
+    num_beams: 1
+experiment_name: pytorch+cuda+float16
+device: cuda
+model: BEE-spoke-data/Mixtral-GQA-400m-v2
+task: text-generation
+hub_kwargs:
+  revision: main
+  cache_dir: null
+  force_download: false
+  local_files_only: false
+  trust_remote_code: true
+environment:
+  optimum_version: 1.16.1
+  optimum_commit: null
+  transformers_version: 4.36.2
+  transformers_commit: null
+  accelerate_version: 0.26.1
+  accelerate_commit: null
+  diffusers_version: null
+  diffusers_commit: null
+  python_version: 3.10.12
+  system: Linux
+  cpu: ' AMD Ryzen 9 7950X 16-Core Processor'
+  cpu_count: 32
+  cpu_ram_mb: 134796
+  gpus:
+  - NVIDIA GeForce RTX 4090

audace/pytorch+cuda+float16/BEE-spoke-data/Mixtral-GQA-400m-v2/inference_results.csv

@@ -0,0 +1,2 @@
+forward.latency(s),forward.throughput(samples/s),forward.peak_memory(MB),forward.max_memory_used(MB),forward.max_memory_allocated(MB),forward.max_memory_reserved(MB),forward.energy_consumption(kWh/sample),forward.carbon_emissions(kgCO2eq/sample),generate.latency(s),generate.throughput(tokens/s),decode.latency(s),decode.throughput(tokens/s),generate.peak_memory(MB),generate.max_memory_used(MB),generate.max_memory_allocated(MB),generate.max_memory_reserved(MB),generate.energy_consumption(kWh/token),generate.carbon_emissions(kgCO2eq/token)
+0.0143,69.9,4967,4967,4080,4108,8.17e-07,5.51e-08,1.61,159.0,1.6,159.0,4971,4971,4083,4112,2.62e-07,1.77e-08

audace/pytorch+cuda+float16/BEE-spoke-data/NanoLlama-GQA-L10-A32_KV8-v13-KI/.hydra/config.yaml

@@ -0,0 +1,76 @@
+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
+  delete_cache: false
+  no_weights: true
+  device_map: null
+  torch_dtype: float16
+  eval_mode: ${is_inference:${benchmark.name}}
+  disable_grad: ${is_inference:${benchmark.name}}
+  amp_autocast: false
+  amp_dtype: null
+  torch_compile: false
+  torch_compile_config: {}
+  to_bettertransformer: false
+  use_flash_attention_2: false
+  quantization_scheme: null
+  quantization_config: {}
+  data_parallel: false
+  deepspeed_inference: false
+  deepspeed_inference_config: {}
+  peft_strategy: null
+  peft_config: {}
+benchmark:
+  name: inference
+  _target_: optimum_benchmark.benchmarks.inference.benchmark.InferenceBenchmark
+  duration: 10
+  warmup_runs: 10
+  memory: true
+  energy: true
+  input_shapes:
+    batch_size: 1
+    sequence_length: 256
+    num_choices: 1
+    feature_size: 80
+    nb_max_frames: 3000
+    audio_sequence_length: 16000
+  new_tokens: 256
+  can_diffuse: ${can_diffuse:${task}}
+  can_generate: ${can_generate:${task}}
+  forward_kwargs: {}
+  generate_kwargs: {}
+launcher:
+  name: process
+  _target_: optimum_benchmark.launchers.process.launcher.ProcessLauncher
+  device_isolation: true
+  start_method: spawn
+experiment_name: pytorch+cuda+float16
+device: cuda
+model: BEE-spoke-data/NanoLlama-GQA-L10-A32_KV8-v13-KI
+task: ${infer_task:${model}}
+hub_kwargs:
+  revision: main
+  cache_dir: null
+  force_download: false
+  local_files_only: false
+  trust_remote_code: true
+environment:
+  optimum_version: 1.16.1
+  optimum_commit: null
+  transformers_version: 4.36.2
+  transformers_commit: null
+  accelerate_version: 0.26.1
+  accelerate_commit: null
+  diffusers_version: null
+  diffusers_commit: null
+  python_version: 3.10.12
+  system: Linux
+  cpu: ' AMD Ryzen 9 7950X 16-Core Processor'
+  cpu_count: 32
+  cpu_ram_mb: 134796
+  gpus:
+  - NVIDIA GeForce RTX 4090

audace/pytorch+cuda+float16/BEE-spoke-data/NanoLlama-GQA-L10-A32_KV8-v13-KI/.hydra/hydra.yaml

@@ -0,0 +1,176 @@
+hydra:
+  run:
+    dir: dataset/${oc.env:HOSTNAME}/${experiment_name}/${model}
+  sweep:
+    dir: multirun/${now:%Y-%m-%d}/${now:%H-%M-%S}
+    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: null
+  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: RUN
+  searchpath: []
+  callbacks: {}
+  output_subdir: .hydra
+  overrides:
+    hydra:
+    - hydra.mode=RUN
+    task:
+    - model=BEE-spoke-data/NanoLlama-GQA-L10-A32_KV8-v13-KI
+  job:
+    name: cli
+    chdir: true
+    override_dirname: model=BEE-spoke-data/NanoLlama-GQA-L10-A32_KV8-v13-KI
+    id: ???
+    num: ???
+    config_name: pytorch+cuda+float16
+    env_set:
+      COUNTRY_ISO_CODE: FRA
+      OVERRIDE_BENCHMARKS: '0'
+      CUDA_VISIBLE_DEVICES: '0'
+      CUDA_DEVICE_ORDER: PCI_BUS_ID
+    env_copy: []
+    config:
+      override_dirname:
+        kv_sep: '='
+        item_sep: ','
+        exclude_keys: []
+  runtime:
+    version: 1.3.2
+    version_base: '1.3'
+    cwd: /workspace/llm-perf
+    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: /workspace/llm-perf/configs
+      schema: file
+      provider: command-line
+    - path: ''
+      schema: structured
+      provider: schema
+    output_dir: /workspace/llm-perf/dataset/audace/pytorch+cuda+float16/BEE-spoke-data/NanoLlama-GQA-L10-A32_KV8-v13-KI
+    choices:
+      launcher: process
+      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

audace/pytorch+cuda+float16/BEE-spoke-data/NanoLlama-GQA-L10-A32_KV8-v13-KI/.hydra/overrides.yaml

@@ -0,0 +1 @@
+- model=BEE-spoke-data/NanoLlama-GQA-L10-A32_KV8-v13-KI

audace/pytorch+cuda+float16/BEE-spoke-data/NanoLlama-GQA-L10-A32_KV8-v13-KI/cli.log

@@ -0,0 +1,72 @@
+[2024-01-13 09:57:33,921][launcher][INFO] - Configuring process launcher
+[2024-01-13 09:57:33,921][process][INFO] - Setting multiprocessing start method to spawn.
+[2024-01-13 09:57:33,922][process][INFO] - 	+ Launched worker process with PID 531309.
+[2024-01-13 09:57:33,923][isolation][INFO] - 	+ Launched device(s) isolation process 531310.
+[2024-01-13 09:57:36,017][numexpr.utils][INFO] - Note: NumExpr detected 32 cores but "NUMEXPR_MAX_THREADS" not set, so enforcing safe limit of 8.
+[2024-01-13 09:57:36,017][numexpr.utils][INFO] - NumExpr defaulting to 8 threads.
+[2024-01-13 09:57:36,161][datasets][INFO] - PyTorch version 2.1.2+cu118 available.
+[2024-01-13 09:57:37,523][backend][INFO] - Configuring pytorch backend
+[2024-01-13 09:57:37,523][pytorch][INFO] - 	+ Inferred AutoModel class AutoModelForCausalLM for task text-generation and model_type llama
+[2024-01-13 09:57:37,523][pytorch][INFO] - 	+ Disabling gradients
+[2024-01-13 09:57:37,523][pytorch][INFO] - 	+ Loading model with no weights
+[2024-01-13 09:57:37,523][pytorch][INFO] - 	+ Creating no weights model directory
+[2024-01-13 09:57:37,523][pytorch][INFO] - 	+ Saving pretrained config
+[2024-01-13 09:57:37,524][pytorch][INFO] - 	+ Creating no weights model
+[2024-01-13 09:57:37,524][pytorch][INFO] - 	+ Saving no weights model
+[2024-01-13 09:57:37,525][pytorch][INFO] - 	+ Loading no weights model
+[2024-01-13 09:57:37,525][pytorch][INFO] - 	+ Loading model directly on device: cuda
+[2024-01-13 09:57:37,653][pytorch][INFO] - 	+ Randomizing model weights
+[2024-01-13 09:57:37,653][pytorch][INFO] - 	+ Tying model weights after randomization
+[2024-01-13 09:57:37,654][pytorch][INFO] - 	+ Turning on model's eval mode
+[2024-01-13 09:57:37,713][benchmark][INFO] - Configuring inference benchmark
+[2024-01-13 09:57:37,713][inference][INFO] - Running inference benchmark
+[2024-01-13 09:57:37,713][inference][INFO] - 	+ Updating input shapes with model shapes
+[2024-01-13 09:57:37,713][inference][INFO] - 	+ Preparing backend for inference
+[2024-01-13 09:57:37,713][inference][INFO] - 	+ Creating input generator
+[2024-01-13 09:57:37,713][input-generator][INFO] - Using text-generation task generator
+[2024-01-13 09:57:37,714][inference][INFO] - 	+ Preparing input for the forward pass
+[2024-01-13 09:57:37,714][backend][INFO] - 	+ Moving inputs tensors to device cuda
+[2024-01-13 09:57:37,714][inference][INFO] - 	+ Tracking forward pass peak memory
+[2024-01-13 09:57:37,714][memory][INFO] - Tracking CUDA devices: [0]
+[2024-01-13 09:57:37,714][memory][INFO] - Tracking Pytorch CUDA devices: [0]
+[2024-01-13 09:57:37,866][inference][INFO] - 	+ Forward pass max memory used: 1399 (MB)
+[2024-01-13 09:57:37,866][inference][INFO] - 	+ Forward pass max memory reserved: 541 (MB)
+[2024-01-13 09:57:37,866][inference][INFO] - 	+ Forward pass max memory allocated: 500 (MB)
+[2024-01-13 09:57:37,866][inference][INFO] - 	+ Preparing input for the generation pass
+[2024-01-13 09:57:37,866][backend][INFO] - 	+ Moving inputs tensors to device cuda
+[2024-01-13 09:57:37,867][inference][INFO] - 	+ Tracking generation pass peak memory
+[2024-01-13 09:57:37,867][memory][INFO] - Tracking CUDA devices: [0]
+[2024-01-13 09:57:37,867][memory][INFO] - Tracking Pytorch CUDA devices: [0]
+[2024-01-13 09:57:39,005][inference][INFO] - 	+ Generation pass max memory used: 1408 (MB)
+[2024-01-13 09:57:39,005][inference][INFO] - 	+ Generation pass max memory reserved: 549 (MB)
+[2024-01-13 09:57:39,005][inference][INFO] - 	+ Generation pass max memory allocated: 500 (MB)
+[2024-01-13 09:57:39,006][inference][INFO] - 	+ Preparing input for the forward pass
+[2024-01-13 09:57:39,006][backend][INFO] - 	+ Moving inputs tensors to device cuda
+[2024-01-13 09:57:39,006][inference][INFO] - 	+ Warming up the forward pass
+[2024-01-13 09:57:39,045][inference][INFO] - 	+ Tracking forward pass latency and throughput
+[2024-01-13 09:57:49,104][inference][INFO] - 	+ Forward pass latency: 3.91e-03 (s)
+[2024-01-13 09:57:49,105][inference][INFO] - 	+ Forward pass throughput: 256.00 (samples/s)
+[2024-01-13 09:57:49,105][inference][INFO] - 	+ Preparing input for the generation pass
+[2024-01-13 09:57:49,105][backend][INFO] - 	+ Moving inputs tensors to device cuda
+[2024-01-13 09:57:49,106][inference][INFO] - 	+ Warming up the generation pass
+[2024-01-13 09:57:50,148][inference][INFO] - 	+ Tracking generation latency and throughput
+[2024-01-13 09:58:00,577][inference][INFO] - 	+ Generation pass latency: 1.04e+00 (s)
+[2024-01-13 09:58:00,577][inference][INFO] - 	+ Generation pass throughput: 246.00 (tokens/s)
+[2024-01-13 09:58:00,577][inference][INFO] - 	+ Preparing input for the forward pass
+[2024-01-13 09:58:00,577][backend][INFO] - 	+ Moving inputs tensors to device cuda
+[2024-01-13 09:58:00,578][inference][INFO] - 	+ Tracking forward pass energy consumption
+[2024-01-13 09:58:15,057][inference][INFO] - 	+ Forward pass energy consumption: 2.01e-07 (kWh/sample)
+[2024-01-13 09:58:15,057][inference][INFO] - 	+ Forward pass carbon emissions: 1.36e-08 (kgCO2eq/sample)
+[2024-01-13 09:58:15,057][inference][INFO] - 	+ Full details in the CodeCarbon report: /workspace/llm-perf/dataset/audace/pytorch+cuda+float16/BEE-spoke-data/NanoLlama-GQA-L10-A32_KV8-v13-KI/forward_codecarbon.csv
+[2024-01-13 09:58:15,057][inference][INFO] - 	+ Preparing input for the generation pass
+[2024-01-13 09:58:15,057][backend][INFO] - 	+ Moving inputs tensors to device cuda
+[2024-01-13 09:58:15,057][inference][INFO] - 	+ Tracking generation pass energy consumption
+[2024-01-13 09:58:29,986][inference][INFO] - 	+ Generation pass energy consumption: 1.35e-07 (kWh/token)
+[2024-01-13 09:58:29,987][inference][INFO] - 	+ Generation pass carbon emissions: 9.13e-09 (kgCO2eq/token)
+[2024-01-13 09:58:29,987][inference][INFO] - 	+ Full details in the CodeCarbon report: /workspace/llm-perf/dataset/audace/pytorch+cuda+float16/BEE-spoke-data/NanoLlama-GQA-L10-A32_KV8-v13-KI/generate_codecarbon.csv
+[2024-01-13 09:58:29,987][inference][INFO] - Saving results
+[2024-01-13 09:58:29,991][backend][INFO] - Cleaning pytorch backend
+[2024-01-13 09:58:29,991][backend][INFO] - 	+ Deleting pretrained model
+[2024-01-13 09:58:30,078][pytorch][INFO] - 	+ Emptying CUDA cache
+[2024-01-13 09:58:30,090][pytorch][INFO] - 	+ Cleaning temporary directory
+[2024-01-13 09:58:30,609][isolation][INFO] - 	+ Closing device(s) isolation process...

audace/pytorch+cuda+float16/BEE-spoke-data/NanoLlama-GQA-L10-A32_KV8-v13-KI/forward_codecarbon.csv

@@ -0,0 +1,2 @@
+timestamp,project_name,run_id,duration,emissions,emissions_rate,cpu_power,gpu_power,ram_power,cpu_energy,gpu_energy,ram_energy,energy_consumed,country_name,country_iso_code,region,cloud_provider,cloud_region,os,python_version,codecarbon_version,cpu_count,cpu_model,gpu_count,gpu_model,longitude,latitude,ram_total_size,tracking_mode,on_cloud,pue
+2024-01-13T09:58:15,codecarbon,37c60079-55d9-44d2-8122-e186a25f741a,10.040909051895142,3.5006568124488985e-05,3.486394303898388e-06,42.5,0.0,0.3632984161376953,0.00011853570060597527,0.0003999472644020674,9.79390304154606e-07,0.0005194623553121974,France,FRA,île-de-france,,,Linux-5.15.0-91-generic-x86_64-with-glibc2.35,3.10.12,2.3.2,32,AMD Ryzen 9 7950X 16-Core Processor,1,1 x NVIDIA GeForce RTX 4090,2.4075,48.8323,125.53921508789062,process,N,1.0

audace/pytorch+cuda+float16/BEE-spoke-data/NanoLlama-GQA-L10-A32_KV8-v13-KI/generate_codecarbon.csv

@@ -0,0 +1,2 @@
+timestamp,project_name,run_id,duration,emissions,emissions_rate,cpu_power,gpu_power,ram_power,cpu_energy,gpu_energy,ram_energy,energy_consumed,country_name,country_iso_code,region,cloud_provider,cloud_region,os,python_version,codecarbon_version,cpu_count,cpu_model,gpu_count,gpu_model,longitude,latitude,ram_total_size,tracking_mode,on_cloud,pue
+2024-01-13T09:58:29,codecarbon,23e1cdcd-e776-4b35-b002-45d33aafe2ae,10.495614528656006,2.3375966630948674e-05,2.2272127627330113e-06,42.5,76.65699446370796,0.36365318298339844,0.00012390379359324775,0.00022193545532611658,1.0366470733449283e-06,0.0003468758959927092,France,FRA,île-de-france,,,Linux-5.15.0-91-generic-x86_64-with-glibc2.35,3.10.12,2.3.2,32,AMD Ryzen 9 7950X 16-Core Processor,1,1 x NVIDIA GeForce RTX 4090,2.4075,48.8323,125.53921508789062,process,N,1.0

audace/pytorch+cuda+float16/BEE-spoke-data/NanoLlama-GQA-L10-A32_KV8-v13-KI/hydra_config.yaml

@@ -0,0 +1,84 @@
+launcher:
+  name: process
+  _target_: optimum_benchmark.launchers.process.launcher.ProcessLauncher
+  device_isolation: true
+  start_method: spawn
+backend:
+  name: pytorch
+  version: 2.1.2+cu118
+  _target_: optimum_benchmark.backends.pytorch.backend.PyTorchBackend
+  seed: 42
+  inter_op_num_threads: null
+  intra_op_num_threads: null
+  delete_cache: false
+  no_weights: true
+  device_map: null
+  torch_dtype: float16
+  eval_mode: true
+  disable_grad: true
+  amp_autocast: false
+  amp_dtype: null
+  torch_compile: false
+  torch_compile_config: {}
+  to_bettertransformer: false
+  use_flash_attention_2: false
+  quantization_scheme: null
+  quantization_config: {}
+  data_parallel: false
+  deepspeed_inference: false
+  deepspeed_inference_config: {}
+  peft_strategy: null
+  peft_config: {}
+benchmark:
+  name: inference
+  _target_: optimum_benchmark.benchmarks.inference.benchmark.InferenceBenchmark
+  duration: 10
+  warmup_runs: 10
+  memory: true
+  energy: true
+  input_shapes:
+    batch_size: 1
+    sequence_length: 256
+    num_choices: 1
+    feature_size: 80
+    nb_max_frames: 3000
+    audio_sequence_length: 16000
+  new_tokens: 256
+  can_diffuse: false
+  can_generate: true
+  forward_kwargs: {}
+  generate_kwargs:
+    num_return_sequences: 1
+    max_new_tokens: 256
+    min_new_tokens: 256
+    do_sample: false
+    use_cache: true
+    pad_token_id: 0
+    temperature: 1.0
+    num_beams: 1
+experiment_name: pytorch+cuda+float16
+device: cuda
+model: BEE-spoke-data/NanoLlama-GQA-L10-A32_KV8-v13-KI
+task: text-generation
+hub_kwargs:
+  revision: main
+  cache_dir: null
+  force_download: false
+  local_files_only: false
+  trust_remote_code: true
+environment:
+  optimum_version: 1.16.1
+  optimum_commit: null
+  transformers_version: 4.36.2
+  transformers_commit: null
+  accelerate_version: 0.26.1
+  accelerate_commit: null
+  diffusers_version: null
+  diffusers_commit: null
+  python_version: 3.10.12
+  system: Linux
+  cpu: ' AMD Ryzen 9 7950X 16-Core Processor'
+  cpu_count: 32
+  cpu_ram_mb: 134796
+  gpus:
+  - NVIDIA GeForce RTX 4090

audace/pytorch+cuda+float16/BEE-spoke-data/NanoLlama-GQA-L10-A32_KV8-v13-KI/inference_results.csv

@@ -0,0 +1,2 @@
+forward.latency(s),forward.throughput(samples/s),forward.peak_memory(MB),forward.max_memory_used(MB),forward.max_memory_allocated(MB),forward.max_memory_reserved(MB),forward.energy_consumption(kWh/sample),forward.carbon_emissions(kgCO2eq/sample),generate.latency(s),generate.throughput(tokens/s),decode.latency(s),decode.throughput(tokens/s),generate.peak_memory(MB),generate.max_memory_used(MB),generate.max_memory_allocated(MB),generate.max_memory_reserved(MB),generate.energy_consumption(kWh/token),generate.carbon_emissions(kgCO2eq/token)
+0.00391,256.0,1399,1399,500,541,2.01e-07,1.36e-08,1.04,246.0,1.04,245.0,1408,1408,500,549,1.35e-07,9.13e-09

audace/pytorch+cuda+float16/BEE-spoke-data/smol_llama-101M-GQA/.hydra/config.yaml

@@ -0,0 +1,76 @@
+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
+  delete_cache: false
+  no_weights: true
+  device_map: null
+  torch_dtype: float16
+  eval_mode: ${is_inference:${benchmark.name}}
+  disable_grad: ${is_inference:${benchmark.name}}
+  amp_autocast: false
+  amp_dtype: null
+  torch_compile: false
+  torch_compile_config: {}
+  to_bettertransformer: false
+  use_flash_attention_2: false
+  quantization_scheme: null
+  quantization_config: {}
+  data_parallel: false
+  deepspeed_inference: false
+  deepspeed_inference_config: {}
+  peft_strategy: null
+  peft_config: {}
+benchmark:
+  name: inference
+  _target_: optimum_benchmark.benchmarks.inference.benchmark.InferenceBenchmark
+  duration: 10
+  warmup_runs: 10
+  memory: true
+  energy: true
+  input_shapes:
+    batch_size: 1
+    sequence_length: 256
+    num_choices: 1
+    feature_size: 80
+    nb_max_frames: 3000
+    audio_sequence_length: 16000
+  new_tokens: 256
+  can_diffuse: ${can_diffuse:${task}}
+  can_generate: ${can_generate:${task}}
+  forward_kwargs: {}
+  generate_kwargs: {}
+launcher:
+  name: process
+  _target_: optimum_benchmark.launchers.process.launcher.ProcessLauncher
+  device_isolation: true
+  start_method: spawn
+experiment_name: pytorch+cuda+float16
+device: cuda
+model: BEE-spoke-data/smol_llama-101M-GQA
+task: ${infer_task:${model}}
+hub_kwargs:
+  revision: main
+  cache_dir: null
+  force_download: false
+  local_files_only: false
+  trust_remote_code: true
+environment:
+  optimum_version: 1.16.1
+  optimum_commit: null
+  transformers_version: 4.36.2
+  transformers_commit: null
+  accelerate_version: 0.26.1
+  accelerate_commit: null
+  diffusers_version: null
+  diffusers_commit: null
+  python_version: 3.10.12
+  system: Linux
+  cpu: ' AMD Ryzen 9 7950X 16-Core Processor'
+  cpu_count: 32
+  cpu_ram_mb: 134796
+  gpus:
+  - NVIDIA GeForce RTX 4090

audace/pytorch+cuda+float16/BEE-spoke-data/smol_llama-101M-GQA/.hydra/hydra.yaml

@@ -0,0 +1,176 @@
+hydra:
+  run:
+    dir: dataset/${oc.env:HOSTNAME}/${experiment_name}/${model}
+  sweep:
+    dir: multirun/${now:%Y-%m-%d}/${now:%H-%M-%S}
+    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: null
+  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: RUN
+  searchpath: []
+  callbacks: {}
+  output_subdir: .hydra
+  overrides:
+    hydra:
+    - hydra.mode=RUN
+    task:
+    - model=BEE-spoke-data/smol_llama-101M-GQA
+  job:
+    name: cli
+    chdir: true
+    override_dirname: model=BEE-spoke-data/smol_llama-101M-GQA
+    id: ???
+    num: ???
+    config_name: pytorch+cuda+float16
+    env_set:
+      COUNTRY_ISO_CODE: FRA
+      OVERRIDE_BENCHMARKS: '0'
+      CUDA_VISIBLE_DEVICES: '0'
+      CUDA_DEVICE_ORDER: PCI_BUS_ID
+    env_copy: []
+    config:
+      override_dirname:
+        kv_sep: '='
+        item_sep: ','
+        exclude_keys: []
+  runtime:
+    version: 1.3.2
+    version_base: '1.3'
+    cwd: /workspace/llm-perf
+    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: /workspace/llm-perf/configs
+      schema: file
+      provider: command-line
+    - path: ''
+      schema: structured
+      provider: schema
+    output_dir: /workspace/llm-perf/dataset/audace/pytorch+cuda+float16/BEE-spoke-data/smol_llama-101M-GQA
+    choices:
+      launcher: process
+      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

audace/pytorch+cuda+float16/BEE-spoke-data/smol_llama-101M-GQA/.hydra/overrides.yaml

@@ -0,0 +1 @@
+- model=BEE-spoke-data/smol_llama-101M-GQA

audace/pytorch+cuda+float16/BEE-spoke-data/smol_llama-101M-GQA/cli.log

@@ -0,0 +1,72 @@
+[2024-01-13 09:52:40,736][launcher][INFO] - Configuring process launcher
+[2024-01-13 09:52:40,737][process][INFO] - Setting multiprocessing start method to spawn.
+[2024-01-13 09:52:40,738][process][INFO] - 	+ Launched worker process with PID 519384.
+[2024-01-13 09:52:40,738][isolation][INFO] - 	+ Launched device(s) isolation process 519385.
+[2024-01-13 09:52:42,734][numexpr.utils][INFO] - Note: NumExpr detected 32 cores but "NUMEXPR_MAX_THREADS" not set, so enforcing safe limit of 8.
+[2024-01-13 09:52:42,734][numexpr.utils][INFO] - NumExpr defaulting to 8 threads.
+[2024-01-13 09:52:42,872][datasets][INFO] - PyTorch version 2.1.2+cu118 available.
+[2024-01-13 09:52:44,056][backend][INFO] - Configuring pytorch backend
+[2024-01-13 09:52:44,056][pytorch][INFO] - 	+ Inferred AutoModel class AutoModelForCausalLM for task text-generation and model_type llama
+[2024-01-13 09:52:44,056][pytorch][INFO] - 	+ Disabling gradients
+[2024-01-13 09:52:44,057][pytorch][INFO] - 	+ Loading model with no weights
+[2024-01-13 09:52:44,057][pytorch][INFO] - 	+ Creating no weights model directory
+[2024-01-13 09:52:44,057][pytorch][INFO] - 	+ Saving pretrained config
+[2024-01-13 09:52:44,057][pytorch][INFO] - 	+ Creating no weights model
+[2024-01-13 09:52:44,058][pytorch][INFO] - 	+ Saving no weights model
+[2024-01-13 09:52:44,058][pytorch][INFO] - 	+ Loading no weights model
+[2024-01-13 09:52:44,058][pytorch][INFO] - 	+ Loading model directly on device: cuda
+[2024-01-13 09:52:44,184][pytorch][INFO] - 	+ Randomizing model weights
+[2024-01-13 09:52:44,184][pytorch][INFO] - 	+ Tying model weights after randomization
+[2024-01-13 09:52:44,185][pytorch][INFO] - 	+ Turning on model's eval mode
+[2024-01-13 09:52:44,244][benchmark][INFO] - Configuring inference benchmark
+[2024-01-13 09:52:44,244][inference][INFO] - Running inference benchmark
+[2024-01-13 09:52:44,244][inference][INFO] - 	+ Updating input shapes with model shapes
+[2024-01-13 09:52:44,244][inference][INFO] - 	+ Preparing backend for inference
+[2024-01-13 09:52:44,244][inference][INFO] - 	+ Creating input generator
+[2024-01-13 09:52:44,244][input-generator][INFO] - Using text-generation task generator
+[2024-01-13 09:52:44,245][inference][INFO] - 	+ Preparing input for the forward pass
+[2024-01-13 09:52:44,245][backend][INFO] - 	+ Moving inputs tensors to device cuda
+[2024-01-13 09:52:44,245][inference][INFO] - 	+ Tracking forward pass peak memory
+[2024-01-13 09:52:44,245][memory][INFO] - Tracking CUDA devices: [0]
+[2024-01-13 09:52:44,245][memory][INFO] - Tracking Pytorch CUDA devices: [0]
+[2024-01-13 09:52:44,395][inference][INFO] - 	+ Forward pass max memory used: 1148 (MB)
+[2024-01-13 09:52:44,395][inference][INFO] - 	+ Forward pass max memory reserved: 289 (MB)
+[2024-01-13 09:52:44,396][inference][INFO] - 	+ Forward pass max memory allocated: 269 (MB)
+[2024-01-13 09:52:44,396][inference][INFO] - 	+ Preparing input for the generation pass
+[2024-01-13 09:52:44,396][backend][INFO] - 	+ Moving inputs tensors to device cuda
+[2024-01-13 09:52:44,396][inference][INFO] - 	+ Tracking generation pass peak memory
+[2024-01-13 09:52:44,396][memory][INFO] - Tracking CUDA devices: [0]
+[2024-01-13 09:52:44,396][memory][INFO] - Tracking Pytorch CUDA devices: [0]
+[2024-01-13 09:52:45,179][inference][INFO] - 	+ Generation pass max memory used: 1152 (MB)
+[2024-01-13 09:52:45,179][inference][INFO] - 	+ Generation pass max memory reserved: 293 (MB)
+[2024-01-13 09:52:45,179][inference][INFO] - 	+ Generation pass max memory allocated: 269 (MB)
+[2024-01-13 09:52:45,180][inference][INFO] - 	+ Preparing input for the forward pass
+[2024-01-13 09:52:45,180][backend][INFO] - 	+ Moving inputs tensors to device cuda
+[2024-01-13 09:52:45,180][inference][INFO] - 	+ Warming up the forward pass
+[2024-01-13 09:52:45,202][inference][INFO] - 	+ Tracking forward pass latency and throughput
+[2024-01-13 09:52:55,318][inference][INFO] - 	+ Forward pass latency: 2.32e-03 (s)
+[2024-01-13 09:52:55,320][inference][INFO] - 	+ Forward pass throughput: 431.00 (samples/s)
+[2024-01-13 09:52:55,320][inference][INFO] - 	+ Preparing input for the generation pass
+[2024-01-13 09:52:55,320][backend][INFO] - 	+ Moving inputs tensors to device cuda
+[2024-01-13 09:52:55,320][inference][INFO] - 	+ Warming up the generation pass
+[2024-01-13 09:52:55,956][inference][INFO] - 	+ Tracking generation latency and throughput
+[2024-01-13 09:53:06,162][inference][INFO] - 	+ Generation pass latency: 6.38e-01 (s)
+[2024-01-13 09:53:06,163][inference][INFO] - 	+ Generation pass throughput: 401.00 (tokens/s)
+[2024-01-13 09:53:06,163][inference][INFO] - 	+ Preparing input for the forward pass
+[2024-01-13 09:53:06,163][backend][INFO] - 	+ Moving inputs tensors to device cuda
+[2024-01-13 09:53:06,163][inference][INFO] - 	+ Tracking forward pass energy consumption
+[2024-01-13 09:53:20,645][inference][INFO] - 	+ Forward pass energy consumption: 1.12e-07 (kWh/sample)
+[2024-01-13 09:53:20,646][inference][INFO] - 	+ Forward pass carbon emissions: 7.52e-09 (kgCO2eq/sample)
+[2024-01-13 09:53:20,646][inference][INFO] - 	+ Full details in the CodeCarbon report: /workspace/llm-perf/dataset/audace/pytorch+cuda+float16/BEE-spoke-data/smol_llama-101M-GQA/forward_codecarbon.csv
+[2024-01-13 09:53:20,646][inference][INFO] - 	+ Preparing input for the generation pass
+[2024-01-13 09:53:20,646][backend][INFO] - 	+ Moving inputs tensors to device cuda
+[2024-01-13 09:53:20,646][inference][INFO] - 	+ Tracking generation pass energy consumption
+[2024-01-13 09:53:35,169][inference][INFO] - 	+ Generation pass energy consumption: 7.81e-08 (kWh/token)
+[2024-01-13 09:53:35,169][inference][INFO] - 	+ Generation pass carbon emissions: 5.26e-09 (kgCO2eq/token)
+[2024-01-13 09:53:35,169][inference][INFO] - 	+ Full details in the CodeCarbon report: /workspace/llm-perf/dataset/audace/pytorch+cuda+float16/BEE-spoke-data/smol_llama-101M-GQA/generate_codecarbon.csv
+[2024-01-13 09:53:35,169][inference][INFO] - Saving results
+[2024-01-13 09:53:35,176][backend][INFO] - Cleaning pytorch backend
+[2024-01-13 09:53:35,176][backend][INFO] - 	+ Deleting pretrained model
+[2024-01-13 09:53:35,265][pytorch][INFO] - 	+ Emptying CUDA cache
+[2024-01-13 09:53:35,272][pytorch][INFO] - 	+ Cleaning temporary directory
+[2024-01-13 09:53:35,808][isolation][INFO] - 	+ Closing device(s) isolation process...

audace/pytorch+cuda+float16/BEE-spoke-data/smol_llama-101M-GQA/forward_codecarbon.csv

@@ -0,0 +1,2 @@
+timestamp,project_name,run_id,duration,emissions,emissions_rate,cpu_power,gpu_power,ram_power,cpu_energy,gpu_energy,ram_energy,energy_consumed,country_name,country_iso_code,region,cloud_provider,cloud_region,os,python_version,codecarbon_version,cpu_count,cpu_model,gpu_count,gpu_model,longitude,latitude,ram_total_size,tracking_mode,on_cloud,pue
+2024-01-13T09:53:20,codecarbon,e731a8b9-d0f2-47e5-b0ad-850796786950,10.037946939468384,3.16057820516264e-05,3.1486301175148733e-06,42.5,0.0,0.35999536514282227,0.00011850104083617528,0.00034951833516982944,9.787253684780238e-07,0.0004689981013744828,France,FRA,île-de-france,,,Linux-5.15.0-91-generic-x86_64-with-glibc2.35,3.10.12,2.3.2,32,AMD Ryzen 9 7950X 16-Core Processor,1,1 x NVIDIA GeForce RTX 4090,2.4075,48.8323,125.53921508789062,process,N,1.0

audace/pytorch+cuda+float16/BEE-spoke-data/smol_llama-101M-GQA/generate_codecarbon.csv

@@ -0,0 +1,2 @@
+timestamp,project_name,run_id,duration,emissions,emissions_rate,cpu_power,gpu_power,ram_power,cpu_energy,gpu_energy,ram_energy,energy_consumed,country_name,country_iso_code,region,cloud_provider,cloud_region,os,python_version,codecarbon_version,cpu_count,cpu_model,gpu_count,gpu_model,longitude,latitude,ram_total_size,tracking_mode,on_cloud,pue
+2024-01-13T09:53:35,codecarbon,ac524719-6154-4313-9713-59103e2466cf,10.130835056304932,2.0217237487941266e-05,1.995614120215989e-06,42.5,56.63061379693696,0.3599996566772461,0.00011959752688805264,0.0001794204213139139,9.855758808537492e-07,0.00030000352408282035,France,FRA,île-de-france,,,Linux-5.15.0-91-generic-x86_64-with-glibc2.35,3.10.12,2.3.2,32,AMD Ryzen 9 7950X 16-Core Processor,1,1 x NVIDIA GeForce RTX 4090,2.4075,48.8323,125.53921508789062,process,N,1.0

audace/pytorch+cuda+float16/BEE-spoke-data/smol_llama-101M-GQA/hydra_config.yaml

@@ -0,0 +1,84 @@
+launcher:
+  name: process
+  _target_: optimum_benchmark.launchers.process.launcher.ProcessLauncher
+  device_isolation: true
+  start_method: spawn
+backend:
+  name: pytorch
+  version: 2.1.2+cu118
+  _target_: optimum_benchmark.backends.pytorch.backend.PyTorchBackend
+  seed: 42
+  inter_op_num_threads: null
+  intra_op_num_threads: null
+  delete_cache: false
+  no_weights: true
+  device_map: null
+  torch_dtype: float16
+  eval_mode: true
+  disable_grad: true
+  amp_autocast: false
+  amp_dtype: null
+  torch_compile: false
+  torch_compile_config: {}
+  to_bettertransformer: false
+  use_flash_attention_2: false
+  quantization_scheme: null
+  quantization_config: {}
+  data_parallel: false
+  deepspeed_inference: false
+  deepspeed_inference_config: {}
+  peft_strategy: null
+  peft_config: {}
+benchmark:
+  name: inference
+  _target_: optimum_benchmark.benchmarks.inference.benchmark.InferenceBenchmark
+  duration: 10
+  warmup_runs: 10
+  memory: true
+  energy: true
+  input_shapes:
+    batch_size: 1
+    sequence_length: 256
+    num_choices: 1
+    feature_size: 80
+    nb_max_frames: 3000
+    audio_sequence_length: 16000
+  new_tokens: 256
+  can_diffuse: false
+  can_generate: true
+  forward_kwargs: {}
+  generate_kwargs:
+    num_return_sequences: 1
+    max_new_tokens: 256
+    min_new_tokens: 256
+    do_sample: false
+    use_cache: true
+    pad_token_id: 0
+    temperature: 1.0
+    num_beams: 1
+experiment_name: pytorch+cuda+float16
+device: cuda
+model: BEE-spoke-data/smol_llama-101M-GQA
+task: text-generation
+hub_kwargs:
+  revision: main
+  cache_dir: null
+  force_download: false
+  local_files_only: false
+  trust_remote_code: true
+environment:
+  optimum_version: 1.16.1
+  optimum_commit: null
+  transformers_version: 4.36.2
+  transformers_commit: null
+  accelerate_version: 0.26.1
+  accelerate_commit: null
+  diffusers_version: null
+  diffusers_commit: null
+  python_version: 3.10.12
+  system: Linux
+  cpu: ' AMD Ryzen 9 7950X 16-Core Processor'
+  cpu_count: 32
+  cpu_ram_mb: 134796
+  gpus:
+  - NVIDIA GeForce RTX 4090

audace/pytorch+cuda+float16/BEE-spoke-data/smol_llama-101M-GQA/inference_results.csv

@@ -0,0 +1,2 @@
+forward.latency(s),forward.throughput(samples/s),forward.peak_memory(MB),forward.max_memory_used(MB),forward.max_memory_allocated(MB),forward.max_memory_reserved(MB),forward.energy_consumption(kWh/sample),forward.carbon_emissions(kgCO2eq/sample),generate.latency(s),generate.throughput(tokens/s),decode.latency(s),decode.throughput(tokens/s),generate.peak_memory(MB),generate.max_memory_used(MB),generate.max_memory_allocated(MB),generate.max_memory_reserved(MB),generate.energy_consumption(kWh/token),generate.carbon_emissions(kgCO2eq/token)
+0.00232,431.0,1148,1148,269,289,1.12e-07,7.52e-09,0.638,401.0,0.636,401.0,1152,1152,269,293,7.81e-08,5.26e-09

audace/pytorch+cuda+float16/BEE-spoke-data/smol_llama-220M-GQA/.hydra/config.yaml

@@ -0,0 +1,76 @@
+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
+  delete_cache: false
+  no_weights: true
+  device_map: null
+  torch_dtype: float16
+  eval_mode: ${is_inference:${benchmark.name}}
+  disable_grad: ${is_inference:${benchmark.name}}
+  amp_autocast: false
+  amp_dtype: null
+  torch_compile: false
+  torch_compile_config: {}
+  to_bettertransformer: false
+  use_flash_attention_2: false
+  quantization_scheme: null
+  quantization_config: {}
+  data_parallel: false
+  deepspeed_inference: false
+  deepspeed_inference_config: {}
+  peft_strategy: null
+  peft_config: {}
+benchmark:
+  name: inference
+  _target_: optimum_benchmark.benchmarks.inference.benchmark.InferenceBenchmark
+  duration: 10
+  warmup_runs: 10
+  memory: true
+  energy: true
+  input_shapes:
+    batch_size: 1
+    sequence_length: 256
+    num_choices: 1
+    feature_size: 80
+    nb_max_frames: 3000
+    audio_sequence_length: 16000
+  new_tokens: 256
+  can_diffuse: ${can_diffuse:${task}}
+  can_generate: ${can_generate:${task}}
+  forward_kwargs: {}
+  generate_kwargs: {}
+launcher:
+  name: process
+  _target_: optimum_benchmark.launchers.process.launcher.ProcessLauncher
+  device_isolation: true
+  start_method: spawn
+experiment_name: pytorch+cuda+float16
+device: cuda
+model: BEE-spoke-data/smol_llama-220M-GQA
+task: ${infer_task:${model}}
+hub_kwargs:
+  revision: main
+  cache_dir: null
+  force_download: false
+  local_files_only: false
+  trust_remote_code: true
+environment:
+  optimum_version: 1.16.1
+  optimum_commit: null
+  transformers_version: 4.36.2
+  transformers_commit: null
+  accelerate_version: 0.26.1
+  accelerate_commit: null
+  diffusers_version: null
+  diffusers_commit: null
+  python_version: 3.10.12
+  system: Linux
+  cpu: ' AMD Ryzen 9 7950X 16-Core Processor'
+  cpu_count: 32
+  cpu_ram_mb: 134796
+  gpus:
+  - NVIDIA GeForce RTX 4090

audace/pytorch+cuda+float16/BEE-spoke-data/smol_llama-220M-GQA/.hydra/hydra.yaml

@@ -0,0 +1,176 @@
+hydra:
+  run:
+    dir: dataset/${oc.env:HOSTNAME}/${experiment_name}/${model}
+  sweep:
+    dir: multirun/${now:%Y-%m-%d}/${now:%H-%M-%S}
+    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: null
+  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: RUN
+  searchpath: []
+  callbacks: {}
+  output_subdir: .hydra
+  overrides:
+    hydra:
+    - hydra.mode=RUN
+    task:
+    - model=BEE-spoke-data/smol_llama-220M-GQA
+  job:
+    name: cli
+    chdir: true
+    override_dirname: model=BEE-spoke-data/smol_llama-220M-GQA
+    id: ???
+    num: ???
+    config_name: pytorch+cuda+float16
+    env_set:
+      COUNTRY_ISO_CODE: FRA
+      OVERRIDE_BENCHMARKS: '0'
+      CUDA_VISIBLE_DEVICES: '0'
+      CUDA_DEVICE_ORDER: PCI_BUS_ID
+    env_copy: []
+    config:
+      override_dirname:
+        kv_sep: '='
+        item_sep: ','
+        exclude_keys: []
+  runtime:
+    version: 1.3.2
+    version_base: '1.3'
+    cwd: /workspace/llm-perf
+    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: /workspace/llm-perf/configs
+      schema: file
+      provider: command-line
+    - path: ''
+      schema: structured
+      provider: schema
+    output_dir: /workspace/llm-perf/dataset/audace/pytorch+cuda+float16/BEE-spoke-data/smol_llama-220M-GQA
+    choices:
+      launcher: process
+      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

audace/pytorch+cuda+float16/BEE-spoke-data/smol_llama-220M-GQA/.hydra/overrides.yaml

@@ -0,0 +1 @@
+- model=BEE-spoke-data/smol_llama-220M-GQA

audace/pytorch+cuda+float16/BEE-spoke-data/smol_llama-220M-GQA/cli.log

@@ -0,0 +1,72 @@
+[2024-01-13 10:02:36,957][launcher][INFO] - Configuring process launcher
+[2024-01-13 10:02:36,958][process][INFO] - Setting multiprocessing start method to spawn.
+[2024-01-13 10:02:36,959][process][INFO] - 	+ Launched worker process with PID 544789.
+[2024-01-13 10:02:36,959][isolation][INFO] - 	+ Launched device(s) isolation process 544790.
+[2024-01-13 10:02:38,851][numexpr.utils][INFO] - Note: NumExpr detected 32 cores but "NUMEXPR_MAX_THREADS" not set, so enforcing safe limit of 8.
+[2024-01-13 10:02:38,851][numexpr.utils][INFO] - NumExpr defaulting to 8 threads.
+[2024-01-13 10:02:38,988][datasets][INFO] - PyTorch version 2.1.2+cu118 available.
+[2024-01-13 10:02:40,362][backend][INFO] - Configuring pytorch backend
+[2024-01-13 10:02:40,363][pytorch][INFO] - 	+ Inferred AutoModel class AutoModelForCausalLM for task text-generation and model_type llama
+[2024-01-13 10:02:40,363][pytorch][INFO] - 	+ Disabling gradients
+[2024-01-13 10:02:40,363][pytorch][INFO] - 	+ Loading model with no weights
+[2024-01-13 10:02:40,363][pytorch][INFO] - 	+ Creating no weights model directory
+[2024-01-13 10:02:40,363][pytorch][INFO] - 	+ Saving pretrained config
+[2024-01-13 10:02:40,364][pytorch][INFO] - 	+ Creating no weights model
+[2024-01-13 10:02:40,364][pytorch][INFO] - 	+ Saving no weights model
+[2024-01-13 10:02:40,365][pytorch][INFO] - 	+ Loading no weights model
+[2024-01-13 10:02:40,365][pytorch][INFO] - 	+ Loading model directly on device: cuda
+[2024-01-13 10:02:40,498][pytorch][INFO] - 	+ Randomizing model weights
+[2024-01-13 10:02:40,499][pytorch][INFO] - 	+ Tying model weights after randomization
+[2024-01-13 10:02:40,500][pytorch][INFO] - 	+ Turning on model's eval mode
+[2024-01-13 10:02:40,559][benchmark][INFO] - Configuring inference benchmark
+[2024-01-13 10:02:40,559][inference][INFO] - Running inference benchmark
+[2024-01-13 10:02:40,559][inference][INFO] - 	+ Updating input shapes with model shapes
+[2024-01-13 10:02:40,559][inference][INFO] - 	+ Preparing backend for inference
+[2024-01-13 10:02:40,559][inference][INFO] - 	+ Creating input generator
+[2024-01-13 10:02:40,559][input-generator][INFO] - Using text-generation task generator
+[2024-01-13 10:02:40,560][inference][INFO] - 	+ Preparing input for the forward pass
+[2024-01-13 10:02:40,560][backend][INFO] - 	+ Moving inputs tensors to device cuda
+[2024-01-13 10:02:40,560][inference][INFO] - 	+ Tracking forward pass peak memory
+[2024-01-13 10:02:40,560][memory][INFO] - Tracking CUDA devices: [0]
+[2024-01-13 10:02:40,560][memory][INFO] - Tracking Pytorch CUDA devices: [0]
+[2024-01-13 10:02:40,804][inference][INFO] - 	+ Forward pass max memory used: 1397 (MB)
+[2024-01-13 10:02:40,804][inference][INFO] - 	+ Forward pass max memory reserved: 538 (MB)
+[2024-01-13 10:02:40,805][inference][INFO] - 	+ Forward pass max memory allocated: 497 (MB)
+[2024-01-13 10:02:40,805][inference][INFO] - 	+ Preparing input for the generation pass
+[2024-01-13 10:02:40,805][backend][INFO] - 	+ Moving inputs tensors to device cuda
+[2024-01-13 10:02:40,805][inference][INFO] - 	+ Tracking generation pass peak memory
+[2024-01-13 10:02:40,805][memory][INFO] - Tracking CUDA devices: [0]
+[2024-01-13 10:02:40,805][memory][INFO] - Tracking Pytorch CUDA devices: [0]
+[2024-01-13 10:02:42,011][inference][INFO] - 	+ Generation pass max memory used: 1408 (MB)
+[2024-01-13 10:02:42,011][inference][INFO] - 	+ Generation pass max memory reserved: 549 (MB)
+[2024-01-13 10:02:42,011][inference][INFO] - 	+ Generation pass max memory allocated: 500 (MB)
+[2024-01-13 10:02:42,012][inference][INFO] - 	+ Preparing input for the forward pass
+[2024-01-13 10:02:42,012][backend][INFO] - 	+ Moving inputs tensors to device cuda
+[2024-01-13 10:02:42,012][inference][INFO] - 	+ Warming up the forward pass
+[2024-01-13 10:02:42,051][inference][INFO] - 	+ Tracking forward pass latency and throughput
+[2024-01-13 10:02:52,114][inference][INFO] - 	+ Forward pass latency: 3.88e-03 (s)
+[2024-01-13 10:02:52,115][inference][INFO] - 	+ Forward pass throughput: 258.00 (samples/s)
+[2024-01-13 10:02:52,116][inference][INFO] - 	+ Preparing input for the generation pass
+[2024-01-13 10:02:52,116][backend][INFO] - 	+ Moving inputs tensors to device cuda
+[2024-01-13 10:02:52,116][inference][INFO] - 	+ Warming up the generation pass
+[2024-01-13 10:02:53,161][inference][INFO] - 	+ Tracking generation latency and throughput
+[2024-01-13 10:03:03,606][inference][INFO] - 	+ Generation pass latency: 1.04e+00 (s)
+[2024-01-13 10:03:03,607][inference][INFO] - 	+ Generation pass throughput: 246.00 (tokens/s)
+[2024-01-13 10:03:03,607][inference][INFO] - 	+ Preparing input for the forward pass
+[2024-01-13 10:03:03,607][backend][INFO] - 	+ Moving inputs tensors to device cuda
+[2024-01-13 10:03:03,607][inference][INFO] - 	+ Tracking forward pass energy consumption
+[2024-01-13 10:03:18,072][inference][INFO] - 	+ Forward pass energy consumption: 1.96e-07 (kWh/sample)
+[2024-01-13 10:03:18,072][inference][INFO] - 	+ Forward pass carbon emissions: 1.32e-08 (kgCO2eq/sample)
+[2024-01-13 10:03:18,072][inference][INFO] - 	+ Full details in the CodeCarbon report: /workspace/llm-perf/dataset/audace/pytorch+cuda+float16/BEE-spoke-data/smol_llama-220M-GQA/forward_codecarbon.csv
+[2024-01-13 10:03:18,072][inference][INFO] - 	+ Preparing input for the generation pass
+[2024-01-13 10:03:18,072][backend][INFO] - 	+ Moving inputs tensors to device cuda
+[2024-01-13 10:03:18,072][inference][INFO] - 	+ Tracking generation pass energy consumption
+[2024-01-13 10:03:32,915][inference][INFO] - 	+ Generation pass energy consumption: 1.36e-07 (kWh/token)
+[2024-01-13 10:03:32,915][inference][INFO] - 	+ Generation pass carbon emissions: 9.18e-09 (kgCO2eq/token)
+[2024-01-13 10:03:32,915][inference][INFO] - 	+ Full details in the CodeCarbon report: /workspace/llm-perf/dataset/audace/pytorch+cuda+float16/BEE-spoke-data/smol_llama-220M-GQA/generate_codecarbon.csv
+[2024-01-13 10:03:32,915][inference][INFO] - Saving results
+[2024-01-13 10:03:32,920][backend][INFO] - Cleaning pytorch backend
+[2024-01-13 10:03:32,920][backend][INFO] - 	+ Deleting pretrained model
+[2024-01-13 10:03:33,009][pytorch][INFO] - 	+ Emptying CUDA cache
+[2024-01-13 10:03:33,021][pytorch][INFO] - 	+ Cleaning temporary directory
+[2024-01-13 10:03:33,524][isolation][INFO] - 	+ Closing device(s) isolation process...

audace/pytorch+cuda+float16/BEE-spoke-data/smol_llama-220M-GQA/forward_codecarbon.csv

@@ -0,0 +1,2 @@
+timestamp,project_name,run_id,duration,emissions,emissions_rate,cpu_power,gpu_power,ram_power,cpu_energy,gpu_energy,ram_energy,energy_consumed,country_name,country_iso_code,region,cloud_provider,cloud_region,os,python_version,codecarbon_version,cpu_count,cpu_model,gpu_count,gpu_model,longitude,latitude,ram_total_size,tracking_mode,on_cloud,pue
+2024-01-13T10:03:18,codecarbon,33df1861-3592-49c0-b3b6-984d36e07e43,10.035578966140747,3.502802899658431e-05,3.490384472561685e-06,42.5,0.0,0.35994815826416016,0.00011847292797433006,0.0004003289313736502,9.789537783635183e-07,0.0005197808131263438,France,FRA,île-de-france,,,Linux-5.15.0-91-generic-x86_64-with-glibc2.35,3.10.12,2.3.2,32,AMD Ryzen 9 7950X 16-Core Processor,1,1 x NVIDIA GeForce RTX 4090,2.4075,48.8323,125.53921508789062,process,N,1.0

audace/pytorch+cuda+float16/BEE-spoke-data/smol_llama-220M-GQA/generate_codecarbon.csv

@@ -0,0 +1,2 @@
+timestamp,project_name,run_id,duration,emissions,emissions_rate,cpu_power,gpu_power,ram_power,cpu_energy,gpu_energy,ram_energy,energy_consumed,country_name,country_iso_code,region,cloud_provider,cloud_region,os,python_version,codecarbon_version,cpu_count,cpu_model,gpu_count,gpu_model,longitude,latitude,ram_total_size,tracking_mode,on_cloud,pue
+2024-01-13T10:03:32,codecarbon,e685eaea-e321-476e-a28b-1a62f6e369fd,10.400731801986694,2.34910579187555e-05,2.2585966416582685e-06,42.5,66.37527037995254,0.3602113723754883,0.00012278397960795298,0.00022478351316013168,1.0162424857437222e-06,0.00034858373525382844,France,FRA,île-de-france,,,Linux-5.15.0-91-generic-x86_64-with-glibc2.35,3.10.12,2.3.2,32,AMD Ryzen 9 7950X 16-Core Processor,1,1 x NVIDIA GeForce RTX 4090,2.4075,48.8323,125.53921508789062,process,N,1.0

audace/pytorch+cuda+float16/BEE-spoke-data/smol_llama-220M-GQA/hydra_config.yaml

@@ -0,0 +1,84 @@
+launcher:
+  name: process
+  _target_: optimum_benchmark.launchers.process.launcher.ProcessLauncher
+  device_isolation: true
+  start_method: spawn
+backend:
+  name: pytorch
+  version: 2.1.2+cu118
+  _target_: optimum_benchmark.backends.pytorch.backend.PyTorchBackend
+  seed: 42
+  inter_op_num_threads: null
+  intra_op_num_threads: null
+  delete_cache: false
+  no_weights: true
+  device_map: null
+  torch_dtype: float16
+  eval_mode: true
+  disable_grad: true
+  amp_autocast: false
+  amp_dtype: null
+  torch_compile: false
+  torch_compile_config: {}
+  to_bettertransformer: false
+  use_flash_attention_2: false
+  quantization_scheme: null
+  quantization_config: {}
+  data_parallel: false
+  deepspeed_inference: false
+  deepspeed_inference_config: {}
+  peft_strategy: null
+  peft_config: {}
+benchmark:
+  name: inference
+  _target_: optimum_benchmark.benchmarks.inference.benchmark.InferenceBenchmark
+  duration: 10
+  warmup_runs: 10
+  memory: true
+  energy: true
+  input_shapes:
+    batch_size: 1
+    sequence_length: 256
+    num_choices: 1
+    feature_size: 80
+    nb_max_frames: 3000
+    audio_sequence_length: 16000
+  new_tokens: 256
+  can_diffuse: false
+  can_generate: true
+  forward_kwargs: {}
+  generate_kwargs:
+    num_return_sequences: 1
+    max_new_tokens: 256
+    min_new_tokens: 256
+    do_sample: false
+    use_cache: true
+    pad_token_id: 0
+    temperature: 1.0
+    num_beams: 1
+experiment_name: pytorch+cuda+float16
+device: cuda
+model: BEE-spoke-data/smol_llama-220M-GQA
+task: text-generation
+hub_kwargs:
+  revision: main
+  cache_dir: null
+  force_download: false
+  local_files_only: false
+  trust_remote_code: true
+environment:
+  optimum_version: 1.16.1
+  optimum_commit: null
+  transformers_version: 4.36.2
+  transformers_commit: null
+  accelerate_version: 0.26.1
+  accelerate_commit: null
+  diffusers_version: null
+  diffusers_commit: null
+  python_version: 3.10.12
+  system: Linux
+  cpu: ' AMD Ryzen 9 7950X 16-Core Processor'
+  cpu_count: 32
+  cpu_ram_mb: 134796
+  gpus:
+  - NVIDIA GeForce RTX 4090

audace/pytorch+cuda+float16/BEE-spoke-data/smol_llama-220M-GQA/inference_results.csv

@@ -0,0 +1,2 @@
+forward.latency(s),forward.throughput(samples/s),forward.peak_memory(MB),forward.max_memory_used(MB),forward.max_memory_allocated(MB),forward.max_memory_reserved(MB),forward.energy_consumption(kWh/sample),forward.carbon_emissions(kgCO2eq/sample),generate.latency(s),generate.throughput(tokens/s),decode.latency(s),decode.throughput(tokens/s),generate.peak_memory(MB),generate.max_memory_used(MB),generate.max_memory_allocated(MB),generate.max_memory_reserved(MB),generate.energy_consumption(kWh/token),generate.carbon_emissions(kgCO2eq/token)
+0.00388,258.0,1397,1397,497,538,1.96e-07,1.32e-08,1.04,246.0,1.04,245.0,1408,1408,500,549,1.36e-07,9.18e-09

audace/pytorch+cuda+float16/BEE-spoke-data/smol_llama-81M-tied/.hydra/config.yaml

@@ -0,0 +1,76 @@
+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
+  delete_cache: false
+  no_weights: true
+  device_map: null
+  torch_dtype: float16
+  eval_mode: ${is_inference:${benchmark.name}}
+  disable_grad: ${is_inference:${benchmark.name}}
+  amp_autocast: false
+  amp_dtype: null
+  torch_compile: false
+  torch_compile_config: {}
+  to_bettertransformer: false
+  use_flash_attention_2: false
+  quantization_scheme: null
+  quantization_config: {}
+  data_parallel: false
+  deepspeed_inference: false
+  deepspeed_inference_config: {}
+  peft_strategy: null
+  peft_config: {}
+benchmark:
+  name: inference
+  _target_: optimum_benchmark.benchmarks.inference.benchmark.InferenceBenchmark
+  duration: 10
+  warmup_runs: 10
+  memory: true
+  energy: true
+  input_shapes:
+    batch_size: 1
+    sequence_length: 256
+    num_choices: 1
+    feature_size: 80
+    nb_max_frames: 3000
+    audio_sequence_length: 16000
+  new_tokens: 256
+  can_diffuse: ${can_diffuse:${task}}
+  can_generate: ${can_generate:${task}}
+  forward_kwargs: {}
+  generate_kwargs: {}
+launcher:
+  name: process
+  _target_: optimum_benchmark.launchers.process.launcher.ProcessLauncher
+  device_isolation: true
+  start_method: spawn
+experiment_name: pytorch+cuda+float16
+device: cuda
+model: BEE-spoke-data/smol_llama-81M-tied
+task: ${infer_task:${model}}
+hub_kwargs:
+  revision: main
+  cache_dir: null
+  force_download: false
+  local_files_only: false
+  trust_remote_code: true
+environment:
+  optimum_version: 1.16.1
+  optimum_commit: null
+  transformers_version: 4.36.2
+  transformers_commit: null
+  accelerate_version: 0.26.1
+  accelerate_commit: null
+  diffusers_version: null
+  diffusers_commit: null
+  python_version: 3.10.12
+  system: Linux
+  cpu: ' AMD Ryzen 9 7950X 16-Core Processor'
+  cpu_count: 32
+  cpu_ram_mb: 134796
+  gpus:
+  - NVIDIA GeForce RTX 4090

audace/pytorch+cuda+float16/BEE-spoke-data/smol_llama-81M-tied/.hydra/hydra.yaml

@@ -0,0 +1,176 @@
+hydra:
+  run:
+    dir: dataset/${oc.env:HOSTNAME}/${experiment_name}/${model}
+  sweep:
+    dir: multirun/${now:%Y-%m-%d}/${now:%H-%M-%S}
+    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: null
+  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: RUN
+  searchpath: []
+  callbacks: {}
+  output_subdir: .hydra
+  overrides:
+    hydra:
+    - hydra.mode=RUN
+    task:
+    - model=BEE-spoke-data/smol_llama-81M-tied
+  job:
+    name: cli
+    chdir: true
+    override_dirname: model=BEE-spoke-data/smol_llama-81M-tied
+    id: ???
+    num: ???
+    config_name: pytorch+cuda+float16
+    env_set:
+      COUNTRY_ISO_CODE: FRA
+      OVERRIDE_BENCHMARKS: '0'
+      CUDA_VISIBLE_DEVICES: '0'
+      CUDA_DEVICE_ORDER: PCI_BUS_ID
+    env_copy: []
+    config:
+      override_dirname:
+        kv_sep: '='
+        item_sep: ','
+        exclude_keys: []
+  runtime:
+    version: 1.3.2
+    version_base: '1.3'
+    cwd: /workspace/llm-perf
+    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: /workspace/llm-perf/configs
+      schema: file
+      provider: command-line
+    - path: ''
+      schema: structured
+      provider: schema
+    output_dir: /workspace/llm-perf/dataset/audace/pytorch+cuda+float16/BEE-spoke-data/smol_llama-81M-tied
+    choices:
+      launcher: process
+      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

audace/pytorch+cuda+float16/BEE-spoke-data/smol_llama-81M-tied/.hydra/overrides.yaml

@@ -0,0 +1 @@
+- model=BEE-spoke-data/smol_llama-81M-tied

audace/pytorch+cuda+float16/BEE-spoke-data/smol_llama-81M-tied/cli.log

@@ -0,0 +1,72 @@
+[2024-01-13 09:42:08,442][launcher][INFO] - Configuring process launcher
+[2024-01-13 09:42:08,442][process][INFO] - Setting multiprocessing start method to spawn.
+[2024-01-13 09:42:08,444][process][INFO] - 	+ Launched worker process with PID 495096.
+[2024-01-13 09:42:08,444][isolation][INFO] - 	+ Launched device(s) isolation process 495097.
+[2024-01-13 09:42:10,300][numexpr.utils][INFO] - Note: NumExpr detected 32 cores but "NUMEXPR_MAX_THREADS" not set, so enforcing safe limit of 8.
+[2024-01-13 09:42:10,300][numexpr.utils][INFO] - NumExpr defaulting to 8 threads.
+[2024-01-13 09:42:10,443][datasets][INFO] - PyTorch version 2.1.2+cu118 available.
+[2024-01-13 09:42:11,633][backend][INFO] - Configuring pytorch backend
+[2024-01-13 09:42:11,633][pytorch][INFO] - 	+ Inferred AutoModel class AutoModelForCausalLM for task text-generation and model_type llama
+[2024-01-13 09:42:11,633][pytorch][INFO] - 	+ Disabling gradients
+[2024-01-13 09:42:11,633][pytorch][INFO] - 	+ Loading model with no weights
+[2024-01-13 09:42:11,633][pytorch][INFO] - 	+ Creating no weights model directory
+[2024-01-13 09:42:11,633][pytorch][INFO] - 	+ Saving pretrained config
+[2024-01-13 09:42:11,634][pytorch][INFO] - 	+ Creating no weights model
+[2024-01-13 09:42:11,634][pytorch][INFO] - 	+ Saving no weights model
+[2024-01-13 09:42:11,635][pytorch][INFO] - 	+ Loading no weights model
+[2024-01-13 09:42:11,635][pytorch][INFO] - 	+ Loading model directly on device: cuda
+[2024-01-13 09:42:11,751][pytorch][INFO] - 	+ Randomizing model weights
+[2024-01-13 09:42:11,751][pytorch][INFO] - 	+ Tying model weights after randomization
+[2024-01-13 09:42:11,752][pytorch][INFO] - 	+ Turning on model's eval mode
+[2024-01-13 09:42:11,812][benchmark][INFO] - Configuring inference benchmark
+[2024-01-13 09:42:11,812][inference][INFO] - Running inference benchmark
+[2024-01-13 09:42:11,812][inference][INFO] - 	+ Updating input shapes with model shapes
+[2024-01-13 09:42:11,812][inference][INFO] - 	+ Preparing backend for inference
+[2024-01-13 09:42:11,813][inference][INFO] - 	+ Creating input generator
+[2024-01-13 09:42:11,813][input-generator][INFO] - Using text-generation task generator
+[2024-01-13 09:42:11,813][inference][INFO] - 	+ Preparing input for the forward pass
+[2024-01-13 09:42:11,813][backend][INFO] - 	+ Moving inputs tensors to device cuda
+[2024-01-13 09:42:11,813][inference][INFO] - 	+ Tracking forward pass peak memory
+[2024-01-13 09:42:11,813][memory][INFO] - Tracking CUDA devices: [0]
+[2024-01-13 09:42:11,813][memory][INFO] - Tracking Pytorch CUDA devices: [0]
+[2024-01-13 09:42:11,978][inference][INFO] - 	+ Forward pass max memory used: 1118 (MB)
+[2024-01-13 09:42:11,978][inference][INFO] - 	+ Forward pass max memory reserved: 260 (MB)
+[2024-01-13 09:42:11,978][inference][INFO] - 	+ Forward pass max memory allocated: 232 (MB)
+[2024-01-13 09:42:11,978][inference][INFO] - 	+ Preparing input for the generation pass
+[2024-01-13 09:42:11,978][backend][INFO] - 	+ Moving inputs tensors to device cuda
+[2024-01-13 09:42:11,978][inference][INFO] - 	+ Tracking generation pass peak memory
+[2024-01-13 09:42:11,978][memory][INFO] - Tracking CUDA devices: [0]
+[2024-01-13 09:42:11,978][memory][INFO] - Tracking Pytorch CUDA devices: [0]
+[2024-01-13 09:42:12,772][inference][INFO] - 	+ Generation pass max memory used: 1135 (MB)
+[2024-01-13 09:42:12,772][inference][INFO] - 	+ Generation pass max memory reserved: 276 (MB)
+[2024-01-13 09:42:12,772][inference][INFO] - 	+ Generation pass max memory allocated: 232 (MB)
+[2024-01-13 09:42:12,772][inference][INFO] - 	+ Preparing input for the forward pass
+[2024-01-13 09:42:12,772][backend][INFO] - 	+ Moving inputs tensors to device cuda
+[2024-01-13 09:42:12,772][inference][INFO] - 	+ Warming up the forward pass
+[2024-01-13 09:42:12,794][inference][INFO] - 	+ Tracking forward pass latency and throughput
+[2024-01-13 09:42:22,918][inference][INFO] - 	+ Forward pass latency: 2.24e-03 (s)
+[2024-01-13 09:42:22,920][inference][INFO] - 	+ Forward pass throughput: 446.00 (samples/s)
+[2024-01-13 09:42:22,920][inference][INFO] - 	+ Preparing input for the generation pass
+[2024-01-13 09:42:22,920][backend][INFO] - 	+ Moving inputs tensors to device cuda
+[2024-01-13 09:42:22,920][inference][INFO] - 	+ Warming up the generation pass
+[2024-01-13 09:42:23,547][inference][INFO] - 	+ Tracking generation latency and throughput
+[2024-01-13 09:42:33,554][inference][INFO] - 	+ Generation pass latency: 6.25e-01 (s)
+[2024-01-13 09:42:33,554][inference][INFO] - 	+ Generation pass throughput: 410.00 (tokens/s)
+[2024-01-13 09:42:33,554][inference][INFO] - 	+ Preparing input for the forward pass
+[2024-01-13 09:42:33,554][backend][INFO] - 	+ Moving inputs tensors to device cuda
+[2024-01-13 09:42:33,554][inference][INFO] - 	+ Tracking forward pass energy consumption
+[2024-01-13 09:42:48,010][inference][INFO] - 	+ Forward pass energy consumption: 1.06e-07 (kWh/sample)
+[2024-01-13 09:42:48,010][inference][INFO] - 	+ Forward pass carbon emissions: 7.14e-09 (kgCO2eq/sample)
+[2024-01-13 09:42:48,010][inference][INFO] - 	+ Full details in the CodeCarbon report: /workspace/llm-perf/dataset/audace/pytorch+cuda+float16/BEE-spoke-data/smol_llama-81M-tied/forward_codecarbon.csv
+[2024-01-13 09:42:48,010][inference][INFO] - 	+ Preparing input for the generation pass
+[2024-01-13 09:42:48,010][backend][INFO] - 	+ Moving inputs tensors to device cuda
+[2024-01-13 09:42:48,010][inference][INFO] - 	+ Tracking generation pass energy consumption
+[2024-01-13 09:43:03,094][inference][INFO] - 	+ Generation pass energy consumption: 7.33e-08 (kWh/token)
+[2024-01-13 09:43:03,094][inference][INFO] - 	+ Generation pass carbon emissions: 4.94e-09 (kgCO2eq/token)
+[2024-01-13 09:43:03,094][inference][INFO] - 	+ Full details in the CodeCarbon report: /workspace/llm-perf/dataset/audace/pytorch+cuda+float16/BEE-spoke-data/smol_llama-81M-tied/generate_codecarbon.csv
+[2024-01-13 09:43:03,094][inference][INFO] - Saving results
+[2024-01-13 09:43:03,101][backend][INFO] - Cleaning pytorch backend
+[2024-01-13 09:43:03,101][backend][INFO] - 	+ Deleting pretrained model
+[2024-01-13 09:43:03,186][pytorch][INFO] - 	+ Emptying CUDA cache
+[2024-01-13 09:43:03,195][pytorch][INFO] - 	+ Cleaning temporary directory
+[2024-01-13 09:43:03,689][isolation][INFO] - 	+ Closing device(s) isolation process...