HermesBELARABI
العربيةSearch/CheckingSign in
AcademyDirectorySolutionsStudioProjectsCompareSearch Releases Pricing Account Sign in
Directory → SKILL
SKILLoptionalHermes Optional Skills

Flash Attention

Speed up long-sequence transformer training and inference

OptimizationFlash AttentionAttention OptimizationMemory EfficiencySpeed OptimizationLong ContextPyTorchSDPA
Last registry verification2026-08-18v1.0.1Orchestra Research
Open original source ↗اقرأ بالعربية
Plain meaning

What does it add to Hermes?

Speed up long-sequence transformer training and inference

Flash Attention is a skill related to memory and knowledge. It helps Hermes retrieve past information or stored knowledge instead of starting from zero in every session.

This plain-language explanation is based on the publisher description. The original text remains visible for verification.

Use it when

Use it when your goal in memory and knowledge is clear and you can limit it to the data and actions it actually needs.

Skip it when

Do not add it merely to experiment when Hermes already has a simpler path, or when you cannot review its source and permissions.

Who is it for?

Best for users who want a repeatable way of working inside Hermes.

Safe first test

Add a non-sensitive note, start a new session, and verify retrieval preserves the meaning without inventing details.

Original publisher description

Speed up long-sequence transformer training and inference

✓
Data source

This entry was indexed from Hermes Optional Skills. Our explanation interprets the type and domain without inventing a capability not present upstream.

!
Security review

The source is official or editorially reviewed, but you still need to review permissions and version compatibility.

Safe setup path

Inspect, install, then test.

  1. 01
    Open the source

    Match the publisher, license, and description to your need. Check the real update history.

  2. 02
    Review permissions and secrets

    Never paste a secret value into this site. Use environment-variable names and grant the smallest scope.

  3. 03
    Copy setup only after review

    The controls below copy text. They do not execute commands on your device.

  4. 04
    Test with a non-sensitive task

    Inspect the visible tools, then exclude write or delete tools you do not need.

Install command

Review the command, then copy it.

hermes skills install flash-attention

Hermes Belarabi does not execute this command. Installation happens on your device and remains subject to Hermes scanning and your review.

The full skill definition

Exactly what Hermes loads when this skill runs.

Reproduced from the official documentation. Read it before enabling the skill: this text becomes the agent's instructions.

Speed up long-sequence transformer training and inference.

Skill metadata

A lookup table. Do not read it all; find the row that applies to you.

SourceOptional — install with hermes skills install official/mlops/flash-attention
Pathoptional-skills/mlops/flash-attention
Version1.0.1
AuthorOrchestra Research
LicenseMIT
Dependenciesflash-attn, torch, transformers
Platformslinux, macos
TagsOptimization, Flash Attention, Attention Optimization, Memory Efficiency, Speed Optimization, Long Context, PyTorch, SDPA, H100, FP8, Transformers

Reference: full SKILL.md

Explains the idea itself. Read it slowly; the later sections build on it.

Quick start

Ordered, practical steps. Run one and confirm it worked before moving on.

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+):

Python9 lines



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):

Shell1 line
pip install flash-attn --no-build-isolation
Python4 lines
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

Settings you configure once. Change one at a time so you can see what each does. Set FLASH_ATTENTION in your environment, not in the chat.

Workflow 1: Enable in existing PyTorch model

Copy this checklist:

Text5 lines
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

Shell2 lines
python -c "import torch; print(torch.__version__)"
# Should be ≥2.2.0

If <2.2, upgrade:

Shell1 line
pip install --upgrade torch

Step 2: Enable Flash Attention backend

Replace standard attention:

Python7 lines
# 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)

out = F.scaled_dot_product_attention(q, k, v, attn_mask=mask)

Force Flash Attention backend (torch.backends.cuda.sdp_kernel is deprecated; use torch.nn.attention.sdpa_kernel with SDPBackend):

Python4 lines
from torch.nn.attention import SDPBackend, sdpa_kernel

with sdpa_kernel(SDPBackend.FLASH_ATTENTION):
    out = F.scaled_dot_product_attention(q, k, v)

Step 3: Verify speedup with profiling

Python19 lines


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:
        from torch.nn.attention import SDPBackend, sdpa_kernel
        with sdpa_kernel(SDPBackend.FLASH_ATTENTION):
            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

Python14 lines
# 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:

Text5 lines
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

Shell5 lines
# 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

Python17 lines
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):

Python5 lines
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):

Python6 lines
# 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

Python20 lines

from flash_attn import flash_attn_func


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 Hopper GPUs (H100).

Important: The pip package flash-attn (2.8.x) ships FlashAttention-2 only — it does not contain FA3 or FP8 H100 kernels, and flash_attn_func does not auto-use FP8. FlashAttention-3 is a separate beta build compiled from source from the repo's hopper/ directory, exposed via the flash_attn_interface module. FA3 supports FP16/BF16 forward+backward and FP8 forward only.
Text4 lines
FP8 Setup:
- [ ] Step 1: Verify Hopper (H100) GPU available
- [ ] Step 2: Build & install FlashAttention-3 from source (hopper/)
- [ ] Step 3: Use the FA3 interface (FP8 forward)

Step 1: Verify H100 GPU

Shell2 lines
nvidia-smi --query-gpu=name --format=csv
# Should show "H100" or "H800"

Step 2: Build & install FlashAttention-3 from source

FA3 is NOT included in pip install flash-attn. Build it from the hopper/ subdirectory:

Shell4 lines
git clone https://github.com/Dao-AILab/flash-attention.git
cd flash-attention/hopper
python setup.py install
# (compilation is heavy and requires a CUDA toolchain + Hopper GPU)

Step 3: Use the FA3 interface (FP8 forward)

FA3 exposes its own module flash_attn_interface (distinct from the FA2 flash_attn). FP8 is a forward-only path and expects float8_e4m3fn inputs:

Python15 lines

from flash_attn_interface import flash_attn_func  # FA3 (hopper build), not `flash_attn`

# q, k, v: [batch, seqlen, nheads, headdim]
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)

# FP8 forward (inference / forward-only): cast to float8_e4m3fn
q_fp8 = q.to(torch.float8_e4m3fn)
k_fp8 = k.to(torch.float8_e4m3fn)
v_fp8 = v.to(torch.float8_e4m3fn)

out = flash_attn_func(q_fp8, k_fp8, v_fp8, causal=True)
# FP16/BF16 forward+backward is also supported by the FA3 interface.

When to use vs alternatives

Explains the idea itself. Read it slowly; the later sections build on it.

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 &lt;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

Explains the idea itself. Read it slowly; the later sections build on it.

Issue: ImportError: cannot import flash_attn

Install with no-build-isolation flag:

Shell1 line
pip install flash-attn --no-build-isolation

Or install CUDA toolkit first:

Shell2 lines
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:

  • &lt;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:

Python3 lines

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):

Python1 line
q = q.to(torch.float16)  # Or torch.bfloat16

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

Advanced topics

Explains the idea itself. Read it slowly; the later sections build on it.

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.

Hardware requirements

Explains the idea itself. Read it slowly; the later sections build on it.

  • 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

Explains the idea itself. Read it slowly; the later sections build on it.

  • Paper: "FlashAttention: Fast and Memory-Efficient Exact Attention with IO-Awareness" (NeurIPS 2022)
  • Paper: "FlashAttention-2: Faster Attention with Better Parallelism and Work Partitioning" (ICLR 2024)
  • Blog: https://tridao.me/blog/2024/flash3/
  • GitHub: https://github.com/Dao-AILab/flash-attention
  • PyTorch docs: https://pytorch.org/docs/stable/generated/torch.nn.functional.scaled_dot_product_attention.html