Update olive configs to match new format (#114)

* Update phi2 olive config to match phi3 olive config format * Changes to match updated template * Remove attn_implementation * Remove sliding_window * Unpin phi2 version * Update transformers version and use olive-ai from main (phi2 only as initial test) * Updating phi2 requirements.txt * Tag phi2 package versions * Updating all requirements.txt with latest versions * Update llama-v2-7b olive-config * Update mistral-7b olive-config * Update phi 1.5 olive-config * Formatting/spacing * Update zephyr-7b-beta olive-config * Small updates to phi3 olive-config * Forgot to remove olive-ai * Remove reporting to wandb
microsoft · May 14, 2024 · ad8779d · ad8779d
1 parent d9eebb1
commit ad8779d
Show file tree

Hide file tree

Showing 12 changed files with 289 additions and 167 deletions.
diff --git a/configs/Phi-3-mini-4k-instruct/finetuning/olive-config.json b/configs/Phi-3-mini-4k-instruct/finetuning/olive-config.json
@@ -2,6 +2,9 @@
  "input_model":{
  "type": "PyTorchModel",
  "config": {
+ "model_script": "finetuning/qlora_user_script.py",
+ "io_config": "get_merged_decoder_with_past_io_config",
+ "dummy_inputs_func": "get_merged_decoder_with_past_dummy_inputs",
  "hf_config": {
  "model_name": "model-cache/microsoft/Phi-3-mini-4k-instruct",
  "task": "text-generation",
@@ -23,10 +26,11 @@
  }
  }
  },
- "data_configs": {
- "dataset_default_train": {
- "name": "dataset_default",
+ "data_configs": [
+ {
+ "name": "dataset_default_train",
  "type": "HuggingfaceContainer",
+ "user_script": "finetuning/qlora_user_script.py",
  "params_config": {
 #data_configs_data_files_extension_start
  <!-- "data_name": "<data_configs_data_files_extension>", -->
@@ -46,7 +50,7 @@
  }
  }
  }
- },
+ ],
  "passes": {
  "qlora": {
  "type": "QLoRA",
@@ -75,8 +79,7 @@
  "group_by_length": true,
  "adam_beta2": 0.999,
  "max_grad_norm": 0.3,
- "output_dir": "<output_dir>",
- "report_to": "wandb"
+ "output_dir": "<output_dir>"
  }
  }
  }

diff --git a/configs/Phi-3-mini-4k-instruct/setup/requirements.txt b/configs/Phi-3-mini-4k-instruct/setup/requirements.txt
@@ -1,20 +1,21 @@
 --extra-index-url https://download.pytorch.org/whl/cu118
-torch
-torchvision
-torchaudio
-transformers
-accelerate
-bitsandbytes
-datasets
-peft
-scikit-learn
-sentencepiece
-trl
-protobuf
-ipykernel
-wandb
-onnxruntime-gpu
-olive-ai
-promptflow
-promptflow-tools
-gradio
+torch==2.3.0+cu118
+torchvision==0.18.0+cu118
+torchaudio==2.3.0+cu118
+transformers==4.40.2
+accelerate==0.30.0
+bitsandbytes==0.43.1
+datasets==2.19.1
+peft==0.10.0
+scikit-learn==1.4.2
+sentencepiece==0.2.0
+trl==0.8.6
+protobuf==3.20.3
+ipykernel==6.29.4
+wandb==0.17.0
+onnxruntime-gpu==1.17.1
+einops==0.8.0
+promptflow==1.10.1
+promptflow-tools==1.4.0
+gradio==4.29.0
+git+https://github.com/microsoft/Olive.git@ebe2f7ebbf8278698efddf45803e061fd097babc
diff --git a/configs/llama-v2-7b/finetuning/olive-config.json b/configs/llama-v2-7b/finetuning/olive-config.json
@@ -2,16 +2,46 @@
  "input_model":{
  "type": "PyTorchModel",
  "config": {
+ "model_script": "finetuning/qlora_user_script.py",
+ "io_config": "get_merged_decoder_with_past_io_config",
+ "dummy_inputs_func": "get_merged_decoder_with_past_dummy_inputs",
  "hf_config": {
  "model_name": "model-cache/meta-llama/Llama-2-7b",
- "task": "text-generation"
+ "task": "text-generation",
+ "from_pretrained_args": {
+ "trust_remote_code": true
+ }
  }
  }
  },
- "data_configs": {
- "dataset-default_train": {
- "name": "dataset-default",
+ "systems": {
+ "local_system": {
+ "type": "LocalSystem",
+ "config": {
+ "accelerators": [
+ {
+ "device": "gpu"
+ }
+ ]
+ }
+ },
+ "target_system": {
+ "type": "LocalSystem",
+ "config": {
+ "accelerators": [
+ {
+ "device": "gpu",
+ "execution_providers": ["CPUExecutionProvider"]
+ }
+ ]
+ }
+ }
+ },
+ "data_configs": [
+ {
+ "name": "dataset_default_train",
  "type": "HuggingfaceContainer",
+ "user_script": "finetuning/qlora_user_script.py",
  "params_config": {
 #data_configs_data_files_extension_start
  <!-- "data_name": "<data_configs_data_files_extension>", -->
@@ -31,7 +61,7 @@
  }
  }
  }
- },
+ ],
  "passes": {
  "qlora": {
  "type": "QLoRA",
@@ -42,7 +72,7 @@
  "lora_r": <lora_r>,
  "lora_alpha": <lora_alpha>,
  "lora_dropout": <lora_dropout>,
- "train_data_config": "dataset-default_train",
+ "train_data_config": "dataset_default_train",
  "eval_dataset_size": <eval_dataset_size>,
  "training_args": {
  "seed": <training_args_seed>,
@@ -69,13 +99,8 @@
  "log_severity_level": 0,
  "search_strategy": false,
  "evaluate_input_model": false,
- "target": {
- "type": "LocalSystem",
- "config": {
- "accelerators": ["gpu"]
- }
- },
- "execution_providers": ["CPUExecutionProvider"],
+ "host": "local_system",
+ "target": "target_system",
  "cache_dir": "cache",
  "output_dir" : "models/qlora"
  }

diff --git a/configs/llama-v2-7b/setup/requirements.txt b/configs/llama-v2-7b/setup/requirements.txt
@@ -1,21 +1,21 @@
 --extra-index-url https://download.pytorch.org/whl/cu118
-torch
-torchvision
-torchaudio
-transformers==4.34.1
-accelerate==0.23.0
-bitsandbytes==0.41.1
-datasets==2.14.5
-peft==0.5.0
-scikit-learn==1.3.1
-sentencepiece==0.1.99
-trl==0.7.2
+torch==2.3.0+cu118
+torchvision==0.18.0+cu118
+torchaudio==2.3.0+cu118
+transformers==4.40.2
+accelerate==0.30.0
+bitsandbytes==0.43.1
+datasets==2.19.1
+peft==0.10.0
+scikit-learn==1.4.2
+sentencepiece==0.2.0
+trl==0.8.6
 protobuf==3.20.3
-ipykernel==6.25.2
-wandb==0.15.12
-onnxruntime-gpu==1.16.1
-einops
-olive-ai==0.3.3
-promptflow==0.1.0b8
-promptflow-tools==0.1.0b10
-gradio
+ipykernel==6.29.4
+wandb==0.17.0
+onnxruntime-gpu==1.17.1
+einops==0.8.0
+promptflow==1.10.1
+promptflow-tools==1.4.0
+gradio==4.29.0
+git+https://github.com/microsoft/Olive.git@ebe2f7ebbf8278698efddf45803e061fd097babc
diff --git a/configs/mistral-7b/finetuning/olive-config.json b/configs/mistral-7b/finetuning/olive-config.json
@@ -2,16 +2,46 @@
  "input_model":{
  "type": "PyTorchModel",
  "config": {
+ "model_script": "finetuning/qlora_user_script.py",
+ "io_config": "get_merged_decoder_with_past_io_config",
+ "dummy_inputs_func": "get_merged_decoder_with_past_dummy_inputs",
  "hf_config": {
  "model_name": "model-cache/mistralai/Mistral-7B",
- "task": "text-generation"
+ "task": "text-generation",
+ "from_pretrained_args": {
+ "trust_remote_code": true
+ }
  }
  }
  },
- "data_configs": {
- "dataset-default_train": {
- "name": "dataset-default",
+ "systems": {
+ "local_system": {
+ "type": "LocalSystem",
+ "config": {
+ "accelerators": [
+ {
+ "device": "gpu"
+ }
+ ]
+ }
+ },
+ "target_system": {
+ "type": "LocalSystem",
+ "config": {
+ "accelerators": [
+ {
+ "device": "gpu",
+ "execution_providers": ["CPUExecutionProvider"]
+ }
+ ]
+ }
+ }
+ },
+ "data_configs": [
+ {
+ "name": "dataset_default_train",
  "type": "HuggingfaceContainer",
+ "user_script": "finetuning/qlora_user_script.py",
  "params_config": {
 #data_configs_data_files_extension_start
  <!-- "data_name": "<data_configs_data_files_extension>", -->
@@ -31,7 +61,7 @@
  }
  }
  }
- },
+ ],
  "passes": {
  "qlora": {
  "type": "QLoRA",
@@ -42,7 +72,7 @@
  "lora_r": <lora_r>,
  "lora_alpha": <lora_alpha>,
  "lora_dropout": <lora_dropout>,
- "train_data_config": "dataset-default_train",
+ "train_data_config": "dataset_default_train",
  "eval_dataset_size": <eval_dataset_size>,
  "training_args": {
  "seed": <training_args_seed>,
@@ -69,13 +99,8 @@
  "log_severity_level": 0,
  "search_strategy": false,
  "evaluate_input_model": false,
- "target": {
- "type": "LocalSystem",
- "config": {
- "accelerators": ["gpu"]
- }
- },
- "execution_providers": ["CPUExecutionProvider"],
+ "host": "local_system",
+ "target": "target_system",
  "cache_dir": "cache",
  "output_dir" : "models/qlora"
  }

diff --git a/configs/phi-1_5/finetuning/olive-config.json b/configs/phi-1_5/finetuning/olive-config.json
@@ -2,19 +2,46 @@
  "input_model":{
  "type": "PyTorchModel",
  "config": {
+ "model_script": "finetuning/qlora_user_script.py",
+ "io_config": "get_merged_decoder_with_past_io_config",
+ "dummy_inputs_func": "get_merged_decoder_with_past_dummy_inputs",
  "hf_config": {
  "model_name": "model-cache/microsoft/phi-1_5",
  "task": "text-generation",
- "model_loading_args": {
+ "from_pretrained_args": {
  "trust_remote_code": true
  }
  }
  }
  },
- "data_configs": {
- "dataset-default_train": {
- "name": "dataset-default",
+ "systems": {
+ "local_system": {
+ "type": "LocalSystem",
+ "config": {
+ "accelerators": [
+ {
+ "device": "gpu"
+ }
+ ]
+ }
+ },
+ "target_system": {
+ "type": "LocalSystem",
+ "config": {
+ "accelerators": [
+ {
+ "device": "gpu",
+ "execution_providers": ["CPUExecutionProvider"]
+ }
+ ]
+ }
+ }
+ },
+ "data_configs": [
+ {
+ "name": "dataset_default_train",
  "type": "HuggingfaceContainer",
+ "user_script": "finetuning/qlora_user_script.py",
  "params_config": {
 #data_configs_data_files_extension_start
  <!-- "data_name": "<data_configs_data_files_extension>", -->
@@ -34,7 +61,7 @@
  }
  }
  }
- },
+ ],
  "passes": {
  "qlora": {
  "type": "QLoRA",
@@ -45,7 +72,7 @@
  "lora_r": <lora_r>,
  "lora_alpha": <lora_alpha>,
  "lora_dropout": <lora_dropout>,
- "train_data_config": "dataset-default_train",
+ "train_data_config": "dataset_default_train",
  "eval_dataset_size": <eval_dataset_size>,
  "training_args": {
  "seed": <training_args_seed>,
@@ -72,13 +99,8 @@
  "log_severity_level": 0,
  "search_strategy": false,
  "evaluate_input_model": false,
- "target": {
- "type": "LocalSystem",
- "config": {
- "accelerators": ["gpu"]
- }
- },
- "execution_providers": ["CPUExecutionProvider"],
+ "host": "local_system",
+ "target": "target_system",
  "cache_dir": "cache",
  "output_dir" : "models/qlora"
  }