optimizing-attention-flash

11
1
Source

Optimizes transformer attention with Flash Attention for 2-4x speedup and 10-20x memory reduction. Use when training/running transformers with long sequences (>512 tokens), encountering GPU memory issues with attention, or need faster inference. Supports PyTorch native SDPA, flash-attn library, H100 FP8, and sliding window attention.

Install

mkdir -p .claude/skills/optimizing-attention-flash && curl -L -o skill.zip "https://mcp.directory/api/skills/download/2622" && unzip -o skill.zip -d .claude/skills/optimizing-attention-flash && rm skill.zip

Installs to .claude/skills/optimizing-attention-flash

About this skill

Flash Attention - Fast Memory-Efficient Attention

Quick start

Flash Attention provides 2-4x speedup and 10-20x memory reduction for transformer attention through IO-aware tiling and recomputation.

PyTorch native (easiest, PyTorch 2.2+):

import torch
import torch.nn.functional as F

q = torch.randn(2, 8, 512, 64, device='cuda', dtype=torch.float16)  # [batch, heads, seq, dim]
k = torch.randn(2, 8, 512, 64, device='cuda', dtype=torch.float16)
v = torch.randn(2, 8, 512, 64, device='cuda', dtype=torch.float16)

# Automatically uses Flash Attention if available
out = F.scaled_dot_product_attention(q, k, v)

flash-attn library (more features):

pip install flash-attn --no-build-isolation
from flash_attn import flash_attn_func

# q, k, v: [batch, seqlen, nheads, headdim]
out = flash_attn_func(q, k, v, dropout_p=0.0, causal=True)

Common workflows

Workflow 1: Enable in existing PyTorch model

Copy this checklist:

Flash Attention Integration:
- [ ] Step 1: Check PyTorch version (≥2.2)
- [ ] Step 2: Enable Flash Attention backend
- [ ] Step 3: Verify speedup with profiling
- [ ] Step 4: Test accuracy matches baseline

Step 1: Check PyTorch version

python -c "import torch; print(torch.__version__)"
# Should be ≥2.2.0

If <2.2, upgrade:

pip install --upgrade torch

Step 2: Enable Flash Attention backend

Replace standard attention:

# Before (standard attention)
attn_weights = torch.softmax(q @ k.transpose(-2, -1) / math.sqrt(d_k), dim=-1)
out = attn_weights @ v

# After (Flash Attention)
import torch.nn.functional as F
out = F.scaled_dot_product_attention(q, k, v, attn_mask=mask)

Force Flash Attention backend:

with torch.backends.cuda.sdp_kernel(
    enable_flash=True,
    enable_math=False,
    enable_mem_efficient=False
):
    out = F.scaled_dot_product_attention(q, k, v)

Step 3: Verify speedup with profiling

import torch.utils.benchmark as benchmark

def test_attention(use_flash):
    q, k, v = [torch.randn(2, 8, 2048, 64, device='cuda', dtype=torch.float16) for _ in range(3)]

    if use_flash:
        with torch.backends.cuda.sdp_kernel(enable_flash=True):
            return F.scaled_dot_product_attention(q, k, v)
    else:
        attn = (q @ k.transpose(-2, -1) / 8.0).softmax(dim=-1)
        return attn @ v

# Benchmark
t_flash = benchmark.Timer(stmt='test_attention(True)', globals=globals())
t_standard = benchmark.Timer(stmt='test_attention(False)', globals=globals())

print(f"Flash: {t_flash.timeit(100).mean:.3f}s")
print(f"Standard: {t_standard.timeit(100).mean:.3f}s")

Expected: 2-4x speedup for sequences >512 tokens.

Step 4: Test accuracy matches baseline

# Compare outputs
q, k, v = [torch.randn(1, 8, 512, 64, device='cuda', dtype=torch.float16) for _ in range(3)]

# Flash Attention
out_flash = F.scaled_dot_product_attention(q, k, v)

# Standard attention
attn_weights = torch.softmax(q @ k.transpose(-2, -1) / 8.0, dim=-1)
out_standard = attn_weights @ v

# Check difference
diff = (out_flash - out_standard).abs().max()
print(f"Max difference: {diff:.6f}")
# Should be <1e-3 for float16

Workflow 2: Use flash-attn library for advanced features

For multi-query attention, sliding window, or H100 FP8.

Copy this checklist:

flash-attn Library Setup:
- [ ] Step 1: Install flash-attn library
- [ ] Step 2: Modify attention code
- [ ] Step 3: Enable advanced features
- [ ] Step 4: Benchmark performance

Step 1: Install flash-attn library

# NVIDIA GPUs (CUDA 12.0+)
pip install flash-attn --no-build-isolation

# Verify installation
python -c "from flash_attn import flash_attn_func; print('Success')"

Step 2: Modify attention code

from flash_attn import flash_attn_func

# Input: [batch_size, seq_len, num_heads, head_dim]
# Transpose from [batch, heads, seq, dim] if needed
q = q.transpose(1, 2)  # [batch, seq, heads, dim]
k = k.transpose(1, 2)
v = v.transpose(1, 2)

out = flash_attn_func(
    q, k, v,
    dropout_p=0.1,
    causal=True,  # For autoregressive models
    window_size=(-1, -1),  # No sliding window
    softmax_scale=None  # Auto-scale
)

out = out.transpose(1, 2)  # Back to [batch, heads, seq, dim]

Step 3: Enable advanced features

Multi-query attention (shared K/V across heads):

from flash_attn import flash_attn_func

# q: [batch, seq, num_q_heads, dim]
# k, v: [batch, seq, num_kv_heads, dim]  # Fewer KV heads
out = flash_attn_func(q, k, v)  # Automatically handles MQA

Sliding window attention (local attention):

# Only attend to window of 256 tokens before/after
out = flash_attn_func(
    q, k, v,
    window_size=(256, 256),  # (left, right) window
    causal=True
)

Step 4: Benchmark performance

import torch
from flash_attn import flash_attn_func
import time

q, k, v = [torch.randn(4, 4096, 32, 64, device='cuda', dtype=torch.float16) for _ in range(3)]

# Warmup
for _ in range(10):
    _ = flash_attn_func(q, k, v)

# Benchmark
torch.cuda.synchronize()
start = time.time()
for _ in range(100):
    out = flash_attn_func(q, k, v)
    torch.cuda.synchronize()
end = time.time()

print(f"Time per iteration: {(end-start)/100*1000:.2f}ms")
print(f"Memory allocated: {torch.cuda.max_memory_allocated()/1e9:.2f}GB")

Workflow 3: H100 FP8 optimization (FlashAttention-3)

For maximum performance on H100 GPUs.

FP8 Setup:
- [ ] Step 1: Verify H100 GPU available
- [ ] Step 2: Install flash-attn with FP8 support
- [ ] Step 3: Convert inputs to FP8
- [ ] Step 4: Run with FP8 attention

Step 1: Verify H100 GPU

nvidia-smi --query-gpu=name --format=csv
# Should show "H100" or "H800"

Step 2: Install flash-attn with FP8 support

pip install flash-attn --no-build-isolation
# FP8 support included for H100

Step 3: Convert inputs to FP8

import torch

q = torch.randn(2, 4096, 32, 64, device='cuda', dtype=torch.float16)
k = torch.randn(2, 4096, 32, 64, device='cuda', dtype=torch.float16)
v = torch.randn(2, 4096, 32, 64, device='cuda', dtype=torch.float16)

# Convert to float8_e4m3 (FP8)
q_fp8 = q.to(torch.float8_e4m3fn)
k_fp8 = k.to(torch.float8_e4m3fn)
v_fp8 = v.to(torch.float8_e4m3fn)

Step 4: Run with FP8 attention

from flash_attn import flash_attn_func

# FlashAttention-3 automatically uses FP8 kernels on H100
out = flash_attn_func(q_fp8, k_fp8, v_fp8)
# Result: ~1.2 PFLOPS, 1.5-2x faster than FP16

When to use vs alternatives

Use Flash Attention when:

  • Training transformers with sequences >512 tokens
  • Running inference with long context (>2K tokens)
  • GPU memory constrained (OOM with standard attention)
  • Need 2-4x speedup without accuracy loss
  • Using PyTorch 2.2+ or can install flash-attn

Use alternatives instead:

  • Standard attention: Sequences <256 tokens (overhead not worth it)
  • xFormers: Need more attention variants (not just speed)
  • Memory-efficient attention: CPU inference (Flash Attention needs GPU)

Common issues

Issue: ImportError: cannot import flash_attn

Install with no-build-isolation flag:

pip install flash-attn --no-build-isolation

Or install CUDA toolkit first:

conda install cuda -c nvidia
pip install flash-attn --no-build-isolation

Issue: Slower than expected (no speedup)

Flash Attention benefits increase with sequence length:

  • <512 tokens: Minimal speedup (10-20%)
  • 512-2K tokens: 2-3x speedup
  • 2K tokens: 3-4x speedup

Check sequence length is sufficient.

Issue: RuntimeError: CUDA error

Verify GPU supports Flash Attention:

import torch
print(torch.cuda.get_device_capability())
# Should be ≥(7, 5) for Turing+

Flash Attention requires:

  • Ampere (A100, A10): ✅ Full support
  • Turing (T4): ✅ Supported
  • Volta (V100): ❌ Not supported

Issue: Accuracy degradation

Check dtype is float16 or bfloat16 (not float32):

q = q.to(torch.float16)  # Or torch.bfloat16

Flash Attention uses float16/bfloat16 for speed. Float32 not supported.

Advanced topics

Integration with HuggingFace Transformers: See references/transformers-integration.md for enabling Flash Attention in BERT, GPT, Llama models.

Performance benchmarks: See references/benchmarks.md for detailed speed and memory comparisons across GPUs and sequence lengths.

Algorithm details: See references/algorithm.md for tiling strategy, recomputation, and IO complexity analysis.

Advanced features: See references/advanced-features.md for rotary embeddings, ALiBi, paged KV cache, and custom attention masks.

Hardware requirements

  • GPU: NVIDIA Ampere+ (A100, A10, A30) or AMD MI200+
  • VRAM: Same as standard attention (Flash Attention doesn't increase memory)
  • CUDA: 12.0+ (11.8 minimum)
  • PyTorch: 2.2+ for native support

Not supported: V100 (Volta), CPU inference

Resources

scroll-experience

davila7

Expert in building immersive scroll-driven experiences - parallax storytelling, scroll animations, interactive narratives, and cinematic web experiences. Like NY Times interactives, Apple product pages, and award-winning web experiences. Makes websites feel like experiences, not just pages. Use when: scroll animation, parallax, scroll storytelling, interactive story, cinematic website.

6230

software-architecture

davila7

Guide for quality focused software architecture. This skill should be used when users want to write code, design architecture, analyze code, in any case that relates to software development.

8125

senior-fullstack

davila7

Comprehensive fullstack development skill for building complete web applications with React, Next.js, Node.js, GraphQL, and PostgreSQL. Includes project scaffolding, code quality analysis, architecture patterns, and complete tech stack guidance. Use when building new projects, analyzing code quality, implementing design patterns, or setting up development workflows.

8122

senior-security

davila7

Comprehensive security engineering skill for application security, penetration testing, security architecture, and compliance auditing. Includes security assessment tools, threat modeling, crypto implementation, and security automation. Use when designing security architecture, conducting penetration tests, implementing cryptography, or performing security audits.

6819

game-development

davila7

Game development orchestrator. Routes to platform-specific skills based on project needs.

5414

2d-games

davila7

2D game development principles. Sprites, tilemaps, physics, camera.

4812

You might also like

flutter-development

aj-geddes

Build beautiful cross-platform mobile apps with Flutter and Dart. Covers widgets, state management with Provider/BLoC, navigation, API integration, and material design.

643969

drawio-diagrams-enhanced

jgtolentino

Create professional draw.io (diagrams.net) diagrams in XML format (.drawio files) with integrated PMP/PMBOK methodologies, extensive visual asset libraries, and industry-standard professional templates. Use this skill when users ask to create flowcharts, swimlane diagrams, cross-functional flowcharts, org charts, network diagrams, UML diagrams, BPMN, project management diagrams (WBS, Gantt, PERT, RACI), risk matrices, stakeholder maps, or any other visual diagram in draw.io format. This skill includes access to custom shape libraries for icons, clipart, and professional symbols.

591705

ui-ux-pro-max

nextlevelbuilder

"UI/UX design intelligence. 50 styles, 21 palettes, 50 font pairings, 20 charts, 8 stacks (React, Next.js, Vue, Svelte, SwiftUI, React Native, Flutter, Tailwind). Actions: plan, build, create, design, implement, review, fix, improve, optimize, enhance, refactor, check UI/UX code. Projects: website, landing page, dashboard, admin panel, e-commerce, SaaS, portfolio, blog, mobile app, .html, .tsx, .vue, .svelte. Elements: button, modal, navbar, sidebar, card, table, form, chart. Styles: glassmorphism, claymorphism, minimalism, brutalism, neumorphism, bento grid, dark mode, responsive, skeuomorphism, flat design. Topics: color palette, accessibility, animation, layout, typography, font pairing, spacing, hover, shadow, gradient."

318398

godot

bfollington

This skill should be used when working on Godot Engine projects. It provides specialized knowledge of Godot's file formats (.gd, .tscn, .tres), architecture patterns (component-based, signal-driven, resource-based), common pitfalls, validation tools, code templates, and CLI workflows. The `godot` command is available for running the game, validating scripts, importing resources, and exporting builds. Use this skill for tasks involving Godot game development, debugging scene/resource files, implementing game systems, or creating new Godot components.

339397

nano-banana-pro

garg-aayush

Generate and edit images using Google's Nano Banana Pro (Gemini 3 Pro Image) API. Use when the user asks to generate, create, edit, modify, change, alter, or update images. Also use when user references an existing image file and asks to modify it in any way (e.g., "modify this image", "change the background", "replace X with Y"). Supports both text-to-image generation and image-to-image editing with configurable resolution (1K default, 2K, or 4K for high resolution). DO NOT read the image file first - use this skill directly with the --input-image parameter.

451339

fastapi-templates

wshobson

Create production-ready FastAPI projects with async patterns, dependency injection, and comprehensive error handling. Use when building new FastAPI applications or setting up backend API projects.

304231

Stay ahead of the MCP ecosystem

Get weekly updates on new skills and servers.