New to Claude Skills? Learn how to install them →

davila7 on GitHub

Speculative Decoding

Free

Accelerate LLM inference with advanced techniques.

Get this skill

Free · Opens the source repo

What Speculative Decoding does

Speculative Decoding is a skill designed to enhance the inference speed of large language models (LLMs) by leveraging advanced techniques like Medusa, lookahead decoding, and draft models. This skill is particularly beneficial for developers and data scientists who require faster response times in applications such as chatbots, code generation, or any real-time processing tasks. By applying these techniques, users can achieve inference speedups of 1.5 to 3.6 times without compromising the quality of the generated outputs.

The core of this skill revolves around using a smaller, faster draft model to generate candidate tokens, which are then evaluated by a larger target model. This parallel evaluation allows for rapid decision-making while maintaining high accuracy. Additionally, the Medusa approach incorporates multiple prediction heads to forecast future tokens, further optimizing the decoding process. The lookahead decoding technique, based on Jacobi iteration, also contributes to speed improvements by allowing the model to make educated guesses about upcoming tokens based on a specified lookahead window.

Installation is straightforward, requiring the setup of necessary libraries and models. The skill provides code snippets for quick implementation, enabling users to integrate speculative decoding into their own projects seamlessly. The documentation includes practical examples that illustrate how to use the skill effectively, whether through basic speculative decoding or advanced Medusa and lookahead techniques.

This skill is ideal for those working with large language models who need to optimize performance for applications that demand low latency and high throughput. It is especially useful in scenarios where computational resources are limited, allowing for efficient deployment of LLMs without the need for extensive hardware upgrades.

When to use it

Use this skill when you need to speed up LLM inference for real-time applications or when deploying models on limited hardware.

When not to use it

This skill may not be suitable for applications that do not require high-speed inference or where model quality is the highest priority without regard for speed.

What you can build with it

Real-time Chatbot Deployment

Utilize speculative decoding to enhance the response time of a chatbot, ensuring quick and accurate replies to user queries.

Optimizing Code Generation

Implement this skill in a code generation tool to reduce latency, allowing for faster code suggestions and completions.

Efficient Model Deployment on Limited Hardware

Deploy large language models on devices with restricted computational resources while maintaining high throughput and low latency.

How to install Speculative Decoding

View source

1. Install with the skills CLI

npx skills add davila7/claude-code-templates/emerging-techniques-speculative-decoding --agent claude-code

2. Or install it manually

Download the skill folder and drop it into ~/.claude/skills/ for all projects, or .claude/skills/ to scope it to one repo. Restart Claude Code so it picks up the new skill.

Anthropic's agentic coding CLI, and the reference implementation of Agent Skills. Drop a skill folder into ~/.claude/skills and Claude Code loads it automatically whenever a task matches the skill's description. Claude Code docs

Inside SKILL.md

Written by davila7

Speculative Decoding: Accelerating LLM Inference

When to Use This Skill

Use Speculative Decoding when you need to:

  • Speed up inference by 1.5-3.6× without quality loss
  • Reduce latency for real-time applications (chatbots, code generation)
  • Optimize throughput for high-volume serving
  • Deploy efficiently on limited hardware
  • Generate faster without changing model architecture

Key Techniques: Draft model speculative decoding, Medusa (multiple heads), Lookahead Decoding (Jacobi iteration)

Papers: Medusa (arXiv 2401.10774), Lookahead Decoding (ICML 2024), Speculative Decoding Survey (ACL 2024)

Installation

# Standard speculative decoding (transformers)
pip install transformers accelerate

# Medusa (multiple decoding heads)
git clone https://github.com/FasterDecoding/Medusa
cd Medusa
pip install -e .

# Lookahead Decoding
git clone https://github.com/hao-ai-lab/LookaheadDecoding
cd LookaheadDecoding
pip install -e .

# Optional: vLLM with speculative decoding
pip install vllm

Quick Start

Basic Speculative Decoding (Draft Model)

from transformers import AutoModelForCausalLM, AutoTokenizer

# Load target model (large, slow)
target_model = AutoModelForCausalLM.from_pretrained(
    "meta-llama/Llama-2-70b-hf",
    device_map="auto",
    torch_dtype=torch.float16
)

# Load draft model (small, fast)
draft_model = AutoModelForCausalLM.from_pretrained(
    "meta-llama/Llama-2-7b-hf",
    device_map="auto",
    torch_dtype=torch.float16
)

tokenizer = AutoTokenizer.from_pretrained("meta-llama/Llama-2-70b-hf")

# Generate with speculative decoding
prompt = "Explain quantum computing in simple terms:"
inputs = tokenizer(prompt, return_tensors="pt").to("cuda")

# Transformers 4.36+ supports assisted generation
outputs = target_model.generate(
    **inputs,
    assistant_model=draft_model,  # Enable speculative decoding
    max_new_tokens=256,
    do_sample=True,
    temperature=0.7,
)

response = tokenizer.decode(outputs[0], skip_special_tokens=True)
print(response)

Medusa (Multiple Decoding Heads)

from medusa.model.medusa_model import MedusaModel

# Load Medusa-enhanced model
model = MedusaModel.from_pretrained(
    "FasterDecoding/medusa-vicuna-7b-v1.3",  # Pre-trained with Medusa heads
    torch_dtype=torch.float16,
    device_map="auto"
)

tokenizer = AutoTokenizer.from_pretrained("FasterDecoding/medusa-vicuna-7b-v1.3")

# Generate with Medusa (2-3× speedup)
prompt = "Write a Python function to calculate fibonacci numbers:"
inputs = tokenizer(prompt, return_tensors="pt").to("cuda")

outputs = model.medusa_generate(
    **inputs,
    max_new_tokens=256,
    temperature=0.7,
    posterior_threshold=0.09,  # Acceptance threshold
    posterior_alpha=0.3,       # Tree construction parameter
)

response = tokenizer.decode(outputs[0], skip_special_tokens=True)

Lookahead Decoding (Jacobi Iteration)

from lookahead.lookahead_decoding import LookaheadDecoding

# Load model
model = AutoModelForCausalLM.from_pretrained(
    "meta-llama/Llama-2-7b-hf",
    torch_dtype=torch.float16,
    device_map="auto"
)
tokenizer = AutoTokenizer.from_pretrained("meta-llama/Llama-2-7b-hf")

# Initialize lookahead decoding
lookahead = LookaheadDecoding(
    model=model,
    tokenizer=tokenizer,
    window_size=15,    # Lookahead window (W)
    ngram_size=5,      # N-gram size (N)
    guess_size=5       # Number of parallel guesses
)

# Generate (1.5-2.3× speedup)
prompt = "Implement quicksort in Python:"
output = lookahead.generate(prompt, max_new_tokens=256)
print(output)

Core Concepts

1. Speculative Decoding (Draft Model)

Idea: Use small draft model to generate candidates, large target model to verify in parallel.

Algorithm:

  1. Draft model generates K tokens speculatively
  2. Target model evaluates all K tokens in parallel (single forward pass)
  3. Accept tokens where draft and target agree
  4. Reject first disagreement, continue from there
def speculative_decode(target_model, draft_model, prompt, K=4):
    """Speculative decoding algorithm."""
    # 1. Generate K draft tokens
    draft_tokens = draft_model.generate(prompt, max_new_tokens=K)

    # 2. Target model evaluates all K tokens in one forward pass
    target_logits = target_model(draft_tokens)  # Parallel!

    # 3. Accept/reject based on probability match
    accepted = []
    for i in range(K):
        p_draft = softmax(draft_model.logits[i])
        p_target = softmax(target_logits[i])

        # Acceptance probability
        if random.random() < min(1, p_target[draft_tokens[i]] / p_draft[draft_tokens[i]]):
            accepted.append(draft_tokens[i])
        else:
            break  # Reject, resample from target

    return accepted

Performance:

  • Speedup: 1.5-2× with good draft model
  • Zero quality loss (mathematically equivalent to target model)
  • Best when draft model is 5-10× smaller than target

2. Medusa (Multiple Decoding Heads)

Source: arXiv 2401.10774 (2024)

Innovation: Add multiple prediction heads to existing model, predict future tokens without separate draft model.

Architecture:

Input → Base LLM (frozen) → Hidden State
                                ├→ Head 1 (predicts token t+1)
                                ├→ Head 2 (predicts token t+2)
                                ├→ Head 3 (predicts token t+3)
                                └→ Head 4 (predicts token t+4)

Training:

  • Medusa-1: Freeze base LLM, train only heads
    • 2.2× speedup, lossless
  • Medusa-2: Fine-tune base LLM + heads together
    • 2.3-3.6× speedup, better quality

Tree-based Attention:

# Medusa constructs tree of candidates
# Example: Predict 2 steps ahead with top-2 per step

#         Root
#        /    \
#      T1a    T1b  (Step 1: 2 candidates)
#     /  \    / \
#  T2a  T2b T2c T2d  (Step 2: 4 candidates total)

# Single forward pass evaluates entire tree!

Advantages:

  • No separate draft model needed
  • Minimal training (only heads)
  • Compatible with any LLM

3. Lookahead Decoding (Jacobi Iteration)

Source: ICML 2024

Core idea: Reformulate autoregressive decoding as solving system of equations, solve in parallel using Jacobi iteration.

Mathematical formulation:

Traditional:  y_t = f(x, y_1, ..., y_{t-1})  (sequential)
Jacobi:       y_t^{(k+1)} = f(x, y_1^{(k)}, ..., y_{t-1}^{(k)})  (parallel)

Two branches:

  1. Lookahead Branch: Generate n-grams in parallel

    • Window size W: How many steps to look ahead
    • N-gram size N: How many past tokens to use
  2. Verification Branch: Verify promising n-grams

    • Match n-grams with generated tokens
    • Accept if first token matches
class LookaheadDecoding:
    def __init__(self, model, window_size=15, ngram_size=5):
        self.model = model
        self.W = window_size  # Lookahead window
        self.N = ngram_size   # N-gram size

    def generate_step(self, tokens):
        # Lookahead branch: Generate W × N candidates
        candidates = {}
        for w in range(1, self.W + 1):
            for n in range(1, self.N + 1):
                # Generate n-gram starting at position w
                ngram = self.generate_ngram(tokens, start=w, length=n)
                candidates[(w, n)] = ngram

        # Verification branch: Find matching n-grams
        verified = []
        for ngram in candidates.values():
            if ngram[0] == tokens[-1]:  # First token matches last input
                if self.verify(tokens, ngram):
                    verified.append(ngram)

        # Accept longest verified n-gram
        return max(verified, key=len) if verified else [self.model.generate_next(tokens)]

Performance:

  • Speedup: 1.5-2.3× (up to 3.6× for code generation)
  • No draft model or training needed
  • Works out-of-the-box with any model

Method Comparison

MethodSpeedupTraining NeededDraft ModelQuality Loss
Draft Model Speculative1.5-2×NoYes (external)None
Medusa2-3.6×Minimal (heads only)No (built-in heads)None
Lookahead1.5-2.3×NoneNoNone
Naive Batching1.2-1.5×NoNoNone

Advanced Patterns

Training Medusa Heads

from medusa.model.medusa_model import MedusaModel
from medusa.model.kv_cache import initialize_past_key_values
import torch.nn as nn

# 1. Load base model
base_model = AutoModelForCausalLM.from_pretrained(
    "lmsys/vicuna-7b-v1.3",
    torch_dtype=torch.float16
)

# 2. Add Medusa heads
num_heads = 4
medusa_heads = nn.ModuleList([
    nn.Linear(base_model.config.hidden_size, base_model.config.vocab_size, bias=False)
    for _ in range(num_heads)
])

# 3. Training loop (freeze base model for Medusa-1)
for param in base_model.parameters():
    param.requires_grad = False  # Freeze base

optimizer = torch.optim.Adam(medusa_heads.parameters(), lr=1e-3)

for batch in dataloader:
    # Forward pass
    hidden_states = base_model(**batch, output_hidden_states=True).hidden_states[-1]

    # Predict future tokens with each head
    loss = 0
    for i, head in enumerate(medusa_heads):
        logits = head(hidden_states)
        # Target: tokens shifted by (i+1) positions
        target = batch['input_ids'][:, i+1:]
        loss += F.cross_entropy(logits[:, :-i-1], target)

    # Backward
    optimizer.zero_grad()
    loss.backward()
    optimizer.step()

Hybrid: Speculative + Medusa

# Use Medusa as draft model for speculative decoding
draft_medusa = MedusaModel.from_pretrained("medusa-vicuna-7b")
target_model = AutoModelForCausalLM.from_pretrained("vicuna-33b")

# Draft generates multiple candidates with Medusa
draft_tokens = draft_medusa.medusa_generate(prompt, max_new_tokens=5)

# Target verifies in single forward pass
outputs = target_model.generate(
    prompt,
    assistant_model=draft_medusa,  # Use Medusa as draft
    max_new_tokens=256
)

# Combines benefits: Medusa speed + large model quality

Optimal Draft Model Selection

def select_draft_model(target_model_size, target):
    """Select optimal draft model for speculative decoding."""
    # Rule: Draft should be 5-10× smaller
    if target_model_size == "70B":
        return "7B"  # 10× smaller
    elif target_model_size == "33B":
        return "7B"  # 5× smaller
    elif target_model_size == "13B":
        return "1B"  # 13× smaller
    else:
        return None  # Target too small, use Medusa/Lookahead instead

# Example
draft = select_draft_model("70B", target_model)
# Returns "7B" → Use Llama-2-7b as draft for Llama-2-70b

Best Practices

1. Choose the Right Method

# New deployment → Medusa (best overall speedup, no draft model)
if deploying_new_model:
    use_method = "Medusa"

# Existing deployment with small model available → Draft speculative
elif have_small_version_of_model:
    use_method = "Draft Model Speculative"

# Want zero training/setup → Lookahead
elif want_plug_and_play:
    use_method = "Lookahead Decoding"

2. Hyperparameter Tuning

Draft Model Speculative:

# K = number of speculative tokens
K = 4  # Good default
K = 2  # Conservative (higher acceptance)
K = 8  # Aggressive (lower acceptance, but more when accepted)

# Rule: Larger K → more speedup IF draft model is good

Medusa:

# Posterior threshold (acceptance confidence)
posterior_threshold = 0.09  # Standard (from paper)
posterior_threshold = 0.05  # More conservative (slower, higher quality)
posterior_threshold = 0.15  # More aggressive (faster, may degrade quality)

# Tree depth (how many steps ahead)
medusa_choices = [[0], [0, 0], [0, 1], [0, 0, 0]]  # Depth 3 (standard)

Lookahead:

# Window size W (lookahead distance)
# N-gram size N (context for generation)

# 7B model (more resources)
W, N = 15, 5

# 13B model (moderate)
W, N = 10, 5

# 33B+ model (limited resources)
W, N = 7, 5

3. Production Deployment

# vLLM with speculative decoding
from vllm import LLM, SamplingParams

# Initialize with draft model
llm = LLM(
    model="meta-llama/Llama-2-70b-hf",
    speculative_model="meta-llama/Llama-2-7b-hf",  # Draft model
    num_speculative_tokens=5,
    use_v2_block_manager=True,
)

# Generate
prompts = ["Tell me about AI:", "Explain quantum physics:"]
sampling_params = SamplingParams(temperature=0.7, max_tokens=256)

outputs = llm.generate(prompts, sampling_params)
for output in outputs:
    print(output.outputs[0].text)

Resources

See Also

  • references/draft_model.md - Draft model selection and training
  • references/medusa.md - Medusa architecture and training
  • references/lookahead.md - Lookahead decoding implementation details

Frequently asked questions about Speculative Decoding

Similar skills