dsalt 0.2.11

Dynamic Sparse Attention with Landmark Tokens — High-performance Triton implementation

DSALT: Dynamic Sparse Attention with Landmark Tokens

DSALT is a high‑performance PyTorch library that implements Dynamic Sparse Attention with Landmark Tokens – a memory‑efficient attention mechanism for transformers. It relies on Triton kernels and supports distributed training.

Install: pip install dsalt
Source: https://github.com/LeonardoCofone/dsalt-library
Paper: https://zenodo.org/records/19312826
Feature guide: See FEATURE.md here: https://github.com/LeonardoCofone/dsalt-library/blob/main/FEATURE.md

🚀 Key Features

• Memory‑efficient sparse attention – Triton‑accelerated kernels provide 4–8× memory savings compared to dense attention.
• Adaptive local windows – Token‑wise window sizes that grow with sequence position.
• Global landmark tokens – Top‑k informative tokens per head selected via a hybrid energy scoring function.
• Production‑ready training – Mixed‑precision, gradient checkpointing, and validation support.
• Distributed training – Full DDP and FSDP support for multi‑GPU setups.
• Numerical verification – CPU/GPU equivalence tests and gradient stability checks.

---

📋 Table of Contents

1. Installation
2. Quick Start
3. Architecture Overview
4. Training & Generation
5. API Reference
6. Hyperparameter Guide
7. Testing
8. Performance & Benchmarks
9. Citation
10. Contributing
11. License

---

🛠️ Installation

Requirements

• Python 3.8+
• PyTorch 2.0+
• CUDA 11.0+ (GPU) – CPU fallback is available
• Triton 2.0+ (optional, enables GPU kernels)

From PyPI

pip install dsalt

From source

git clone https://github.com/LeonardoCofone/dsalt-library.git
cd dsalt-pytorch
pip install -e .

Development setup

pip install -r requirements-dev.txt

---

🚀 Quick Start

1. Language‑model inference

import torch
from dsalt.model import DSALTLMHeadModel

model = DSALTLMHeadModel(
    vocab_size=32000,
    d_model=1024,
    n_layers=24,
    n_heads=16,
    n_min=32,
    n_max=512,
    k_lmk=64,
)

input_ids = torch.randint(0, 32000, (1, 1024))  # [batch, seq_len]
logits = model(input_ids)                     # [1, 1024, 32000]
print(logits.shape)

# With labels – loss is computed internally
labels = torch.randint(0, 32000, (1, 1024))
outputs = model(input_ids, labels=labels)
loss = outputs.loss
loss.backward()

2. Single‑GPU training

import torch
from torch.utils.data import DataLoader, TensorDataset
from dsalt.model import DSALTLMHeadModel
from dsalt.training import DSALTTrainer

vocab_size = 32000
seq_len = 512
train_dataset = TensorDataset(
    torch.randint(0, vocab_size, (1000, seq_len))
)
train_loader = DataLoader(train_dataset, batch_size=8, shuffle=True)

model = DSALTLMHeadModel(
    vocab_size=vocab_size,
    d_model=768,
    n_layers=12,
    n_heads=12,
    n_min=32,
    n_max=256,
    k_lmk=32,
)

trainer = DSALTTrainer(
    model=model,
    train_loader=train_loader,
    lr=3e-4,
    total_steps=10_000,
    save_dir="checkpoints",
    dtype=torch.bfloat16,
    log_every=50,
)
trainer.train()

3. Multi‑GPU with DataParallel

import torch
import torch.nn as nn
from dsalt.model import DSALTLMHeadModel
from dsalt.training import DSALTTrainer

model = DSALTLMHeadModel(...).to("cuda")
model = nn.DataParallel(model)  # uses all available GPUs

trainer = DSALTTrainer(
    model=model,
    train_loader=train_loader,
    lr=3e-4,
    total_steps=100_000,
    dtype=torch.bfloat16,
    save_dir="checkpoints",
)
trainer.train()

4. Multi‑GPU with FSDP (model sharding)

torchrun --nproc_per_node=2 train.py

Then configure the trainer with fsdp=True.

---

🏗️ Architecture Overview

DSALT combines local causal windows (adaptive per token) with global landmark tokens (top‑k per head):

┌─ Local window (adaptive) ──┬─ Global landmarks ──┐
│ Recent N tokens            │ Top‑K informative │
│ (window size grows)        │ tokens per head   │
└────────────────────────────┴────────────────────┘
                ↓                     ↓
            Sparse attention output

Key components:

1. DSALTAttention – multi‑head sparse attention with adaptive windows and landmark selection.
2. WindowSizePredictor – learns per‑token window sizes.
3. HybridEnergyScorer (kernel) – computes landmark scores.
4. DSALTTransformer – stack of attention + feed‑forward layers.
5. Triton kernels – fused forward and backward passes for speed and memory efficiency.

---

🎯 Training & Generation

See the code snippets above for full training loops. The DSALTTrainer handles:

• Mixed‑precision (BF16 default)
• Gradient checkpointing
• Learning‑rate warm‑up and cosine decay
• Optional window‑entropy regularisation (window_reg_coef)
• Checkpointing and logging utilities

---

📚 API Reference (excerpt)

from dsalt.model import DSALTLMHeadModel
model = DSALTLMHeadModel(vocab_size=32000, d_model=1024, n_layers=24,
                         n_heads=16, n_min=32, n_max=512, k_lmk=64)
logits, windows = model(input_ids, return_window=True)

Low‑level kernel call:

from dsalt.kernels import dsalt_attention
out = dsalt_attention(Q, K, V, window_sizes, landmark_idx)

---

📊 Performance & Benchmarks (May 2026)

Attention type	Approx. memory (GB)	Relative speed
Dense (O(N²))	~3.5	1.0×
FlashAttention 2	~1.8	0.5×
DSALT	~0.6	0.17×

---

📖 Hyperparameter Guide

All hyperparameters are documented in FEATURE.md. Typical configurations are provided for:

• Mobile / Edge – tiny models, low memory.
• Consumer GPU – e.g., RTX 4090, 24 GB.
• Enterprise – H100 80 GB, optional FSDP.
• Research – multi‑node, large models.

---

🧪 Testing

make test-cov          # Full test suite with coverage report
pytest tests/ -v       # Run tests directly

Key test modules:

• tests/test_sparse_attn.py – kernel equivalence and backward.
• tests/test_hybrid_energy.py – landmark scoring.
• tests/test_dsalt_lm.py – language‑model wrapper.
• tests/test_main.py – end‑to‑end smoke test.

---

📄 License

See here: https://github.com/LeonardoCofone/dsalt-library/blob/main/LICENSE

---

🤝 Contributing

Contributions are welcome! Please read CONTRIBUTING.md for guidelines. Areas where help is especially valuable:

• Triton kernel optimisation
• New model architectures (encoder, encoder‑decoder)
• Additional training strategies and samplers
• Documentation and tutorials
• Bug reports and fixes

---

📞 Support & Questions

• Issues: https://github.com/LeonardoCofone/dsalt-library/issues
• Discussions: https://github.com/LeonardoCofone/dsalt-library/discussions
• Paper: https://zenodo.org/records/19312826

---

Last Updated: May 2026