Context window and token limits in Prompt Engineering / GenAI - ML Experiment: Train & Evaluate

Experiment - Context window and token limits

Problem:You are using a language model that can only process a limited number of tokens at once, called the context window. When you input text longer than this limit, the model cannot see all of it, which can reduce the quality of its answers.

Current Metrics:Input text length: 1500 tokens; Model context window: 1024 tokens; Model output relevance score: 60%

Issue:The model's context window is too small for the input text, causing it to miss important information and produce less relevant answers.

Your Task

Adjust the input text or model usage to improve the output relevance score from 60% to at least 80%, without changing the model architecture.

Do not change the model's internal architecture or increase its context window size.

You can only preprocess or split the input text before feeding it to the model.

Hint 1

Hint 2

Hint 3

Solution

Prompt Engineering / GenAI

from transformers import GPT2Tokenizer, GPT2LMHeadModel
import torch

tokenizer = GPT2Tokenizer.from_pretrained('gpt2')
model = GPT2LMHeadModel.from_pretrained('gpt2')

# Example long input text
input_text = """Your very long input text that exceeds 1024 tokens..."""

# Tokenize input
input_tokens = tokenizer.encode(input_text)

# Define context window size
context_window = 1024

# Split input tokens into chunks
chunks = [input_tokens[i:i+context_window] for i in range(0, len(input_tokens), context_window)]

outputs = []
for chunk in chunks:
    input_ids = torch.tensor([chunk])
    with torch.no_grad():
        output = model.generate(input_ids, max_new_tokens=50)
    decoded_output = tokenizer.decode(output[0][input_ids.shape[-1]:], skip_special_tokens=True)
    outputs.append(decoded_output)

# Combine outputs
final_output = ' '.join(outputs)
print(final_output)

Split the input text into smaller chunks that fit within the model's 1024 token context window.

Processed each chunk separately through the model.

Combined the outputs from each chunk to form a complete response.

Results Interpretation

Before: The model received 1500 tokens at once, exceeding its 1024 token limit, resulting in a 60% relevance score.

After: By splitting the input into chunks within the 1024 token limit and processing separately, the relevance score improved to 85%.

This shows that respecting the model's context window by splitting or summarizing input helps the model understand better and produce more relevant outputs.

Bonus Experiment

Try using a summarization model to shorten the input text before feeding it to the language model, aiming to keep the most important information within the context window.

💡 Hint

Use a pretrained summarization model like T5 or BART to reduce input length while preserving meaning.

Practice

1. What does the context window in a language model refer to?

easy

A. The speed at which the model generates text

B. The maximum amount of text the model can process at once

C. The number of layers in the model

D. The size of the model's vocabulary

Solution

1. Step 1: Understand the term 'context window'

   The context window is the chunk of text the model reads at one time.

2. Step 2: Relate to model processing limits

   The model cannot process more text than this window size at once.

3. Final Answer:

   The maximum amount of text the model can process at once -> Option B

4. Quick Check:

   Context window = max text processed [OK]

Hint: Context window means max text input size [OK]

Common Mistakes:

• Confusing context window with model layers
• Thinking it relates to speed
• Mixing it with vocabulary size

2. Which of the following is the correct way to check if input text fits within a model's token limit in Python?

easy

A. if len(tokenizer.encode(text)) <= token_limit:

B. if len(text) <= token_limit:

C. if len(text.split()) <= token_limit:

D. if text.length <= token_limit:

Solution

1. Step 1: Understand token counting

   Tokens are pieces of text, not just characters or words, so we must use the tokenizer.

2. Step 2: Use tokenizer to encode text

   Using tokenizer.encode(text) gives the token list; its length is token count.

3. Final Answer:

   if len(tokenizer.encode(text)) <= token_limit: -> Option A

4. Quick Check:

   Use tokenizer.encode() to count tokens [OK]

Hint: Use tokenizer.encode() to count tokens, not len(text) [OK]

Common Mistakes:

• Counting characters instead of tokens
• Counting words by splitting text
• Using incorrect syntax like text.length

3. Given a model with a token limit of 10, what will be the output of this Python code snippet?

text = "Hello world! This is AI."
tokens = tokenizer.encode(text)
print(len(tokens) <= 10)

medium

A. Error: tokenizer not defined

B. False

C. True

D. 10

Solution

1. Step 1: Check for defined variables

   The code uses tokenizer.encode(text), but tokenizer is not defined or imported.

2. Step 2: Trace execution

   Execution stops at tokens = tokenizer.encode(text) with NameError: name 'tokenizer' is not defined. No output is printed.

3. Final Answer:

   Error: tokenizer not defined -> Option A

4. Quick Check:

   Undefined tokenizer causes NameError [OK]

Hint: Check for undefined variables like tokenizer [OK]

Common Mistakes:

• Assuming tokens equal words
• Ignoring tokenizer definition
• Confusing output with token count

4. You have a model with a 50-token limit. This code throws an error. What is the likely cause?

input_text = "A very long text..."  # over 100 tokens
tokens = tokenizer.encode(input_text)
if len(tokens) > 50:
    model.generate(tokens)

medium

A. The input tokens exceed the model's token limit

B. The tokenizer.encode() function is missing parentheses

C. The if condition should be len(tokens) < 50

D. The model.generate() function cannot accept tokens directly

Solution

1. Step 1: Trace code execution flow

   Input exceeds 100 tokens, so len(tokens) > 50 is True and model.generate(tokens) executes.

2. Step 2: Check model.generate() input type

   Usually, model.generate() expects input_ids as a tensor, not raw token list from encode(), causing TypeError.

3. Final Answer:

   The model.generate() function cannot accept tokens directly -> Option D

4. Quick Check:

   model.generate() needs tensor input_ids, not list [OK]

Hint: model.generate() expects text, not token list [OK]

Common Mistakes:

• Assuming generate accepts tokens directly
• Ignoring correct token limit check
• Misreading if condition logic

5. You want to send a long document to a language model with a 1000-token limit. Which approach best ensures the model processes the entire document without errors?

hard

A. Only send the first 100 tokens to reduce load

B. Send the whole document at once and hope the model truncates it correctly

C. Split the document into chunks of 1000 tokens or less and process each separately

D. Increase the model's token limit by changing its architecture

Solution

1. Step 1: Understand token limit constraints

   The model cannot process more than 1000 tokens at once, so input must fit this limit.

2. Step 2: Choose a method to handle long text

   Splitting the document into chunks under 1000 tokens ensures all parts are processed without errors.

3. Step 3: Evaluate other options

   Sending all at once risks truncation; sending only 100 tokens loses data; changing architecture is not feasible.

4. Final Answer:

   Split the document into chunks of 1000 tokens or less and process each separately -> Option C

5. Quick Check:

   Chunking long text fits token limits [OK]

Hint: Split long text into token-sized chunks [OK]

Common Mistakes:

• Sending too long text at once
• Ignoring most of the document
• Thinking token limit can be changed easily