Cookbook: Supervised Fine-Tuning

Run supervised fine-tuning on Fireworks using the cookbook sft_loop recipe to handle dataset loading, tokenization, gradient accumulation, and checkpointing.

​

What this is

Supervised Fine-Tuning (SFT) trains the model to produce desired outputs by minimizing cross-entropy loss on (prompt, response) pairs. The cookbook’s sft_loop recipe handles data loading, tokenization, batching, gradient accumulation, and checkpointing automatically.

​

Using the recipe

from training.recipes.sft_loop import Config, main
from training.utils import TrainerConfig

cfg = Config(
    log_path="./sft_logs",
    base_model="accounts/fireworks/models/qwen3-8b",
    dataset="/path/to/training_data.jsonl",
    tokenizer_model="Qwen/Qwen3-8B",
    max_seq_len=4096,
    epochs=1,
    batch_size=4,
    learning_rate=1e-5,
    trainer=TrainerConfig(
        training_shape_id="accounts/fireworks/trainingShapes/qwen3-8b-128k",
    ),
)

main(cfg)

batch_size controls examples per optimizer batch. SFT keeps controlled dataset batches and does not need rollout streaming.

​

Dataset format

SFT datasets use the standard messages format (JSONL with one example per line):

{"messages": [
  {"role": "user", "content": "What is the capital of France?"},
  {"role": "assistant", "content": "The capital of France is Paris."}
]}

Multi-turn conversations are supported:

{"messages": [
  {"role": "system", "content": "You are a helpful assistant."},
  {"role": "user", "content": "Hello"},
  {"role": "assistant", "content": "Hi! How can I help?"},
  {"role": "user", "content": "What is 2+2?"},
  {"role": "assistant", "content": "2+2 = 4"}
]}

The recipe automatically tokenizes conversations using the chat template, masking context tokens and normalizing trainable response loss within each example.

​

Vision datasets

The SFT recipe also supports vision-language model training. Use multimodal content arrays with image_url objects in your JSONL, and specify a VLM training shape and tokenizer. See Vision Inputs for dataset format details and a full walkthrough.

​

Checkpointing and resume

By default, sft_loop needs only a trainer. It supports DCP checkpointing and resume; optional generation_eval=True samples held-out text after training using SDK-managed sampling resources:

from training.utils import TrainerConfig, WandBConfig

cfg = Config(
    log_path="./sft_logs",
    base_model="accounts/fireworks/models/qwen3-8b",
    dataset="/path/to/training_data.jsonl",
    tokenizer_model="Qwen/Qwen3-8B",
    max_seq_len=4096,
    epochs=1,
    batch_size=4,
    trainer=TrainerConfig(
        training_shape_id="accounts/fireworks/trainingShapes/qwen3-8b-128k",
    ),
    dcp_save_interval=50,
    init_from_checkpoint="previous-job-id:step-100",  # optional
    wandb=WandBConfig(entity="my-team", project="sft-experiment"),
)

main(cfg)

​

Operational guidance

• Set trainer.training_shape_id — cookbook trainer launches use training shapes.
• Only one trainer job needed — SFT does not require a reference trainer.
• No sampling deployment is required by default — optional generation evaluation uses the SDK-managed sampler.
• Use batch_size to control the number of examples per optimizer step.
• Resume: Attach to the same trainer with TrainerConfig(job_id=...) and keep the same log_path; the recipe restores the saved training state and data cursor. See checkpoints.
• Periodic checkpoints are disabled by default (dcp_save_interval=0); successful runs save final state by default. Set a positive interval (e.g., 50) to recover progress from an interrupted run.

​

Related guides

• Vision Inputs — VLM training with image and text data
• Cookbook DPO — preference optimization
• Cookbook RL (GRPO) — reinforcement learning recipes
• Cookbook Reference — all config classes and parameters
• Loss Functions — API-level SFT loss details