Recent advancements in reasoning models, such as OpenAI\u2019s o1<\/a> and DeepSeek R1<\/a>, have propelled LLMs to achieve impressive performance through techniques like Chain of Thought<\/a> (CoT). However, the verbose nature of CoT leads to increased computational costs and latency. A novel paper published by Zoom Communications presents a new prompting technique called Chain of Draft (CoD). CoD focuses on concise, dense reasoning steps, reducing verbosity while maintaining accuracy. This approach mirrors human reasoning by prioritizing minimal, informative outputs, optimizing efficiency for real-world In this guide article we will explore this new prompting technique thoroughly and implement it\u00a0 using Gemini, Groq and Cohere API. And understand the differences between other prompting techniques and Chain of Draft prompting technique.<\/p>\n This article was published as a part of the\u00a0<\/strong><\/em>Data Science Blogathon.<\/strong><\/em><\/a><\/p>\n Chain of Draft (CoD) prompting is a novel approach to reasoning in large language models<\/a> (LLMs), inspired by how humans tackle complex tasks. Rather than generating verbose, step-by-step explanations like the Chain of Thought (CoT) method, CoD focuses on producing concise, critical insights at each step. This minimalist approach allows LLMs to advance toward solutions more efficiently, using fewer tokens and reducing latency, all while maintaining or even improving accuracy. <\/p>\n Introduced by researchers at Zoom Communications, CoD has shown significant improvements in cost-effectiveness and speed across tasks like arithmetic, common-sense reasoning, and symbolic problem-solving, making it a practical technique for real-world applications. One can read the published paper in detail here<\/a>.<\/p>\n Large Language Models (LLMs) have significantly advanced in their ability to perform complex reasoning tasks, owing much of their progress to various structured reasoning frameworks. One foundational method, Chain-of-Thought (CoT) reasoning, encourages models to articulate intermediate steps, thereby enhancing problem-solving capabilities. Building upon this, more sophisticated structures like tree and graph-based reasoning have been developed, allowing LLMs to tackle increasingly intricate problems by representing hierarchical and relational data more effectively. <\/p>\n Additionally, approaches such as self-consistency CoT incorporate verification and reflection mechanisms to bolster reasoning reliability, while ReAct integrates tool usage into the reasoning process, enabling LLMs to access external resources and knowledge. These innovations collectively expand the reasoning capabilities of LLMs across a diverse range of applications.\u00a0<\/p>\n Chain of Draft (CoD) Prompting is a minimalist reasoning technique designed to optimize the performance of large language models (LLMs) by reducing verbosity during the reasoning process while maintaining accuracy. The core idea behind CoD is inspired by how humans approach problem-solving: instead of articulating every detail in a step-by-step manner, we tend to use concise, shorthand notes or drafts that capture only the most crucial pieces of information. This approach helps to reduce cognitive load and enables faster progress toward a solution.<\/p>\n Concise Intermediate Steps<\/b>: CoD focuses on generating compact, dense outputs for each reasoning step, which capture only the essential information needed to move forward. This results in minimalistic drafts that help guide the model through problem-solving without unnecessary detail.<\/p>\n Cognitive Scaffolding<\/b>: Just as humans use shorthand to track their ideas, CoD externalizes critical Problem<\/b>: Jason had 20 lollipops. He gave Denny some. Now he has 12 left. How many did Jason give to Denny?\u00a0\u00a0<\/p>\n Response [CoD] :\u00a0<\/b>20\u201312 = 8 \u2192 Final Answer: 8.<\/p>\n<\/blockquote>\n As we can see above the response for the problem had very concise symbolic reasoning steps similar to what we do when we are doing problem solving.<\/p>\n Different prompting techniques enhance LLM reasoning in unique ways, from step-by-step logic to external knowledge integration and structured thought processes.<\/p>\n In standard prompting, the LLM generates a direct answer to a query without showing the intermediate reasoning steps. It provides the final output without revealing the thought process behind it.<\/p>\n Although this approach is efficient in terms of token usage, it lacks transparency. Without insight into how the model reached its conclusion, verifying correctness or identifying reasoning errors becomes With CoT prompting, the model offers an in-depth explanation of its reasoning process.<\/p>\n This response is thorough and transparent, outlining every step of the reasoning process. However, it is overly detailed, including redundant information that doesn\u2019t contribute computationally. This excess verbosity greatly increases token usage, resulting in higher latency and cost.<\/p>\n With CoD prompting, the model focuses exclusively on the essential reasoning steps, providing only the most critical information. This approach eliminates unnecessary details, ensuring efficiency while maintaining accuracy.<\/p>\n Below we will look into the advantages of chain of draft prompting:<\/p>\n Now we will see how we can implement the Chain of Draft prompting using different LLMs and methods.<\/p>\n We can implement Chain of Draft in different ways let us g through them:<\/p>\n We will now implement this in code using two different LLM Gemini and Groq API. Gr<\/p>\n Let us now implement these prompting techniques using Gemini to enhance reasoning, decision-making, and problem-solving capabilities.<\/p>\n For Gemini API Key visit Gemini Site<\/a>\u00a0Click on get an API Key\u00a0<\/i>\u00a0button as shown below in pic. You will be We basically need to install google genai library.<\/p>\n We import relevant packages and add API key as a environment variable.<\/p>\n Now we define the generate function and configure model, contents and generate_content_config .<\/i><\/p>\n Note in\u00a0generate_content_config\u00a0 we pass system instruction as \u201d Think step by step, but only keep a minimum draft for each thinking step, with 5 words at most. Return the answer at the end of the response after a separator ####.\u201d<\/p>\n Now we can execute the code using two methods one passing only system instruction prompt and question directly. Another is by passing one-shot example in prompt along with question and system instruction.<\/p>\n Response for Zero-shot CoD prompt from Gemini:<\/p>\n
applications.<\/p>\nLearning Objectives<\/h3>\n
\n
Introducing Chain of Draft Prompting\u00a0<\/h2>\n
Background on Other Prompting Techniques<\/h2>\n
Different Prompting Techniques<\/h3>\n
\n
Explaining Chain of Draft Prompting<\/h2>\n
Human-Centric Inspiration<\/h3>\n
\n
Mechanism of CoD<\/h3>\n
thoughts while avoiding the verbosity that typically burdens traditional reasoning models. The goal is to maintain the integrity of the reasoning pathway without overloading the model with excessive tokens.<\/p>\nExample of CoD<\/h4>\n
\n
Comparison Between different Prompting Techniques<\/h2>\n
Standard Prompting<\/h3>\n
![\"standard](\"https:\/\/cdn.analyticsvidhya.com\/wp-content\/uploads\/2025\/03\/standard-prompting.webp\")
<\/figure>\n![\"Standard](\"https:\/\/cdn.analyticsvidhya.com\/wp-content\/uploads\/2025\/03\/Standard-Prompting-Example.webp\")
<\/figure>\n
challenging, particularly for complex problems that require step-by-step reasoning.<\/p>\nChain of Thought(CoT) Prompting<\/h3>\n
![\"Chain](\"https:\/\/cdn.analyticsvidhya.com\/wp-content\/uploads\/2025\/03\/Chain-of-ThoughtCoT-Prompting.webp\")
<\/figure>\n![\"Chain](\"https:\/\/cdn.analyticsvidhya.com\/wp-content\/uploads\/2025\/03\/Chain-of-Thought-Prompting-example.webp\")
<\/figure>\nChain of Draft (CoD) Prompting<\/h3>\n
![\"Chain](\"https:\/\/cdn.analyticsvidhya.com\/wp-content\/uploads\/2025\/03\/Chain-of-Draft-CoD-Prompting.webp\")
<\/figure>\n![\"Chain](\"https:\/\/cdn.analyticsvidhya.com\/wp-content\/uploads\/2025\/03\/Chain-of-Draft-Prompting-example.webp\")
<\/figure>\nAdvantages of Chain of Draft (CoD) Prompting<\/h2>\n
\n
Code Implementation of CoD<\/h2>\n
Methods to Implement CoD<\/h3>\n
\n
Implementation using Gemini<\/h2>\n
Step 1: Generate Gemini API Key<\/h3>\n
redirected Google AI Studio<\/a>\u00a0where you will need to use your google account login and then find your API Key generated.<\/p>\n![\"Generate](\"https:\/\/cdn.analyticsvidhya.com\/wp-content\/uploads\/2025\/03\/Generate-Gemini-API-Key.webp\")
<\/figure>\nStep 2: Install Libraries<\/h3>\n
pip install google-genai<\/code><\/pre>\nStep 3: Import Packages and Setup API Key<\/h3>\n
import base64\nimport os\nfrom google import genai\nfrom google.genai import types\n\nos.environ[\"GEMINI_API_KEY\"] = \"Your Gemini API Key\"\n<\/code><\/pre>\nStep 4: Create Generate Function<\/h3>\n
def generate_gemini(example,question):\n client = genai.Client(\n api_key=os.environ.get(\"GEMINI_API_KEY\"),\n )\n\n model = \"gemini-2.0-flash\"\n contents = [\n types.Content(\n role=\"user\",\n parts=[\n types.Part.from_text(text=example),\n types.Part.from_text(text=question),\n ],\n ),\n ]\n generate_content_config = types.GenerateContentConfig(\n temperature=1,\n top_p=0.95,\n top_k=40,\n max_output_tokens=8192,\n response_mime_type=\"text\/plain\",\n system_instruction=[\n types.Part.from_text(text=\"\"\"Think step by step, but only keep a minimum draft for each thinking step, with 5 words at most. Return the answer at the end of the response after a separator ####.\"\"\"),\n ],\n )\n\n# Now pass the parameters to generate_content_stream function\n for chunk in client.models.generate_content_stream(\n model=model,\n contents=contents,\n config=generate_content_config,\n ):\n print(chunk.text, end=\"\")<\/code><\/pre>\nStep 5: Execute the Code\u00a0<\/h3>\n
if __name__ == \"__main__\":\n example = \"\"\"\"\"\"\n question =\"\"\"Q: Anita bought 3 apples and 4 oranges. Each apple costs $1.20 and each orange costs $0.80. How much did she spend in total?\nA:\"\"\"\n generate_gemini(example,question)<\/code><\/pre>\nApples cost: 3 * $1.20\nOranges cost: 4 * $0.80\nTotal: sum of both\n#### $6.80<\/code><\/pre>\nif __name__ == \"__main__\":\n example = \"\"\"Q: Jason had 20 lollipops. He gave Denny some lollipops. Now Jason has 12 lollipops. How many lollipops did Jason give to Denny?\nA: 20 - x = 12; x = 8. #### 8\"\"\"\n question =\"\"\"Q: Anita bought 3 apples and 4 oranges. Each apple costs $1.20 and each orange costs $0.80. How much did she spend in total?\nA:\"\"\"\n generate_gemini(example,question)\n<\/code><\/pre>\n
![\"Creating](\"https:\/\/cdn.analyticsvidhya.com\/wp-content\/uploads\/2025\/03\/Creating-Groq-API-Key.webp\")
<\/figure>\n![\"Ritika\"](\"https:\/\/av-eks-lekhak.s3.amazonaws.com\/media\/lekhak-profile-images\/converted_image_oHzb1Uq.webp\")
<\/p>\n