Skip to content

Commit

Permalink
address comments
Browse files Browse the repository at this point in the history
  • Loading branch information
@srsaggam
srsaggam committed Sep 3, 2024
1 parent 1a7cf3d commit e928398
Showing 1 changed file with 17 additions and 16 deletions.
33 changes: 17 additions & 16 deletions ...on/jobs/finetuning/standalone/chat-completion/chat_completion_with_model_as_service.ipynb
View file
Original file line number Diff line number Diff line change
Expand Up @@ -110,9 +110,9 @@
"source": [
"### 2. Pick a foundation model to fine tune\n",
"\n",
"`Phi-3-mini-4k-instruct` is a 3.8B parameters, lightweight, state-of-the-art open model built upon datasets used for Phi-2. The model belongs to the Phi-3 model family, and the Mini version comes in two variants 4K and 128K which is the context length (in tokens) it can support, we need to finetune the model for our specific purpose in order to use it. You can browse these models in the Model Catalog in the AzureML Studio, filtering by the `chat-completion` task. In this example, we use the `Phi-3-mini-4k-instruct` model. If you have opened this notebook for a different model, replace the model name and version accordingly.\n",
"`Phi-3-mini-4k-instruct` is a 3.8B parameters, lightweight, state-of-the-art open model built upon datasets used for Phi-2. The model belongs to the Phi-3 model family, and the Mini version comes in two variants 4K and 128K which is the context length (in tokens) it can support. You can browse these models in the Model Catalog in the Azure AI Studio, filtering by the `chat-completion` task. In this example, we use the `Phi-3-mini-4k-instruct` model. If you have opened this notebook for a different model, replace the model name and version accordingly.\n",
"\n",
"Note the model id property of the model. This will be passed as input to the fine tuning job. This is also available as the `Asset ID` field in model details page in AzureML Studio Model Catalog."
"Note the model id property of the model. This will be passed as input to the fine tuning job. This is also available as the `Asset ID` field in model details page in Azure AI Studio Model Catalog."
]
},
{
Expand All @@ -121,7 +121,7 @@
"metadata": {},
"outputs": [],
"source": [
"model_name = \"Meta-Llama-3.1-8B-Instruct\" # \"Phi-3-mini-4k-instruct\"\n",
"model_name = \"Phi-3-mini-4k-instruct\" #\"Meta-Llama-3.1-8B-Instruct\" \n",
"foundation_model = registry_ml_client.models.get(model_name, label=\"latest\")\n",
"print(\n",
" \"\\n\\nUsing model name: {0}, version: {1}, id: {2} for fine tuning\".format(\n",
Expand All @@ -148,8 +148,8 @@
"source": [
"### 3. Prepare data\n",
"\n",
"- The [download-dataset.py](./download-dataset.py) is used to download the ultrachat_200k dataset and transform the dataset into finetune pipeline component consumable format. Also as the dataset is large, hence we here have only part of the dataset.\n",
"- Running the below script only downloads 5% of the data. This can be increased by changing `dataset_split_pc` parameter to desired percenetage.\n",
"- The [download-dataset.py](./download-dataset.py) is used to download the ultrachat_200k dataset and transform the dataset into the format expected by model. Also as the dataset is large, hence we here have only part of the dataset.\n",
"- Running the below script downloads only 1% of the data because the dataset is very large. This can be increased by changing `dataset_split_pc` parameter to desired percentage.\n",
"\n",
"**Note** : Some language models have different language codes and hence the column names in the dataset should reflect the same.\n",
"\n",
Expand Down Expand Up @@ -238,7 +238,7 @@
"dataset_version = \"1\"\n",
"train_dataset_name = f\"{dataset_dir}_train\"\n",
"try:\n",
" train_data_created = workspace_ml_client.data.get(\n",
" train_data_asset= workspace_ml_client.data.get(\n",
" train_dataset_name, version=dataset_version\n",
" )\n",
" print(f\"Dataset {train_dataset_name} already exists\")\n",
Expand All @@ -251,7 +251,7 @@
" name=train_dataset_name,\n",
" version=\"1\",\n",
" )\n",
" train_data_created = workspace_ml_client.data.create_or_update(train_data)"
" train_data_asset = workspace_ml_client.data.create_or_update(train_data)"
]
},
{
Expand All @@ -265,7 +265,7 @@
"dataset_version = \"1\"\n",
"validation_dataset_name = f\"{dataset_dir}_validation\"\n",
"try:\n",
" validation_data_created = workspace_ml_client.data.get(\n",
" validation_data_asset= workspace_ml_client.data.get(\n",
" validation_dataset_name, version=dataset_version\n",
" )\n",
" print(f\"Dataset {validation_dataset_name} already exists\")\n",
Expand All @@ -278,7 +278,7 @@
" name=validation_dataset_name,\n",
" version=\"1\",\n",
" )\n",
" validation_data_created = workspace_ml_client.data.create_or_update(validation_data)"
" validation_data_asset = workspace_ml_client.data.create_or_update(validation_data)"
]
},
{
Expand All @@ -290,12 +290,12 @@
"from azure.ai.ml.entities._inputs_outputs import Input\n",
"\n",
"training_data = Input(\n",
" type=train_data_created.type,\n",
" path=f\"azureml://locations/{workspace.location}/workspaces/{workspace._workspace_id}/data/{train_data_created.name}/versions/{train_data_created.version}\",\n",
" type=train_data_asset.type,\n",
" path=f\"azureml://locations/{workspace.location}/workspaces/{workspace._workspace_id}/data/{train_data_asset.name}/versions/{train_data_asset.version}\",\n",
")\n",
"validation_data = Input(\n",
" type=validation_data_created.type,\n",
" path=f\"azureml://locations/{workspace.location}/workspaces/{workspace._workspace_id}/data/{validation_data_created.name}/versions/{validation_data_created.version}\",\n",
" type=validation_data_asset.type,\n",
" path=f\"azureml://locations/{workspace.location}/workspaces/{workspace._workspace_id}/data/{validation_data_asset.name}/versions/{validation_data_asset.version}\",\n",
")"
]
},
Expand All @@ -313,7 +313,7 @@
"metadata": {},
"source": [
"##### Create marketplace subscription for 3P models\n",
"Note: Skip this step for 1P models that are offered on Azure. Example: Phi family of models"
"Note: Skip this step for 1P(Microsoft) models that are offered on Azure. Example: Phi family of models"
]
},
{
Expand Down Expand Up @@ -365,8 +365,9 @@
"\n",
"1. `model` - Base model to finetune.\n",
"2. `training_data` - Training data for finetuning the base model.\n",
"3. `task` - FineTuning task to perform. eg. TEXT_COMPLETION for text-generation/text-generation finetuning jobs.\n",
"4. `outputs`- Output registered model name.\n",
"3. `validation_data` - Validation data for finetuning the base model.\n",
"4. `task` - FineTuning task to perform. eg. TEXT_COMPLETION for text-generation/text-generation finetuning jobs.\n",
"5. `outputs`- Output registered model name.\n",
"\n",
"##### Following parameters are optional:\n",
"\n",
Expand Down

0 comments on commit e928398

Please sign in to comment.