Tutorial: PhariaFinetuning

import json

from datasets import load_dataset

def convert_to_messages_format(conversation):
    """Convert a conversation with multiple turns to the desired message format."""
    messages = []

    # Process each turn in the conversation
    for turn in conversation:
        # Add user message
        messages.append({"role": "user", "content": turn["input"]})

        # Add assistant message
        messages.append({"role": "assistant", "content": turn["output"]})

    # Create the final dictionary structure
    return {"messages": messages}

def convert_dataset_to_jsonl(output_file="formatted_data.jsonl"):
    """Convert the entire dataset to JSONL format."""
    # Load the dataset
    ds = load_dataset("LDJnr/Pure-Dove")

    # Open the output file in write mode
    with open(output_file, "w", encoding="utf-8") as f:
        # Process each example in the training set
        for example in ds["train"]:
            # Convert the conversation to the desired format
            formatted_data = convert_to_messages_format(example["conversation"])

            # Write the formatted data as a JSON line
            json.dump(formatted_data, f, ensure_ascii=False)
            f.write("\n")

if __name__ == "__main__":
    # Convert the dataset and save to formatted_data.jsonl
    convert_dataset_to_jsonl()
    print("Conversion completed! Check formatted_data.jsonl for results.")

models:
  - name: models-<your-model-name> # must be lowercase
    pvcSize: 100Gi
    weights:
      - s3:
          # TODO: align this with PhariaFinetuning API Job object. This is the first part of the checkpoints field in the PhariaFinetuning API Job object
          endpoint: <your-storage-endpoint> # your storage endpoint
          # this is the second part of the checkpoints field in the PhariaFinetuning API Job object
          folder: <path to your model weights inside your storage> # has to end with checkpoint.ckpt
          targetDirectory: <your-model-name>

s3Credentials: # use the same credentials you use in the pharia-ai Helm chart values in pharia-finetuning
  accessKeyId: ""
  secretAccessKey: ""
  profile: "" # can be left empty
  region: ""

inference-worker:
  ...
  checkpoints:
    ...
    - generator:
        type: vllm
        pipeline_parallel_size: 1
        tensor_parallel_size: 1
        model_path: /models/<your-model-name>
      queue: <your-model-name>
      replicas: 1
      modelVolumeClaim: models-<your-model-name>

inference-api:
  ...
  modelsOverride:
    ...
      <your-model-name>:
        checkpoint: <your-model-name>
        experimental: false
        multimodal_enabled: false
        completion_type: full
        embedding_type: null
        maximum_completion_tokens: 8192
        adapter_name: null
        bias_name: null
        softprompt_name: null
        description: Your description here
        aligned: false
        chat_template:
          template: |-
            {% set loop_messages = messages %}{% for message in loop_messages %}{% set content = '<|start_header_id|>' + message['role'] + '<|end_header_id|>

            '+ message['content'] | trim + '<|eot_id|>' %}{% if loop.index0 == 0 %}{% set content = bos_token + content %}{% endif %}{{ content }}{% endfor %}{{ '<|start_header_id|>assistant<|end_header_id|>

            ' }}
          bos_token: <|begin_of_text|>
          eos_token: <|endoftext|>
        worker_type: vllm # this needs to be the same worker type as defined in step 2
        prompt_template: |-
          <|begin_of_text|>{% for message in messages %}<|start_header_id|>{{message.role}}<|end_header_id|>

          {% promptrange instruction %}{{message.content}}{% endpromptrange %}<|eot_id|>{% endfor %}<|start_header_id|>assistant<|end_header_id|>

          {% if response_prefix %}{{response_prefix}}{% endif %}