Blame for README.md

---
license: apache-2.0
language:
- en
- zh
library_name: transformers
pipeline_tag: text-generation
tags:
- ksa
- ksa-v2
- custom_code
- long-context
---



<div align="center">
  <h1>KSA V2 4B</h1>
  <p><strong>Summary tokens as long-range memory and sparse indexes</strong></p>
  <p>
    <a href="https://arxiv.org/abs/2610.18654"><img alt="KSA V2 Paper" src="https://img.shields.io/badge/KSA_V2-arXiv%3A2610.18654-b31b1b?logo=arxiv" /></a>
    <a href="https://arxiv.org/abs/2604.24432"><img alt="KSA Paper" src="https://img.shields.io/badge/KSA-arXiv%3A2604.24432-b31b1b?logo=arxiv" /></a>
    <a href="https://github.com/Kuaishou-OneRec/KSA"><img alt="GitHub" src="https://img.shields.io/badge/GitHub-Kuaishou--OneRec-black?logo=github" /></a>
    <a href="LICENSE"><img alt="License" src="https://img.shields.io/badge/License-Apache%202.0-green" /></a>
  </p>
  <p><a href="README.md">English</a> | <a href="README_zh.md">中文</a></p>
</div>

## 📖 Introduction

**Kwai Summary Attention (KSA)** is our attention architecture for efficient
long-context modeling. Learnable Summary tokens at fixed chunk boundaries
preserve a compressed representation of distant history.

**KSA V2** gives these Summaries two complementary roles:

- **Compressed KSA (C-KSA)** aggregates Summary KV into compact Meta KV for
  more efficient long-range memory.
- **Sparse KSA (S-KSA)** uses Summaries to build block-level routing keys.
  A lightweight Indexer selects relevant blocks, whose original KV is read
  directly by the main attention.

We retain Hybrid KSA's 3:1 layer layout, replacing KSA layers with C-KSA and
Full layers with S-KSA.

<p align="center"><img src="assets/figures/ksa_v2_overview.png" width="100%" alt="From Hybrid KSA to KSA V2: the same Summary layout, with C-KSA compression and S-KSA block routing." /></p>
<p align="center"><em>From Hybrid KSA to KSA V2: shared Summaries, layer-local KV compression and sparse original-token access.</em></p>

This base model builds on our **4B Hybrid KSA backbone continually pretrained from Qwen3-4B**. It is not an instruction-tuned model.

## 🔥 News

- **2026-10-13** — [KSA V2 technical report](https://arxiv.org/abs/2610.18654) released.
- **2026-05-08** — [KSA-4B-base](https://huggingface.co/OpenOneRec/KSA-4B-base) weights released.
- **2026-04-28** — [KSA technical report](https://arxiv.org/abs/2604.24432) and training code released.

## ✨ Method

**C-KSA: Summary → Meta KV.** Learnable dimension-wise pooling compresses groups
of Summary KV within each attention layer. Summary tokens continue through
the backbone at their original chunk granularity; stronger KV compression does
not require a new large-chunk backbone. Each query attends to remote Meta,
boundary Summary and local text KV. Meta entries remain local to the layer
that constructs them; they are not propagated as new tokens.

<p align="center"><img src="assets/figures/ksa_v2_compression.png" width="80%" alt="C-KSA compresses complete remote Summary groups into Meta KV while retaining boundary Summaries and local text KV." /></p>
<p align="center"><em>C-KSA hierarchical compression, illustrated with chunk size 4 and two Summaries per Meta entry.</em></p>

**S-KSA: Summary → block index.** A Summary query aggregates each text block
into a representation for low-dimensional routing. Main attention reads the
selected blocks' original KV together with sinks and local context. A
blockwise-max dense-attention teacher supervises the Indexer.

<p align="center"><img src="assets/figures/ksa_v2_sparse.png" width="80%" alt="S-KSA uses Summary-guided block aggregation and an Indexer to select remote text blocks, then attends to their original KV together with sinks and local context." /></p>
<p align="center"><em>S-KSA block routing, illustrated with chunk size 4 and block size 2. Released model configurations are listed below.</em></p>

**Three adaptation stages.** Compression-module warmup → full-parameter C-KSA
CPT → S-KSA aggregation and Indexer warmup. The final stage freezes inherited
parameters; its checkpoint is exported directly for sparse inference.

## 🤖 Model configuration

This directory contains **KSA-v2-4B**.

| Configuration | KSA-v2-1.9B | KSA-v2-4B |
| :--- | :--- | :--- |
| Number of layers | 24 | 36 |
| Hidden size | 2048 | 2560 |
| Intermediate size | 6144 | 9728 |
| Attention heads (Q / KV) | 16 / 16 (MHA) | 32 / 8 (GQA) |
| Head dimension | 128 | 128 |
| Hybrid ratio (C-KSA : S-KSA) | 3 : 1 | 3 : 1 |
| Maximum context (text tokens) | 131072 | 131072 |
| Summary chunk size | 8 | 8 |
| C-KSA sliding chunk number | 128 | 128 |
| C-KSA group size | 128 | 32 |
| S-KSA block size | 4 | 4 |
| S-KSA remote token budget | 4096 | 4096 |
| Sink tokens | 8 | 8 |
| S-KSA local window (text tokens) | 1024 | 1024 |
| Indexer heads × head dimension | 8 × 128 | 8 × 128 |
| Indexer query rank | 512 | 512 |
| Indexer RoPE dimension | 64 | 64 |
| Indexer Q/K RMSNorm | False | True |
| Teacher pooling | max | max |
| Tied embeddings | False | True |

The remote budget includes eight sink tokens; the local window is additional.

Weights are stored in FP32 and loaded in BF16.

## 📈 Results

<p align="center"><img src="assets/figures/ksa_v2_results.png" width="85%" alt="The technical report's first-page 1.9B results: long-context retrieval and general capabilities, and theoretical per-token inference FLOPs and KV cache size." /></p>
<p align="center"><em>1.9B results: long-context and general capabilities (left), theoretical per-token inference FLOPs and KV cache size (right).</em></p>

We evaluate KSA V2 on two Hybrid KSA backbones: a 1.9B model trained from
scratch and a 4B model continually pretrained from Qwen3-4B. Both undergo
three-stage adaptation; original baseline results are reused from KSA V1.

### 1.9B: Long-context retrieval

C-KSA uses group128 and S-KSA uses block4. Hybrid KSA is the parent model before adaptation.

| Model | 4K | 8K | 16K | 32K | 64K | 128K | Mean |
| --- | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
| Full | 76.08 | 72.85 | 73.24 | 69.06 | 65.32 | 48.75 | 67.55 |
| Hybrid SWA | 74.54 | 71.69 | 69.54 | 67.86 | 63.03 | 56.64 | 67.22 |
| Hybrid SCA | 77.72 | 75.22 | 72.55 | 67.74 | 63.54 | 58.01 | 69.13 |
| Hybrid GDN | 79.83 | 76.01 | 74.04 | 70.41 | 69.39 | 59.87 | 71.59 |
| Pure KSA | 70.44 | 65.91 | 66.74 | 62.54 | 57.13 | 39.29 | 60.34 |
| Hybrid KSA | 80.65 | 73.35 | 74.07 | 72.30 | 69.95 | 65.35 | 72.61 |
| C-KSA + Full | 81.52 | 72.24 | 74.44 | 71.93 | 69.61 | 66.31 | 72.68 |
| **C-KSA + S-KSA** | 81.48 | 72.42 | 75.67 | 70.78 | 68.99 | 65.01 | 72.39 |

### 4B: Scaling to a GQA backbone

C-KSA uses group32 and S-KSA uses block4, preserving the backbone's GQA structure.

| Model | 4K | 8K | 16K | 32K | 64K | 128K | Mean |
| --- | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
| Full | 92.88 | 91.38 | 89.12 | 84.74 | 78.16 | 65.86 | 83.69 |
| Hybrid SWA | 91.30 | 88.03 | 82.87 | 78.94 | 73.88 | 66.27 | 80.22 |
| Hybrid SCA | 86.02 | 84.28 | 80.67 | 76.89 | 68.88 | 60.94 | 76.28 |
| Hybrid Linear | 86.39 | 83.86 | 78.06 | 76.48 | 73.50 | 67.98 | 77.71 |
| Pure KSA | 91.55 | 86.78 | 84.78 | 80.30 | 76.09 | 66.81 | 81.05 |
| Hybrid KSA | 92.97 | 90.53 | 88.86 | 86.65 | 76.04 | 71.67 | 84.45 |
| Hybrid KSA + CPT | 92.22 | 89.85 | 87.20 | 83.09 | 74.74 | 72.03 | 83.19 |
| C-KSA + Full | 92.77 | 88.12 | 86.27 | 82.68 | 78.58 | 73.92 | 83.72 |
| **C-KSA + S-KSA** | 93.38 | 87.78 | 85.45 | 83.60 | 80.28 | 74.41 | 84.15 |

Hybrid KSA + CPT is a continuation control with no architectural changes.

See the technical report for further ablations and inference efficiency.

## Quick start

Linux, Python 3.12 and H100/H800. Download the **entire model directory**,
including `kernel/`, and run from it:

```bash
python -m pip install torch==2.6.0 --index-url https://download.pytorch.org/whl/cu126
python -m pip install -r requirements.txt

CUDA_VISIBLE_DEVICES=0 python generate.py \
    --prompt "The main advantage of long-context language models is" \
    --max-new-tokens 128
```

Omit `--prompt` for interactive completion. The optional `--apply-chat-template`
only applies the tokenizer template; it does not add instruction tuning.

Transformers 5.0.0 and the bundled `ksa_attn` / FlashAttention CuTe sources are
used. No training checkout, kernel build or FlashAttention 2 CUDA extension
is required.

### Transformers usage

Set the bundled kernel paths before importing Transformers:

```bash
export PYTHONPATH="$PWD/kernel:$PWD/kernel/deps${PYTHONPATH:+:$PYTHONPATH}"
```

```python
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

tokenizer = AutoTokenizer.from_pretrained(".", local_files_only=True)
model = AutoModelForCausalLM.from_pretrained(
    ".",
    trust_remote_code=True,
    local_files_only=True,
    dtype=torch.bfloat16,
    device_map={"": "cuda:0"},
    attn_implementation="sdpa",
).eval()

inputs = tokenizer("The main advantage of long-context models is", return_tensors="pt")
inputs = inputs.to(model.device)
with torch.inference_mode():
    output = model.generate(
        **inputs,
        max_new_tokens=128,
        do_sample=False,
        pad_token_id=tokenizer.eos_token_id,
    )
print(tokenizer.decode(output[0, inputs.input_ids.shape[1]:], skip_special_tokens=True))
```

## License and citation

Weights and our runtime: [Apache-2.0](LICENSE). Qwen-derived files retain their
copyright notices. Bundled FlashAttention CuTe retains its
[BSD-3-Clause license](kernel/deps/flash_attn/LICENSE) and
[AUTHORS](kernel/deps/flash_attn/AUTHORS); its initializer exposes CuTe without
the separate FlashAttention 2 extension.

```bibtex
@techreport{kwai2026ksav2,
  title       = {Kwai Summary Attention V2 Technical Report},
  author      = {OneRec Team},
  year        = {2026},
  institution = {Kuaishou Technology},
  url         = {https://arxiv.org/abs/2610.18654}
}
```