Adding regression benchmark for the transformers SHA fe3c8ab1af558b95f67f5fafc0c55f09fd2b09db

raw_results/2023-08-11_19:07:08_fe3c8ab1af558b95f67f5fafc0c55f09fd2b09db/llama_1gpu_inference/0/hydra_config.yaml

@@ -0,0 +1,66 @@
+backend:
+  name: pytorch
+  version: 2.0.1+cu117
+  _target_: optimum_benchmark.backends.pytorch.PyTorchBackend
+  inter_op_num_threads: null
+  intra_op_num_threads: null
+  initial_isolation_check: true
+  continous_isolation_check: true
+  delete_cache: false
+  no_weights: false
+  torch_dtype: float16
+  device_map: null
+  load_in_8bit: false
+  load_in_4bit: false
+  bettertransformer: false
+  torch_compile: false
+  torch_compile_config:
+    fullgraph: false
+    dynamic: false
+    backend: inductor
+    mode: null
+    options: null
+    disable: false
+  amp_autocast: false
+  amp_dtype: null
+  disable_grad: true
+  eval_mode: true
+benchmark:
+  name: inference
+  _target_: optimum_benchmark.benchmarks.inference.InferenceBenchmark
+  seed: 42
+  memory: true
+  warmup_runs: 10
+  benchmark_duration: 20
+  input_shapes:
+    batch_size: 1
+    sequence_length: 200
+    num_choices: 4
+    width: 64
+    height: 64
+    num_channels: 3
+    point_batch_size: 3
+    nb_points_per_image: 2
+    feature_size: 80
+    nb_max_frames: 3000
+    audio_sequence_length: 16000
+  new_tokens: 200
+experiment_name: llama_1gpu_inference
+model: togethercomputer/LLaMA-2-7B-32K
+device: cuda
+task: text-generation
+hub_kwargs:
+  revision: main
+  cache_dir: null
+  force_download: false
+  local_files_only: false
+environment:
+  optimum_version: 1.11.0
+  transformers_version: 4.32.0.dev0
+  accelerate_version: 0.21.0
+  diffusers_version: null
+  python_version: 3.10.12
+  system: Linux
+  cpu: ' Intel(R) Xeon(R) Platinum 8275CL CPU @ 3.00GHz'
+  cpu_count: 96
+  cpu_ram_mb: 1204539.797504

raw_results/2023-08-11_19:07:08_fe3c8ab1af558b95f67f5fafc0c55f09fd2b09db/llama_1gpu_inference/0/inference_results.csv

@@ -0,0 +1,2 @@
+,forward.peak_memory(MB),forward.latency(s),forward.throughput(samples/s),generate.latency(s),generate.throughput(tokens/s)
+0,16195.125247999998,0.0313,31.9,5.97,33.5

raw_results/2023-08-11_19:07:08_fe3c8ab1af558b95f67f5fafc0c55f09fd2b09db/llama_1gpu_inference/0/main.log

@@ -0,0 +1,23 @@
+[2023-08-11 20:49:36,796][benchmark][INFO] - Configuring inference benchmark
+[2023-08-11 20:49:36,797][benchmark][INFO] - 	+ Setting seed(42)
+[2023-08-11 20:49:37,090][pytorch][INFO] - 	+ Infered AutoModel class AutoModelForCausalLM for task text-generation and model_type llama
+[2023-08-11 20:49:37,090][backend][INFO] - Configuring pytorch backend
+[2023-08-11 20:49:37,090][backend][INFO] - 	+ Checking initial device isolation
+[2023-08-11 20:49:37,234][backend][INFO] - 	+ Checking contineous device isolation
+[2023-08-11 20:49:37,250][pytorch][INFO] - 	+ Disabling gradients
+[2023-08-11 20:49:37,251][pytorch][INFO] - 	+ Loading pretrained model weights in dtype: float16 on device: cuda
+[2023-08-11 20:50:44,643][pytorch][INFO] - 	+ Turning on eval mode
+[2023-08-11 20:50:44,645][inference][INFO] - Running inference benchmark
+[2023-08-11 20:50:53,511][inference][INFO] - 	+ Tracking forward pass peak memory
+[2023-08-11 20:50:54,789][memory_tracker][INFO] - Peak memory usage: 16195.125247999998 MB
+[2023-08-11 20:50:54,790][inference][INFO] - 	+ Forward pass peak memory: 16195.125247999998 (MB)
+[2023-08-11 20:50:54,791][inference][INFO] - 	+ Warming up the forward pass
+[2023-08-11 20:50:55,104][inference][INFO] - 	+ Tracking forward pass latency and throughput
+[2023-08-11 20:51:15,432][inference][INFO] - 	+ Forward pass latency: 3.13e-02 (s)
+[2023-08-11 20:51:15,433][inference][INFO] - 	+ Forward pass throughput: 31.90 (samples/s)
+[2023-08-11 20:51:15,434][inference][INFO] - 	+ Warming up the generation pass
+[2023-08-11 20:51:22,081][inference][INFO] - 	+ Tracking generation latency and throughput
+[2023-08-11 20:51:45,953][inference][INFO] - 	+ Generation pass latency: 5.97e+00 (s)
+[2023-08-11 20:51:45,956][inference][INFO] - 	+ Generation pass throughput: 33.50 (tokens/s)
+[2023-08-11 20:51:45,956][inference][INFO] - Saving inference results
+[2023-08-11 20:51:45,965][backend][INFO] - Cleaning backend

raw_results/2023-08-11_19:07:08_fe3c8ab1af558b95f67f5fafc0c55f09fd2b09db/llama_1gpu_inference/1/hydra_config.yaml

@@ -0,0 +1,66 @@
+backend:
+  name: pytorch
+  version: 2.0.1+cu117
+  _target_: optimum_benchmark.backends.pytorch.PyTorchBackend
+  inter_op_num_threads: null
+  intra_op_num_threads: null
+  initial_isolation_check: true
+  continous_isolation_check: true
+  delete_cache: false
+  no_weights: false
+  torch_dtype: float32
+  device_map: null
+  load_in_8bit: false
+  load_in_4bit: false
+  bettertransformer: false
+  torch_compile: false
+  torch_compile_config:
+    fullgraph: false
+    dynamic: false
+    backend: inductor
+    mode: null
+    options: null
+    disable: false
+  amp_autocast: false
+  amp_dtype: null
+  disable_grad: true
+  eval_mode: true
+benchmark:
+  name: inference
+  _target_: optimum_benchmark.benchmarks.inference.InferenceBenchmark
+  seed: 42
+  memory: true
+  warmup_runs: 10
+  benchmark_duration: 20
+  input_shapes:
+    batch_size: 1
+    sequence_length: 200
+    num_choices: 4
+    width: 64
+    height: 64
+    num_channels: 3
+    point_batch_size: 3
+    nb_points_per_image: 2
+    feature_size: 80
+    nb_max_frames: 3000
+    audio_sequence_length: 16000
+  new_tokens: 200
+experiment_name: llama_1gpu_inference
+model: togethercomputer/LLaMA-2-7B-32K
+device: cuda
+task: text-generation
+hub_kwargs:
+  revision: main
+  cache_dir: null
+  force_download: false
+  local_files_only: false
+environment:
+  optimum_version: 1.11.0
+  transformers_version: 4.32.0.dev0
+  accelerate_version: 0.21.0
+  diffusers_version: null
+  python_version: 3.10.12
+  system: Linux
+  cpu: ' Intel(R) Xeon(R) Platinum 8275CL CPU @ 3.00GHz'
+  cpu_count: 96
+  cpu_ram_mb: 1204539.797504

raw_results/2023-08-11_19:07:08_fe3c8ab1af558b95f67f5fafc0c55f09fd2b09db/llama_1gpu_inference/1/inference_results.csv

@@ -0,0 +1,2 @@
+,forward.peak_memory(MB),forward.latency(s),forward.throughput(samples/s),generate.latency(s),generate.throughput(tokens/s)
+0,30317.346815999997,0.0642,15.6,5.66,35.3

raw_results/2023-08-11_19:07:08_fe3c8ab1af558b95f67f5fafc0c55f09fd2b09db/llama_1gpu_inference/1/main.log

@@ -0,0 +1,23 @@
+[2023-08-11 20:51:46,444][benchmark][INFO] - Configuring inference benchmark
+[2023-08-11 20:51:46,445][benchmark][INFO] - 	+ Setting seed(42)
+[2023-08-11 20:51:46,706][pytorch][INFO] - 	+ Infered AutoModel class AutoModelForCausalLM for task text-generation and model_type llama
+[2023-08-11 20:51:46,707][backend][INFO] - Configuring pytorch backend
+[2023-08-11 20:51:46,707][backend][INFO] - 	+ Checking initial device isolation
+[2023-08-11 20:51:46,814][backend][INFO] - 	+ Checking contineous device isolation
+[2023-08-11 20:51:46,851][pytorch][INFO] - 	+ Disabling gradients
+[2023-08-11 20:51:46,852][pytorch][INFO] - 	+ Loading pretrained model weights in dtype: float32 on device: cuda
+[2023-08-11 20:52:04,268][pytorch][INFO] - 	+ Turning on eval mode
+[2023-08-11 20:52:04,270][inference][INFO] - Running inference benchmark
+[2023-08-11 20:52:12,744][inference][INFO] - 	+ Tracking forward pass peak memory
+[2023-08-11 20:52:12,820][memory_tracker][INFO] - Peak memory usage: 30317.346815999997 MB
+[2023-08-11 20:52:12,820][inference][INFO] - 	+ Forward pass peak memory: 30317.346815999997 (MB)
+[2023-08-11 20:52:12,821][inference][INFO] - 	+ Warming up the forward pass
+[2023-08-11 20:52:14,964][inference][INFO] - 	+ Tracking forward pass latency and throughput
+[2023-08-11 20:53:21,313][inference][INFO] - 	+ Forward pass latency: 6.42e-02 (s)
+[2023-08-11 20:53:21,314][inference][INFO] - 	+ Forward pass throughput: 15.60 (samples/s)
+[2023-08-11 20:53:21,314][inference][INFO] - 	+ Warming up the generation pass
+[2023-08-11 20:53:26,968][inference][INFO] - 	+ Tracking generation latency and throughput
+[2023-08-11 20:53:49,607][inference][INFO] - 	+ Generation pass latency: 5.66e+00 (s)
+[2023-08-11 20:53:49,610][inference][INFO] - 	+ Generation pass throughput: 35.30 (tokens/s)
+[2023-08-11 20:53:49,610][inference][INFO] - Saving inference results
+[2023-08-11 20:53:49,617][backend][INFO] - Cleaning backend

raw_results/2023-08-11_19:07:08_fe3c8ab1af558b95f67f5fafc0c55f09fd2b09db/llama_1gpu_inference/2/hydra_config.yaml

@@ -0,0 +1,66 @@
+backend:
+  name: pytorch
+  version: 2.0.1+cu117
+  _target_: optimum_benchmark.backends.pytorch.PyTorchBackend
+  inter_op_num_threads: null
+  intra_op_num_threads: null
+  initial_isolation_check: true
+  continous_isolation_check: true
+  delete_cache: false
+  no_weights: false
+  torch_dtype: float16
+  device_map: null
+  load_in_8bit: false
+  load_in_4bit: false
+  bettertransformer: false
+  torch_compile: false
+  torch_compile_config:
+    fullgraph: false
+    dynamic: false
+    backend: inductor
+    mode: null
+    options: null
+    disable: false
+  amp_autocast: false
+  amp_dtype: null
+  disable_grad: true
+  eval_mode: true
+benchmark:
+  name: inference
+  _target_: optimum_benchmark.benchmarks.inference.InferenceBenchmark
+  seed: 42
+  memory: true
+  warmup_runs: 10
+  benchmark_duration: 20
+  input_shapes:
+    batch_size: 2
+    sequence_length: 200
+    num_choices: 4
+    width: 64
+    height: 64
+    num_channels: 3
+    point_batch_size: 3
+    nb_points_per_image: 2
+    feature_size: 80
+    nb_max_frames: 3000
+    audio_sequence_length: 16000
+  new_tokens: 200
+experiment_name: llama_1gpu_inference
+model: togethercomputer/LLaMA-2-7B-32K
+device: cuda
+task: text-generation
+hub_kwargs:
+  revision: main
+  cache_dir: null
+  force_download: false
+  local_files_only: false
+environment:
+  optimum_version: 1.11.0
+  transformers_version: 4.32.0.dev0
+  accelerate_version: 0.21.0
+  diffusers_version: null
+  python_version: 3.10.12
+  system: Linux
+  cpu: ' Intel(R) Xeon(R) Platinum 8275CL CPU @ 3.00GHz'
+  cpu_count: 96
+  cpu_ram_mb: 1204539.797504

raw_results/2023-08-11_19:07:08_fe3c8ab1af558b95f67f5fafc0c55f09fd2b09db/llama_1gpu_inference/2/inference_results.csv

@@ -0,0 +1,2 @@
+,forward.peak_memory(MB),forward.latency(s),forward.throughput(samples/s),generate.latency(s),generate.throughput(tokens/s)
+0,16381.771776,0.0315,63.5,6.18,64.7

raw_results/2023-08-11_19:07:08_fe3c8ab1af558b95f67f5fafc0c55f09fd2b09db/llama_1gpu_inference/2/main.log

@@ -0,0 +1,23 @@
+[2023-08-11 20:53:50,108][benchmark][INFO] - Configuring inference benchmark
+[2023-08-11 20:53:50,109][benchmark][INFO] - 	+ Setting seed(42)
+[2023-08-11 20:53:50,302][pytorch][INFO] - 	+ Infered AutoModel class AutoModelForCausalLM for task text-generation and model_type llama
+[2023-08-11 20:53:50,303][backend][INFO] - Configuring pytorch backend
+[2023-08-11 20:53:50,303][backend][INFO] - 	+ Checking initial device isolation
+[2023-08-11 20:53:50,403][backend][INFO] - 	+ Checking contineous device isolation
+[2023-08-11 20:53:50,438][pytorch][INFO] - 	+ Disabling gradients
+[2023-08-11 20:53:50,438][pytorch][INFO] - 	+ Loading pretrained model weights in dtype: float16 on device: cuda
+[2023-08-11 20:54:01,733][pytorch][INFO] - 	+ Turning on eval mode
+[2023-08-11 20:54:01,734][inference][INFO] - Running inference benchmark
+[2023-08-11 20:54:10,232][inference][INFO] - 	+ Tracking forward pass peak memory
+[2023-08-11 20:54:10,273][memory_tracker][INFO] - Peak memory usage: 16381.771776 MB
+[2023-08-11 20:54:10,273][inference][INFO] - 	+ Forward pass peak memory: 16381.771776 (MB)
+[2023-08-11 20:54:10,274][inference][INFO] - 	+ Warming up the forward pass
+[2023-08-11 20:54:10,742][inference][INFO] - 	+ Tracking forward pass latency and throughput
+[2023-08-11 20:54:41,250][inference][INFO] - 	+ Forward pass latency: 3.15e-02 (s)
+[2023-08-11 20:54:41,251][inference][INFO] - 	+ Forward pass throughput: 63.50 (samples/s)
+[2023-08-11 20:54:41,251][inference][INFO] - 	+ Warming up the generation pass
+[2023-08-11 20:54:48,239][inference][INFO] - 	+ Tracking generation latency and throughput
+[2023-08-11 20:55:12,962][inference][INFO] - 	+ Generation pass latency: 6.18e+00 (s)
+[2023-08-11 20:55:12,964][inference][INFO] - 	+ Generation pass throughput: 64.70 (tokens/s)
+[2023-08-11 20:55:12,964][inference][INFO] - Saving inference results
+[2023-08-11 20:55:12,972][backend][INFO] - Cleaning backend

raw_results/2023-08-11_19:07:08_fe3c8ab1af558b95f67f5fafc0c55f09fd2b09db/llama_1gpu_inference/3/hydra_config.yaml

@@ -0,0 +1,66 @@
+backend:
+  name: pytorch
+  version: 2.0.1+cu117
+  _target_: optimum_benchmark.backends.pytorch.PyTorchBackend
+  inter_op_num_threads: null
+  intra_op_num_threads: null
+  initial_isolation_check: true
+  continous_isolation_check: true
+  delete_cache: false
+  no_weights: false
+  torch_dtype: float32
+  device_map: null
+  load_in_8bit: false
+  load_in_4bit: false
+  bettertransformer: false
+  torch_compile: false
+  torch_compile_config:
+    fullgraph: false
+    dynamic: false
+    backend: inductor
+    mode: null
+    options: null
+    disable: false
+  amp_autocast: false
+  amp_dtype: null
+  disable_grad: true
+  eval_mode: true
+benchmark:
+  name: inference
+  _target_: optimum_benchmark.benchmarks.inference.InferenceBenchmark
+  seed: 42
+  memory: true
+  warmup_runs: 10
+  benchmark_duration: 20
+  input_shapes:
+    batch_size: 2
+    sequence_length: 200
+    num_choices: 4
+    width: 64
+    height: 64
+    num_channels: 3
+    point_batch_size: 3
+    nb_points_per_image: 2
+    feature_size: 80
+    nb_max_frames: 3000
+    audio_sequence_length: 16000
+  new_tokens: 200
+experiment_name: llama_1gpu_inference
+model: togethercomputer/LLaMA-2-7B-32K
+device: cuda
+task: text-generation
+hub_kwargs:
+  revision: main
+  cache_dir: null
+  force_download: false
+  local_files_only: false
+environment:
+  optimum_version: 1.11.0
+  transformers_version: 4.32.0.dev0
+  accelerate_version: 0.21.0
+  diffusers_version: null
+  python_version: 3.10.12
+  system: Linux
+  cpu: ' Intel(R) Xeon(R) Platinum 8275CL CPU @ 3.00GHz'
+  cpu_count: 96
+  cpu_ram_mb: 1204539.797504

raw_results/2023-08-11_19:07:08_fe3c8ab1af558b95f67f5fafc0c55f09fd2b09db/llama_1gpu_inference/3/inference_results.csv

@@ -0,0 +1,2 @@
+,forward.peak_memory(MB),forward.latency(s),forward.throughput(samples/s),generate.latency(s),generate.throughput(tokens/s)
+0,30778.720255999997,0.109,18.3,7.05,56.7

raw_results/2023-08-11_19:07:08_fe3c8ab1af558b95f67f5fafc0c55f09fd2b09db/llama_1gpu_inference/3/main.log

@@ -0,0 +1,23 @@
+[2023-08-11 20:55:13,446][benchmark][INFO] - Configuring inference benchmark
+[2023-08-11 20:55:13,447][benchmark][INFO] - 	+ Setting seed(42)
+[2023-08-11 20:55:13,641][pytorch][INFO] - 	+ Infered AutoModel class AutoModelForCausalLM for task text-generation and model_type llama
+[2023-08-11 20:55:13,641][backend][INFO] - Configuring pytorch backend
+[2023-08-11 20:55:13,641][backend][INFO] - 	+ Checking initial device isolation
+[2023-08-11 20:55:13,744][backend][INFO] - 	+ Checking contineous device isolation
+[2023-08-11 20:55:13,779][pytorch][INFO] - 	+ Disabling gradients
+[2023-08-11 20:55:13,780][pytorch][INFO] - 	+ Loading pretrained model weights in dtype: float32 on device: cuda
+[2023-08-11 20:55:31,032][pytorch][INFO] - 	+ Turning on eval mode
+[2023-08-11 20:55:31,034][inference][INFO] - Running inference benchmark
+[2023-08-11 20:55:39,761][inference][INFO] - 	+ Tracking forward pass peak memory
+[2023-08-11 20:55:39,886][memory_tracker][INFO] - Peak memory usage: 30778.720255999997 MB
+[2023-08-11 20:55:39,886][inference][INFO] - 	+ Forward pass peak memory: 30778.720255999997 (MB)
+[2023-08-11 20:55:39,887][inference][INFO] - 	+ Warming up the forward pass
+[2023-08-11 20:55:43,714][inference][INFO] - 	+ Tracking forward pass latency and throughput
+[2023-08-11 20:56:54,313][inference][INFO] - 	+ Forward pass latency: 1.09e-01 (s)
+[2023-08-11 20:56:54,313][inference][INFO] - 	+ Forward pass throughput: 18.30 (samples/s)
+[2023-08-11 20:56:54,314][inference][INFO] - 	+ Warming up the generation pass
+[2023-08-11 20:57:01,375][inference][INFO] - 	+ Tracking generation latency and throughput
+[2023-08-11 20:57:22,517][inference][INFO] - 	+ Generation pass latency: 7.05e+00 (s)
+[2023-08-11 20:57:22,520][inference][INFO] - 	+ Generation pass throughput: 56.70 (tokens/s)
+[2023-08-11 20:57:22,520][inference][INFO] - Saving inference results
+[2023-08-11 20:57:22,528][backend][INFO] - Cleaning backend

raw_results/2023-08-11_19:07:08_fe3c8ab1af558b95f67f5fafc0c55f09fd2b09db/llama_1gpu_inference/4/hydra_config.yaml

@@ -0,0 +1,66 @@
+backend:
+  name: pytorch
+  version: 2.0.1+cu117
+  _target_: optimum_benchmark.backends.pytorch.PyTorchBackend
+  inter_op_num_threads: null
+  intra_op_num_threads: null
+  initial_isolation_check: true
+  continous_isolation_check: true
+  delete_cache: false
+  no_weights: false
+  torch_dtype: float16
+  device_map: null
+  load_in_8bit: false
+  load_in_4bit: false
+  bettertransformer: false
+  torch_compile: false
+  torch_compile_config:
+    fullgraph: false
+    dynamic: false
+    backend: inductor
+    mode: null
+    options: null
+    disable: false
+  amp_autocast: false
+  amp_dtype: null
+  disable_grad: true
+  eval_mode: true
+benchmark:
+  name: inference
+  _target_: optimum_benchmark.benchmarks.inference.InferenceBenchmark
+  seed: 42
+  memory: true
+  warmup_runs: 10
+  benchmark_duration: 20
+  input_shapes:
+    batch_size: 4
+    sequence_length: 200
+    num_choices: 4
+    width: 64
+    height: 64
+    num_channels: 3
+    point_batch_size: 3
+    nb_points_per_image: 2
+    feature_size: 80
+    nb_max_frames: 3000
+    audio_sequence_length: 16000
+  new_tokens: 200
+experiment_name: llama_1gpu_inference
+model: togethercomputer/LLaMA-2-7B-32K
+device: cuda
+task: text-generation
+hub_kwargs:
+  revision: main
+  cache_dir: null
+  force_download: false
+  local_files_only: false
+environment:
+  optimum_version: 1.11.0
+  transformers_version: 4.32.0.dev0
+  accelerate_version: 0.21.0
+  diffusers_version: null
+  python_version: 3.10.12
+  system: Linux
+  cpu: ' Intel(R) Xeon(R) Platinum 8275CL CPU @ 3.00GHz'
+  cpu_count: 96
+  cpu_ram_mb: 1204539.797504

raw_results/2023-08-11_19:07:08_fe3c8ab1af558b95f67f5fafc0c55f09fd2b09db/llama_1gpu_inference/4/inference_results.csv

@@ -0,0 +1,2 @@
+,forward.peak_memory(MB),forward.latency(s),forward.throughput(samples/s),generate.latency(s),generate.throughput(tokens/s)
+0,17000.431615999998,0.0317,126.0,6.19,129.0

raw_results/2023-08-11_19:07:08_fe3c8ab1af558b95f67f5fafc0c55f09fd2b09db/llama_1gpu_inference/4/main.log

@@ -0,0 +1,23 @@
+[2023-08-11 20:57:23,044][benchmark][INFO] - Configuring inference benchmark
+[2023-08-11 20:57:23,045][benchmark][INFO] - 	+ Setting seed(42)
+[2023-08-11 20:57:23,243][pytorch][INFO] - 	+ Infered AutoModel class AutoModelForCausalLM for task text-generation and model_type llama
+[2023-08-11 20:57:23,243][backend][INFO] - Configuring pytorch backend
+[2023-08-11 20:57:23,243][backend][INFO] - 	+ Checking initial device isolation
+[2023-08-11 20:57:23,349][backend][INFO] - 	+ Checking contineous device isolation
+[2023-08-11 20:57:23,387][pytorch][INFO] - 	+ Disabling gradients
+[2023-08-11 20:57:23,388][pytorch][INFO] - 	+ Loading pretrained model weights in dtype: float16 on device: cuda
+[2023-08-11 20:57:34,727][pytorch][INFO] - 	+ Turning on eval mode
+[2023-08-11 20:57:34,729][inference][INFO] - Running inference benchmark
+[2023-08-11 20:57:43,214][inference][INFO] - 	+ Tracking forward pass peak memory
+[2023-08-11 20:57:43,262][memory_tracker][INFO] - Peak memory usage: 17000.431615999998 MB
+[2023-08-11 20:57:43,262][inference][INFO] - 	+ Forward pass peak memory: 17000.431615999998 (MB)
+[2023-08-11 20:57:43,262][inference][INFO] - 	+ Warming up the forward pass
+[2023-08-11 20:57:44,021][inference][INFO] - 	+ Tracking forward pass latency and throughput
+[2023-08-11 20:58:31,578][inference][INFO] - 	+ Forward pass latency: 3.17e-02 (s)
+[2023-08-11 20:58:31,580][inference][INFO] - 	+ Forward pass throughput: 126.00 (samples/s)
+[2023-08-11 20:58:31,580][inference][INFO] - 	+ Warming up the generation pass
+[2023-08-11 20:58:38,781][inference][INFO] - 	+ Tracking generation latency and throughput
+[2023-08-11 20:59:03,524][inference][INFO] - 	+ Generation pass latency: 6.19e+00 (s)
+[2023-08-11 20:59:03,527][inference][INFO] - 	+ Generation pass throughput: 129.00 (tokens/s)
+[2023-08-11 20:59:03,527][inference][INFO] - Saving inference results
+[2023-08-11 20:59:03,536][backend][INFO] - Cleaning backend

raw_results/2023-08-11_19:07:08_fe3c8ab1af558b95f67f5fafc0c55f09fd2b09db/llama_1gpu_inference/5/hydra_config.yaml

@@ -0,0 +1,66 @@
+backend:
+  name: pytorch
+  version: 2.0.1+cu117
+  _target_: optimum_benchmark.backends.pytorch.PyTorchBackend
+  inter_op_num_threads: null
+  intra_op_num_threads: null
+  initial_isolation_check: true
+  continous_isolation_check: true
+  delete_cache: false
+  no_weights: false
+  torch_dtype: float32
+  device_map: null
+  load_in_8bit: false
+  load_in_4bit: false
+  bettertransformer: false
+  torch_compile: false
+  torch_compile_config:
+    fullgraph: false
+    dynamic: false
+    backend: inductor
+    mode: null
+    options: null
+    disable: false
+  amp_autocast: false
+  amp_dtype: null
+  disable_grad: true
+  eval_mode: true
+benchmark:
+  name: inference
+  _target_: optimum_benchmark.benchmarks.inference.InferenceBenchmark
+  seed: 42
+  memory: true
+  warmup_runs: 10
+  benchmark_duration: 20
+  input_shapes:
+    batch_size: 4
+    sequence_length: 200
+    num_choices: 4
+    width: 64
+    height: 64
+    num_channels: 3
+    point_batch_size: 3
+    nb_points_per_image: 2
+    feature_size: 80
+    nb_max_frames: 3000
+    audio_sequence_length: 16000
+  new_tokens: 200
+experiment_name: llama_1gpu_inference
+model: togethercomputer/LLaMA-2-7B-32K
+device: cuda
+task: text-generation
+hub_kwargs:
+  revision: main
+  cache_dir: null
+  force_download: false
+  local_files_only: false
+environment:
+  optimum_version: 1.11.0
+  transformers_version: 4.32.0.dev0
+  accelerate_version: 0.21.0
+  diffusers_version: null
+  python_version: 3.10.12
+  system: Linux
+  cpu: ' Intel(R) Xeon(R) Platinum 8275CL CPU @ 3.00GHz'
+  cpu_count: 96
+  cpu_ram_mb: 1204539.797504

raw_results/2023-08-11_19:07:08_fe3c8ab1af558b95f67f5fafc0c55f09fd2b09db/llama_1gpu_inference/5/inference_results.csv

@@ -0,0 +1,2 @@
+,forward.peak_memory(MB),forward.latency(s),forward.throughput(samples/s),generate.latency(s),generate.throughput(tokens/s)
+0,31481.266175999997,0.187,21.4,7.67,104.0

raw_results/2023-08-11_19:07:08_fe3c8ab1af558b95f67f5fafc0c55f09fd2b09db/llama_1gpu_inference/5/main.log

@@ -0,0 +1,23 @@
+[2023-08-11 20:59:04,065][benchmark][INFO] - Configuring inference benchmark
+[2023-08-11 20:59:04,067][benchmark][INFO] - 	+ Setting seed(42)
+[2023-08-11 20:59:04,255][pytorch][INFO] - 	+ Infered AutoModel class AutoModelForCausalLM for task text-generation and model_type llama
+[2023-08-11 20:59:04,256][backend][INFO] - Configuring pytorch backend
+[2023-08-11 20:59:04,256][backend][INFO] - 	+ Checking initial device isolation
+[2023-08-11 20:59:04,363][backend][INFO] - 	+ Checking contineous device isolation
+[2023-08-11 20:59:04,400][pytorch][INFO] - 	+ Disabling gradients
+[2023-08-11 20:59:04,401][pytorch][INFO] - 	+ Loading pretrained model weights in dtype: float32 on device: cuda
+[2023-08-11 20:59:21,875][pytorch][INFO] - 	+ Turning on eval mode
+[2023-08-11 20:59:21,876][inference][INFO] - Running inference benchmark
+[2023-08-11 20:59:30,357][inference][INFO] - 	+ Tracking forward pass peak memory
+[2023-08-11 20:59:30,570][memory_tracker][INFO] - Peak memory usage: 31481.266175999997 MB
+[2023-08-11 20:59:30,570][inference][INFO] - 	+ Forward pass peak memory: 31481.266175999997 (MB)
+[2023-08-11 20:59:30,575][inference][INFO] - 	+ Warming up the forward pass
+[2023-08-11 20:59:37,498][inference][INFO] - 	+ Tracking forward pass latency and throughput
+[2023-08-11 21:00:52,157][inference][INFO] - 	+ Forward pass latency: 1.87e-01 (s)
+[2023-08-11 21:00:52,160][inference][INFO] - 	+ Forward pass throughput: 21.40 (samples/s)
+[2023-08-11 21:00:52,160][inference][INFO] - 	+ Warming up the generation pass
+[2023-08-11 21:00:59,914][inference][INFO] - 	+ Tracking generation latency and throughput
+[2023-08-11 21:01:22,918][inference][INFO] - 	+ Generation pass latency: 7.67e+00 (s)
+[2023-08-11 21:01:22,921][inference][INFO] - 	+ Generation pass throughput: 104.00 (tokens/s)
+[2023-08-11 21:01:22,921][inference][INFO] - Saving inference results
+[2023-08-11 21:01:22,928][backend][INFO] - Cleaning backend

raw_results/2023-08-11_19:07:08_fe3c8ab1af558b95f67f5fafc0c55f09fd2b09db/llama_1gpu_inference/6/hydra_config.yaml

@@ -0,0 +1,66 @@
+backend:
+  name: pytorch
+  version: 2.0.1+cu117
+  _target_: optimum_benchmark.backends.pytorch.PyTorchBackend
+  inter_op_num_threads: null
+  intra_op_num_threads: null
+  initial_isolation_check: true
+  continous_isolation_check: true
+  delete_cache: false
+  no_weights: false
+  torch_dtype: float16
+  device_map: null
+  load_in_8bit: false
+  load_in_4bit: false
+  bettertransformer: false
+  torch_compile: false
+  torch_compile_config:
+    fullgraph: false
+    dynamic: false
+    backend: inductor
+    mode: null
+    options: null
+    disable: false
+  amp_autocast: false
+  amp_dtype: null
+  disable_grad: true
+  eval_mode: true
+benchmark:
+  name: inference
+  _target_: optimum_benchmark.benchmarks.inference.InferenceBenchmark
+  seed: 42
+  memory: true
+  warmup_runs: 10
+  benchmark_duration: 20
+  input_shapes:
+    batch_size: 16
+    sequence_length: 200
+    num_choices: 4
+    width: 64
+    height: 64
+    num_channels: 3
+    point_batch_size: 3
+    nb_points_per_image: 2
+    feature_size: 80
+    nb_max_frames: 3000
+    audio_sequence_length: 16000
+  new_tokens: 200
+experiment_name: llama_1gpu_inference
+model: togethercomputer/LLaMA-2-7B-32K
+device: cuda
+task: text-generation
+hub_kwargs:
+  revision: main
+  cache_dir: null
+  force_download: false
+  local_files_only: false
+environment:
+  optimum_version: 1.11.0
+  transformers_version: 4.32.0.dev0
+  accelerate_version: 0.21.0
+  diffusers_version: null
+  python_version: 3.10.12
+  system: Linux
+  cpu: ' Intel(R) Xeon(R) Platinum 8275CL CPU @ 3.00GHz'
+  cpu_count: 96
+  cpu_ram_mb: 1204539.797504

raw_results/2023-08-11_19:07:08_fe3c8ab1af558b95f67f5fafc0c55f09fd2b09db/llama_1gpu_inference/6/inference_results.csv

@@ -0,0 +1,2 @@
+,forward.peak_memory(MB),forward.latency(s),forward.throughput(samples/s),generate.latency(s),generate.throughput(tokens/s)
+0,19498.139648,0.0982,163.0,6.42,498.0

raw_results/2023-08-11_19:07:08_fe3c8ab1af558b95f67f5fafc0c55f09fd2b09db/llama_1gpu_inference/6/main.log

@@ -0,0 +1,23 @@
+[2023-08-11 21:01:23,498][benchmark][INFO] - Configuring inference benchmark
+[2023-08-11 21:01:23,499][benchmark][INFO] - 	+ Setting seed(42)
+[2023-08-11 21:01:23,697][pytorch][INFO] - 	+ Infered AutoModel class AutoModelForCausalLM for task text-generation and model_type llama
+[2023-08-11 21:01:23,698][backend][INFO] - Configuring pytorch backend
+[2023-08-11 21:01:23,698][backend][INFO] - 	+ Checking initial device isolation
+[2023-08-11 21:01:23,802][backend][INFO] - 	+ Checking contineous device isolation
+[2023-08-11 21:01:23,837][pytorch][INFO] - 	+ Disabling gradients
+[2023-08-11 21:01:23,838][pytorch][INFO] - 	+ Loading pretrained model weights in dtype: float16 on device: cuda
+[2023-08-11 21:01:36,731][pytorch][INFO] - 	+ Turning on eval mode
+[2023-08-11 21:01:36,732][inference][INFO] - Running inference benchmark
+[2023-08-11 21:01:45,591][inference][INFO] - 	+ Tracking forward pass peak memory
+[2023-08-11 21:01:45,703][memory_tracker][INFO] - Peak memory usage: 19498.139648 MB
+[2023-08-11 21:01:45,703][inference][INFO] - 	+ Forward pass peak memory: 19498.139648 (MB)
+[2023-08-11 21:01:45,704][inference][INFO] - 	+ Warming up the forward pass
+[2023-08-11 21:01:48,360][inference][INFO] - 	+ Tracking forward pass latency and throughput
+[2023-08-11 21:02:42,434][inference][INFO] - 	+ Forward pass latency: 9.82e-02 (s)
+[2023-08-11 21:02:42,435][inference][INFO] - 	+ Forward pass throughput: 163.00 (samples/s)
+[2023-08-11 21:02:42,435][inference][INFO] - 	+ Warming up the generation pass
+[2023-08-11 21:02:49,098][inference][INFO] - 	+ Tracking generation latency and throughput
+[2023-08-11 21:03:14,760][inference][INFO] - 	+ Generation pass latency: 6.42e+00 (s)
+[2023-08-11 21:03:14,763][inference][INFO] - 	+ Generation pass throughput: 498.00 (tokens/s)
+[2023-08-11 21:03:14,763][inference][INFO] - Saving inference results
+[2023-08-11 21:03:14,770][backend][INFO] - Cleaning backend

raw_results/2023-08-11_19:07:08_fe3c8ab1af558b95f67f5fafc0c55f09fd2b09db/llama_1gpu_inference/7/hydra_config.yaml

@@ -0,0 +1,66 @@
+backend:
+  name: pytorch
+  version: 2.0.1+cu117
+  _target_: optimum_benchmark.backends.pytorch.PyTorchBackend
+  inter_op_num_threads: null
+  intra_op_num_threads: null
+  initial_isolation_check: true
+  continous_isolation_check: true
+  delete_cache: false
+  no_weights: false
+  torch_dtype: float32
+  device_map: null
+  load_in_8bit: false
+  load_in_4bit: false
+  bettertransformer: false
+  torch_compile: false
+  torch_compile_config:
+    fullgraph: false
+    dynamic: false
+    backend: inductor
+    mode: null
+    options: null
+    disable: false
+  amp_autocast: false
+  amp_dtype: null
+  disable_grad: true
+  eval_mode: true
+benchmark:
+  name: inference
+  _target_: optimum_benchmark.benchmarks.inference.InferenceBenchmark
+  seed: 42
+  memory: true
+  warmup_runs: 10
+  benchmark_duration: 20
+  input_shapes:
+    batch_size: 16
+    sequence_length: 200
+    num_choices: 4
+    width: 64
+    height: 64
+    num_channels: 3
+    point_batch_size: 3
+    nb_points_per_image: 2
+    feature_size: 80
+    nb_max_frames: 3000
+    audio_sequence_length: 16000
+  new_tokens: 200
+experiment_name: llama_1gpu_inference
+model: togethercomputer/LLaMA-2-7B-32K
+device: cuda
+task: text-generation
+hub_kwargs:
+  revision: main
+  cache_dir: null
+  force_download: false
+  local_files_only: false
+environment:
+  optimum_version: 1.11.0
+  transformers_version: 4.32.0.dev0
+  accelerate_version: 0.21.0
+  diffusers_version: null
+  python_version: 3.10.12
+  system: Linux
+  cpu: ' Intel(R) Xeon(R) Platinum 8275CL CPU @ 3.00GHz'
+  cpu_count: 96
+  cpu_ram_mb: 1204539.797504

raw_results/2023-08-11_19:07:08_fe3c8ab1af558b95f67f5fafc0c55f09fd2b09db/llama_1gpu_inference/7/inference_results.csv

@@ -0,0 +1,2 @@
+,forward.peak_memory(MB),forward.latency(s),forward.throughput(samples/s),generate.latency(s),generate.throughput(tokens/s)
+0,35824.467968,0.684,23.4,12.9,248.0

raw_results/2023-08-11_19:07:08_fe3c8ab1af558b95f67f5fafc0c55f09fd2b09db/llama_1gpu_inference/7/main.log

@@ -0,0 +1,23 @@
+[2023-08-11 21:03:15,362][benchmark][INFO] - Configuring inference benchmark
+[2023-08-11 21:03:15,364][benchmark][INFO] - 	+ Setting seed(42)
+[2023-08-11 21:03:15,555][pytorch][INFO] - 	+ Infered AutoModel class AutoModelForCausalLM for task text-generation and model_type llama
+[2023-08-11 21:03:15,555][backend][INFO] - Configuring pytorch backend
+[2023-08-11 21:03:15,555][backend][INFO] - 	+ Checking initial device isolation
+[2023-08-11 21:03:15,660][backend][INFO] - 	+ Checking contineous device isolation
+[2023-08-11 21:03:15,696][pytorch][INFO] - 	+ Disabling gradients
+[2023-08-11 21:03:15,697][pytorch][INFO] - 	+ Loading pretrained model weights in dtype: float32 on device: cuda
+[2023-08-11 21:03:33,461][pytorch][INFO] - 	+ Turning on eval mode
+[2023-08-11 21:03:33,463][inference][INFO] - Running inference benchmark
+[2023-08-11 21:03:42,061][inference][INFO] - 	+ Tracking forward pass peak memory
+[2023-08-11 21:03:42,774][memory_tracker][INFO] - Peak memory usage: 35824.467968 MB
+[2023-08-11 21:03:42,774][inference][INFO] - 	+ Forward pass peak memory: 35824.467968 (MB)
+[2023-08-11 21:03:42,791][inference][INFO] - 	+ Warming up the forward pass
+[2023-08-11 21:04:08,164][inference][INFO] - 	+ Tracking forward pass latency and throughput
+[2023-08-11 21:05:26,189][inference][INFO] - 	+ Forward pass latency: 6.84e-01 (s)
+[2023-08-11 21:05:26,190][inference][INFO] - 	+ Forward pass throughput: 23.40 (samples/s)
+[2023-08-11 21:05:26,191][inference][INFO] - 	+ Warming up the generation pass
+[2023-08-11 21:05:39,863][inference][INFO] - 	+ Tracking generation latency and throughput
+[2023-08-11 21:06:05,763][inference][INFO] - 	+ Generation pass latency: 1.29e+01 (s)
+[2023-08-11 21:06:05,766][inference][INFO] - 	+ Generation pass throughput: 248.00 (tokens/s)
+[2023-08-11 21:06:05,766][inference][INFO] - Saving inference results
+[2023-08-11 21:06:05,773][backend][INFO] - Cleaning backend

raw_results/2023-08-11_19:07:08_fe3c8ab1af558b95f67f5fafc0c55f09fd2b09db/pytorch_bert_inference/0/hydra_config.yaml

@@ -0,0 +1,66 @@
+backend:
+  name: pytorch
+  version: 2.0.1+cu117
+  _target_: optimum_benchmark.backends.pytorch.PyTorchBackend
+  inter_op_num_threads: null
+  intra_op_num_threads: null
+  initial_isolation_check: true
+  continous_isolation_check: true
+  delete_cache: false
+  no_weights: false
+  torch_dtype: null
+  device_map: null
+  load_in_8bit: false
+  load_in_4bit: false
+  bettertransformer: false
+  torch_compile: false
+  torch_compile_config:
+    fullgraph: false
+    dynamic: false
+    backend: inductor
+    mode: null
+    options: null
+    disable: false
+  amp_autocast: false
+  amp_dtype: null
+  disable_grad: true
+  eval_mode: true
+benchmark:
+  name: inference
+  _target_: optimum_benchmark.benchmarks.inference.InferenceBenchmark
+  seed: 42
+  memory: true
+  warmup_runs: 10
+  benchmark_duration: 10
+  input_shapes:
+    batch_size: 1
+    sequence_length: 16
+    num_choices: 4
+    width: 64
+    height: 64
+    num_channels: 3
+    point_batch_size: 3
+    nb_points_per_image: 2
+    feature_size: 80
+    nb_max_frames: 3000
+    audio_sequence_length: 16000
+  new_tokens: 100
+experiment_name: pytorch_bert_inference
+model: hf-internal-testing/tiny-random-bert
+device: cpu
+task: text-classification
+hub_kwargs:
+  revision: main
+  cache_dir: null
+  force_download: false
+  local_files_only: false
+environment:
+  optimum_version: 1.11.0
+  transformers_version: 4.32.0.dev0
+  accelerate_version: 0.21.0
+  diffusers_version: null
+  python_version: 3.10.12
+  system: Linux
+  cpu: ' Intel(R) Xeon(R) Platinum 8275CL CPU @ 3.00GHz'
+  cpu_count: 96
+  cpu_ram_mb: 1204539.797504

raw_results/2023-08-11_19:07:08_fe3c8ab1af558b95f67f5fafc0c55f09fd2b09db/pytorch_bert_inference/0/inference_results.csv

@@ -0,0 +1,2 @@
+,forward.peak_memory(MB),forward.latency(s),forward.throughput(samples/s)
+0,459.993088,0.00362,276.0

raw_results/2023-08-11_19:07:08_fe3c8ab1af558b95f67f5fafc0c55f09fd2b09db/pytorch_bert_inference/0/main.log

@@ -0,0 +1,20 @@
+[2023-08-11 21:06:10,615][benchmark][INFO] - Configuring inference benchmark
+[2023-08-11 21:06:10,616][benchmark][INFO] - 	+ Setting seed(42)
+[2023-08-11 21:06:10,802][pytorch][INFO] - 	+ Infered AutoModel class AutoModelForSequenceClassification for task text-classification and model_type bert
+[2023-08-11 21:06:10,802][backend][INFO] - Configuring pytorch backend
+[2023-08-11 21:06:10,803][backend][INFO] - 	+ Checking initial device isolation
+[2023-08-11 21:06:10,803][backend][INFO] - 	+ Checking contineous device isolation
+[2023-08-11 21:06:10,804][pytorch][INFO] - 	+ Disabling gradients
+[2023-08-11 21:06:10,805][pytorch][INFO] - 	+ Loading pretrained model weights in dtype: None on device: cpu
+[2023-08-11 21:06:11,401][pytorch][INFO] - 	+ Turning on eval mode
+[2023-08-11 21:06:11,402][inference][INFO] - Running inference benchmark
+[2023-08-11 21:06:11,528][dummy_input][INFO] - Generating dummy input for: ['input_ids', 'attention_mask', 'token_type_ids']
+[2023-08-11 21:06:11,529][inference][INFO] - 	+ Tracking forward pass peak memory
+[2023-08-11 21:06:11,577][inference][INFO] - 	+ Forward pass peak memory: 459.993088 (MB)
+[2023-08-11 21:06:11,578][dummy_input][INFO] - Generating dummy input for: ['input_ids', 'attention_mask', 'token_type_ids']
+[2023-08-11 21:06:11,580][inference][INFO] - 	+ Warming up the forward pass
+[2023-08-11 21:06:11,615][inference][INFO] - 	+ Tracking forward pass latency and throughput
+[2023-08-11 21:06:21,714][inference][INFO] - 	+ Forward pass latency: 3.62e-03 (s)
+[2023-08-11 21:06:21,716][inference][INFO] - 	+ Forward pass throughput: 276.00 (samples/s)
+[2023-08-11 21:06:21,716][inference][INFO] - Saving inference results
+[2023-08-11 21:06:21,728][backend][INFO] - Cleaning backend

raw_results/2023-08-11_19:07:08_fe3c8ab1af558b95f67f5fafc0c55f09fd2b09db/pytorch_gpt2_inference/0/hydra_config.yaml

@@ -0,0 +1,66 @@
+backend:
+  name: pytorch
+  version: 2.0.1+cu117
+  _target_: optimum_benchmark.backends.pytorch.PyTorchBackend
+  inter_op_num_threads: null
+  intra_op_num_threads: null
+  initial_isolation_check: true
+  continous_isolation_check: true
+  delete_cache: false
+  no_weights: false
+  torch_dtype: null
+  device_map: null
+  load_in_8bit: false
+  load_in_4bit: false
+  bettertransformer: false
+  torch_compile: false
+  torch_compile_config:
+    fullgraph: false
+    dynamic: false
+    backend: inductor
+    mode: null
+    options: null
+    disable: false
+  amp_autocast: false
+  amp_dtype: null
+  disable_grad: true
+  eval_mode: true
+benchmark:
+  name: inference
+  _target_: optimum_benchmark.benchmarks.inference.InferenceBenchmark
+  seed: 42
+  memory: true
+  warmup_runs: 10
+  benchmark_duration: 10
+  input_shapes:
+    batch_size: 1
+    sequence_length: 16
+    num_choices: 4
+    width: 64
+    height: 64
+    num_channels: 3
+    point_batch_size: 3
+    nb_points_per_image: 2
+    feature_size: 80
+    nb_max_frames: 3000
+    audio_sequence_length: 16000
+  new_tokens: 100
+experiment_name: pytorch_gpt2_inference
+model: hf-internal-testing/tiny-random-gpt2
+device: cpu
+task: text-generation
+hub_kwargs:
+  revision: main
+  cache_dir: null
+  force_download: false
+  local_files_only: false
+environment:
+  optimum_version: 1.11.0
+  transformers_version: 4.32.0.dev0
+  accelerate_version: 0.21.0
+  diffusers_version: null
+  python_version: 3.10.12
+  system: Linux
+  cpu: ' Intel(R) Xeon(R) Platinum 8275CL CPU @ 3.00GHz'
+  cpu_count: 96
+  cpu_ram_mb: 1204539.797504

raw_results/2023-08-11_19:07:08_fe3c8ab1af558b95f67f5fafc0c55f09fd2b09db/pytorch_gpt2_inference/0/inference_results.csv

@@ -0,0 +1,2 @@
+,forward.peak_memory(MB),forward.latency(s),forward.throughput(samples/s),generate.latency(s),generate.throughput(tokens/s)
+0,463.560704,0.00379,264.0,0.521,192.0

raw_results/2023-08-11_19:07:08_fe3c8ab1af558b95f67f5fafc0c55f09fd2b09db/pytorch_gpt2_inference/0/main.log

@@ -0,0 +1,22 @@
+[2023-08-11 21:06:26,642][benchmark][INFO] - Configuring inference benchmark
+[2023-08-11 21:06:26,643][benchmark][INFO] - 	+ Setting seed(42)
+[2023-08-11 21:06:26,830][pytorch][INFO] - 	+ Infered AutoModel class AutoModelForCausalLM for task text-generation and model_type gpt2
+[2023-08-11 21:06:26,830][backend][INFO] - Configuring pytorch backend
+[2023-08-11 21:06:26,830][backend][INFO] - 	+ Checking initial device isolation
+[2023-08-11 21:06:26,830][backend][INFO] - 	+ Checking contineous device isolation
+[2023-08-11 21:06:26,832][pytorch][INFO] - 	+ Disabling gradients
+[2023-08-11 21:06:26,832][pytorch][INFO] - 	+ Loading pretrained model weights in dtype: None on device: cpu
+[2023-08-11 21:06:27,473][pytorch][INFO] - 	+ Turning on eval mode
+[2023-08-11 21:06:27,474][inference][INFO] - Running inference benchmark
+[2023-08-11 21:06:27,680][inference][INFO] - 	+ Tracking forward pass peak memory
+[2023-08-11 21:06:27,731][inference][INFO] - 	+ Forward pass peak memory: 463.560704 (MB)
+[2023-08-11 21:06:27,733][inference][INFO] - 	+ Warming up the forward pass
+[2023-08-11 21:06:27,765][inference][INFO] - 	+ Tracking forward pass latency and throughput
+[2023-08-11 21:06:37,859][inference][INFO] - 	+ Forward pass latency: 3.79e-03 (s)
+[2023-08-11 21:06:37,862][inference][INFO] - 	+ Forward pass throughput: 264.00 (samples/s)
+[2023-08-11 21:06:37,863][inference][INFO] - 	+ Warming up the generation pass
+[2023-08-11 21:06:38,448][inference][INFO] - 	+ Tracking generation latency and throughput
+[2023-08-11 21:06:48,860][inference][INFO] - 	+ Generation pass latency: 5.21e-01 (s)
+[2023-08-11 21:06:48,861][inference][INFO] - 	+ Generation pass throughput: 192.00 (tokens/s)
+[2023-08-11 21:06:48,861][inference][INFO] - Saving inference results
+[2023-08-11 21:06:48,874][backend][INFO] - Cleaning backend