The Full Machine Learning Pipeline

From raw bytes to deployed model, every ML system follows the same ten-stage pipeline. Master it and you can read any architecture paper.

50 min · Reviewed 2026

Ten Stages, One Pipeline

Nearly every production ML system, from spam filters to LLMs, flows through the same skeleton. Knowing the skeleton lets you place any new paper or product into context quickly.

The stages

Data collection
Data cleaning and labeling
Feature engineering or tokenization
Train/validation/test split
Model architecture selection
Training with a loss function and optimizer
Evaluation on held-out data
Fine-tuning or post-training alignment
Deployment to inference infrastructure
Monitoring, feedback collection, and retraining

Where time actually goes

Newcomers assume training is the main event. In practice, data prep and evaluation consume 70 to 90 percent of engineering time. Training is usually a well-defined, scriptable step. Cleaning messy real-world data is not.

Stage	Typical time share
Data collection and cleaning	30-50%
Feature/token engineering	10-20%
Model training	5-15%
Evaluation and iteration	20-30%
Deployment and monitoring	10-20%

Concrete example: fine-tuning for a legal use case

from datasets import load_dataset
from transformers import AutoTokenizer, AutoModelForCausalLM, Trainer, TrainingArguments

dataset = load_dataset("legal_memos", split="train")
tok = AutoTokenizer.from_pretrained("base-model")
model = AutoModelForCausalLM.from_pretrained("base-model")

def preprocess(ex):
    return tok(ex["text"], truncation=True, max_length=2048)

ds = dataset.map(preprocess, batched=True)

args = TrainingArguments(
    output_dir="./legal-tuned",
    per_device_train_batch_size=4,
    num_train_epochs=3,
    learning_rate=2e-5,
    evaluation_strategy="epoch",
)

trainer = Trainer(model=model, args=args, train_dataset=ds)
trainer.train()A minimal fine-tuning loop using the Hugging Face stack.

Train vs. inference infrastructure

Training: massive parallel GPUs, weeks of runtime, measured in FLOPs
Inference: latency-sensitive, often needs quantization or distillation
Fine-tuning: smaller GPUs, hours to days, often with LoRA adapters
Monitoring: logging, drift detection, A/B tests in production

Common pipeline failures

Train/serve skew: features built differently in training vs. production
Data leakage: test data accidentally appears in training
Distribution drift: real inputs change over time, model decays
Unmonitored bias: groups see worse outcomes and nobody notices

Production ML is 5 percent machine learning and 95 percent engineering.
— A Google research engineer

The big idea: the ML pipeline is the real substrate of AI products. Papers describe stages 5 and 6. Careers are built on stages 1, 2, 7, 9, and 10.

End-of-lesson check

15 questions · take it digitally for instant feedback at tendril.neural-forge.io/learn/quiz/end-creators-full-ml-pipeline

What is the core idea behind "The Full Machine Learning Pipeline"?
1. From raw bytes to deployed model, every ML system follows the same ten-stage pipeline. Master it and you can read any architecture paper.
2. Following a set of rules you just gave it
3. letters
4. Apply RoPE in your foundations workflow to get better results
Which term best describes a foundational idea in "The Full Machine Learning Pipeline"?
1. preprocessing
2. pipeline
3. fine-tuning
4. inference
A learner studying The Full Machine Learning Pipeline would need to understand which concept?
1. pipeline
2. fine-tuning
3. preprocessing
4. inference
Which of these is directly relevant to The Full Machine Learning Pipeline?
1. pipeline
2. preprocessing
3. inference
4. fine-tuning
Which of the following is a key point about The Full Machine Learning Pipeline?
1. Data collection
2. Data cleaning and labeling
3. Feature engineering or tokenization
4. Train/validation/test split
Which of these does NOT belong in a discussion of The Full Machine Learning Pipeline?
1. Data collection
2. Data cleaning and labeling
3. Feature engineering or tokenization
4. Following a set of rules you just gave it
Which statement is accurate regarding The Full Machine Learning Pipeline?
1. Inference: latency-sensitive, often needs quantization or distillation
2. Fine-tuning: smaller GPUs, hours to days, often with LoRA adapters
3. Training: massive parallel GPUs, weeks of runtime, measured in FLOPs
4. Monitoring: logging, drift detection, A/B tests in production
Which of these does NOT belong in a discussion of The Full Machine Learning Pipeline?
1. Following a set of rules you just gave it
2. Fine-tuning: smaller GPUs, hours to days, often with LoRA adapters
3. Training: massive parallel GPUs, weeks of runtime, measured in FLOPs
4. Inference: latency-sensitive, often needs quantization or distillation
What is the key insight about "Post-training is where behavior is shaped" in the context of The Full Machine Learning Pipeline?
1. Pretraining gives the model raw capability. Instruction tuning, RLHF, and DPO shape its behavior.
2. Following a set of rules you just gave it
3. letters
4. Apply RoPE in your foundations workflow to get better results
What is the recommended tip about "Ground your practice in fundamentals" in the context of The Full Machine Learning Pipeline?
1. Following a set of rules you just gave it
2. Every AI capability has an underlying mechanism. Understanding that mechanism tells you where it'll fail — which is more…
3. letters
4. Apply RoPE in your foundations workflow to get better results
Which statement accurately describes an aspect of The Full Machine Learning Pipeline?
1. Following a set of rules you just gave it
2. letters
3. Nearly every production ML system, from spam filters to LLMs, flows through the same skeleton.
4. Apply RoPE in your foundations workflow to get better results
What does working with The Full Machine Learning Pipeline typically involve?
1. Following a set of rules you just gave it
2. letters
3. Apply RoPE in your foundations workflow to get better results
4. Newcomers assume training is the main event. In practice, data prep and evaluation consume 70 to 90 percent of engineering time.
Which of the following is true about The Full Machine Learning Pipeline?
1. The big idea: the ML pipeline is the real substrate of AI products. Papers describe stages 5 and 6.
2. Following a set of rules you just gave it
3. letters
4. Apply RoPE in your foundations workflow to get better results
Which best describes the scope of "The Full Machine Learning Pipeline"?
1. It is unrelated to foundations workflows
2. It focuses on From raw bytes to deployed model, every ML system follows the same ten-stage pipeline. Master it and
3. It applies only to the opposite beginner tier
4. It was deprecated in 2024 and no longer relevant
Which section heading best belongs in a lesson about The Full Machine Learning Pipeline?
1. Following a set of rules you just gave it
2. letters
3. The stages
4. Apply RoPE in your foundations workflow to get better results

← Back to interactive lesson

Tendril · Creators · AI Foundations

The Full Machine Learning Pipeline

From raw bytes to deployed model, every ML system follows the same ten-stage pipeline. Master it and you can read any architecture paper.

50 min · Reviewed 2026

Ten Stages, One Pipeline

Nearly every production ML system, from spam filters to LLMs, flows through the same skeleton. Knowing the skeleton lets you place any new paper or product into context quickly.

The stages

Data collection
Data cleaning and labeling
Feature engineering or tokenization
Train/validation/test split
Model architecture selection
Training with a loss function and optimizer
Evaluation on held-out data
Fine-tuning or post-training alignment
Deployment to inference infrastructure
Monitoring, feedback collection, and retraining

Where time actually goes

Stage	Typical time share
Data collection and cleaning	30-50%
Feature/token engineering	10-20%
Model training	5-15%
Evaluation and iteration	20-30%
Deployment and monitoring	10-20%

Concrete example: fine-tuning for a legal use case

from datasets import load_dataset
from transformers import AutoTokenizer, AutoModelForCausalLM, Trainer, TrainingArguments

dataset = load_dataset("legal_memos", split="train")
tok = AutoTokenizer.from_pretrained("base-model")
model = AutoModelForCausalLM.from_pretrained("base-model")

def preprocess(ex):
    return tok(ex["text"], truncation=True, max_length=2048)

ds = dataset.map(preprocess, batched=True)

args = TrainingArguments(
    output_dir="./legal-tuned",
    per_device_train_batch_size=4,
    num_train_epochs=3,
    learning_rate=2e-5,
    evaluation_strategy="epoch",
)

trainer = Trainer(model=model, args=args, train_dataset=ds)
trainer.train()A minimal fine-tuning loop using the Hugging Face stack.

Train vs. inference infrastructure

Training: massive parallel GPUs, weeks of runtime, measured in FLOPs
Inference: latency-sensitive, often needs quantization or distillation
Fine-tuning: smaller GPUs, hours to days, often with LoRA adapters
Monitoring: logging, drift detection, A/B tests in production

Common pipeline failures

Train/serve skew: features built differently in training vs. production
Data leakage: test data accidentally appears in training
Distribution drift: real inputs change over time, model decays
Unmonitored bias: groups see worse outcomes and nobody notices

Production ML is 5 percent machine learning and 95 percent engineering.
— A Google research engineer

The big idea: the ML pipeline is the real substrate of AI products. Papers describe stages 5 and 6. Careers are built on stages 1, 2, 7, 9, and 10.

End-of-lesson check

15 questions · take it digitally for instant feedback at tendril.neural-forge.io/learn/quiz/end-creators-full-ml-pipeline

What is the core idea behind "The Full Machine Learning Pipeline"?
1. From raw bytes to deployed model, every ML system follows the same ten-stage pipeline. Master it and you can read any architecture paper.
2. Following a set of rules you just gave it
3. letters
4. Apply RoPE in your foundations workflow to get better results
Which term best describes a foundational idea in "The Full Machine Learning Pipeline"?
1. preprocessing
2. pipeline
3. fine-tuning
4. inference
A learner studying The Full Machine Learning Pipeline would need to understand which concept?
1. pipeline
2. fine-tuning
3. preprocessing
4. inference
Which of these is directly relevant to The Full Machine Learning Pipeline?
1. pipeline
2. preprocessing
3. inference
4. fine-tuning
Which of the following is a key point about The Full Machine Learning Pipeline?
1. Data collection
2. Data cleaning and labeling
3. Feature engineering or tokenization
4. Train/validation/test split
Which of these does NOT belong in a discussion of The Full Machine Learning Pipeline?
1. Data collection
2. Data cleaning and labeling
3. Feature engineering or tokenization
4. Following a set of rules you just gave it
Which statement is accurate regarding The Full Machine Learning Pipeline?
1. Inference: latency-sensitive, often needs quantization or distillation
2. Fine-tuning: smaller GPUs, hours to days, often with LoRA adapters
3. Training: massive parallel GPUs, weeks of runtime, measured in FLOPs
4. Monitoring: logging, drift detection, A/B tests in production
Which of these does NOT belong in a discussion of The Full Machine Learning Pipeline?
1. Following a set of rules you just gave it
2. Fine-tuning: smaller GPUs, hours to days, often with LoRA adapters
3. Training: massive parallel GPUs, weeks of runtime, measured in FLOPs
4. Inference: latency-sensitive, often needs quantization or distillation
What is the key insight about "Post-training is where behavior is shaped" in the context of The Full Machine Learning Pipeline?
1. Pretraining gives the model raw capability. Instruction tuning, RLHF, and DPO shape its behavior.
2. Following a set of rules you just gave it
3. letters
4. Apply RoPE in your foundations workflow to get better results
What is the recommended tip about "Ground your practice in fundamentals" in the context of The Full Machine Learning Pipeline?
1. Following a set of rules you just gave it
2. Every AI capability has an underlying mechanism. Understanding that mechanism tells you where it'll fail — which is more…
3. letters
4. Apply RoPE in your foundations workflow to get better results
Which statement accurately describes an aspect of The Full Machine Learning Pipeline?
1. Following a set of rules you just gave it
2. letters
3. Nearly every production ML system, from spam filters to LLMs, flows through the same skeleton.
4. Apply RoPE in your foundations workflow to get better results
What does working with The Full Machine Learning Pipeline typically involve?
1. Following a set of rules you just gave it
2. letters
3. Apply RoPE in your foundations workflow to get better results
4. Newcomers assume training is the main event. In practice, data prep and evaluation consume 70 to 90 percent of engineering time.
Which of the following is true about The Full Machine Learning Pipeline?
1. The big idea: the ML pipeline is the real substrate of AI products. Papers describe stages 5 and 6.
2. Following a set of rules you just gave it
3. letters
4. Apply RoPE in your foundations workflow to get better results
Which best describes the scope of "The Full Machine Learning Pipeline"?
1. It is unrelated to foundations workflows
2. It focuses on From raw bytes to deployed model, every ML system follows the same ten-stage pipeline. Master it and
3. It applies only to the opposite beginner tier
4. It was deprecated in 2024 and no longer relevant
Which section heading best belongs in a lesson about The Full Machine Learning Pipeline?
1. Following a set of rules you just gave it
2. letters
3. The stages
4. Apply RoPE in your foundations workflow to get better results

← Back to interactive lesson