fla-org/flash-linear-attention

💥 Flash Linear Attention

This repo aims at providing a collection of efficient Triton-based implementations for state-of-the-art linear attention models. All implementations are written purely in PyTorch and Triton, making them platform-agnostic. Currently verified platforms include NVIDIA, AMD, and Intel. Any pull requests are welcome!

News
Models
Installation
Usage
Training
Evaluation
Benchmarks
Citation
Star History
Acknowledgments

News

$\texttt{[2025-09]}$: 🐻 Thrilled to announce that GDN has been integrated into Qwen3-Next. Check out the PR and their blog post for more infos!
$\texttt{[2025-08]}$: 🌲 Add Log-Linear Attention implementation to fla (paper).
$\texttt{[2025-08]}$: 🎓 Add MoM implementation to fla (paper).
$\texttt{[2025-07]}$: 🐳 Add MLA implementation to fla (paper).
$\texttt{[2025-07]}$: 🛣️ Added PaTH Attention to fla (paper).
$\texttt{[2025-06]}$: 🎉 Added MesaNet to fla (paper).
$\texttt{[2025-06]}$: 🐍 Add Comba implementation to fla (paper).
$\texttt{[2025-05]}$: 🎉 Add Rodimus* implementation to fla (paper).
$\texttt{[2025-04]}$: 🎉 Add DeltaProduct implementation to fla (paper).
$\texttt{[2025-04]}$: 🎉 Add FoX implementation to fla (paper).
$\texttt{[2025-03]}$: ~~We have changed the default initializer_range to the magic 🐳 0.006~~ The initializer_range was rolled back to the default value of 0.02. For actual training, we recommend trying both.
$\texttt{[2025-02]}$: 🐳 Add NSA implementations to fla. See kernels here.
$\texttt{[2025-01]}$: 🔥 We are migrating to torchtitan-based training framework. Check out the flame repo for more details.
$\texttt{[2025-01]}$: 🦅 Add RWKV7 implementations (both kernels and models) to fla.
$\texttt{[2024-12]}$: Integrated flash-bidirectional-attention to fla-org (repo)
$\texttt{[2024-12]}$: 🎉 Add Gated DeltaNet implementation to fla (paper).
$\texttt{[2024-12]}$: 🚀 fla now officially supports kernels with variable-length inputs.
$\texttt{[2024-11]}$: The inputs are now switched from head-first to seq-first format.
$\texttt{[2024-11]}$: 💥 fla now provides a flexible way for training hybrid models.
$\texttt{[2024-10]}$: 🔥 Announcing flame, a minimal and scalable framework for training fla models. Check out the details here.
$\texttt{[2024-09]}$: fla now includes a fused linear and cross-entropy layer, significantly reducing memory usage during training.
$\texttt{[2024-09]}$: 🎉 Add GSA implementation to fla (paper).
$\texttt{[2024-05]}$: 🎉 Add DeltaNet implementation to fla (paper).
$\texttt{[2024-05]}$: 💥 fla v0.1: a variety of subquadratic kernels/layers/models integrated (RetNet/GLA/Mamba/HGRN/HGRN2/RWKV6, etc., see Models).
$\texttt{[2023-12]}$: 💥 Launched fla, offering a collection of implementations for state-of-the-art linear attention models.

Models

Roughly sorted according to the timeline supported in fla. The recommended training mode is chunk when available.

Year	Venue	Model	Paper	Code
2023		RetNet	Retentive network: a successor to transformer for large language models	official	fla
2024	ICML	GLA	Gated Linear Attention Transformers with Hardware-Efficient Training	official	fla
2024	ICML	Based	Simple linear attention language models balance the recall-throughput tradeoff	official	fla
2024	ACL	Rebased	Linear Transformers with Learnable Kernel Functions are Better In-Context Models	official	fla
2024	NeurIPS	DeltaNet	Parallelizing Linear Transformers with Delta Rule over Sequence Length	official	fla
2022	ACL	ABC	ABC: Attention with Bounded-memory Control		fla
2023	NeurIPS	HGRN	Hierarchically Gated Recurrent Neural Network for Sequence Modeling	official	fla
2024	COLM	HGRN2	HGRN2: Gated Linear RNNs with State Expansion	official	fla
2024	COLM	RWKV6	Eagle and Finch: RWKV with Matrix-Valued States and Dynamic Recurrence	official	fla
2024		LightNet	You Only Scan Once: Efficient Multi-dimension Sequential Modeling with LightNet	official	fla
2025	ICLR	Samba	Samba: Simple Hybrid State Space Models for Efficient Unlimited Context Language Modeling	official	fla
2024	ICML	Mamba2	Transformers are SSMs: Generalized Models and Efficient Algorithms Through Structured State Space Duality	official	fla
2024	NeurIPS	GSA	Gated Slot Attention for Efficient Linear-Time Sequence Modeling	official	fla
2025	ICLR	Gated DeltaNet	Gated Delta Networks: Improving Mamba2 with Delta Rule	official	fla
2025		RWKV7	RWKV-7 “Goose” with Expressive Dynamic State Evolution	official	fla
2025		NSA	Native Sparse Attention: Hardware-Aligned and Natively Trainable Sparse Attention		fla
2025	ICLR	FoX	Forgetting Transformer: Softmax Attention with a Forget Gate	official	fla
2025		DeltaProduct	DeltaProduct: Improving State-Tracking in Linear RNNs via Householder Products		fla
2025	ICLR	Rodimus*	Rodimus*: Breaking the Accuracy-Efficiency Trade-Off with Efficient Attentions	official	fla
2025		MesaNet	MesaNet: Sequence Modeling by Locally Optimal Test-Time Training		fla
2025		Comba	Comba: Improving Bilinear RNNs with Closed-loop Control	official	fla
2025		PaTH	PaTH Attention: Position Encoding via Accumulating Householder Transformations		fla
2025		MoM	MoM: Linear Sequence Modeling with Mixture-of-Memories	official	fla
2025		Log-Linear Attention	Log-Linear Attention	official	fla

Installation

The following requirements should be satisfied

PyTorch >= 2.5
Triton >=3.0 (or nightly version, see FAQs)
einops
transformers >=4.45.0
datasets >=3.3.0

Starting from v0.3.2, the packages published on PyPI are fla-core and flash-linear-attention. The former contains all our customized kernels and only depends on PyTorch, Triton, and einops. The latter is an extension package of the former, containing fla/layers and fla/models, and depends on transformers. We also provide Triton implementations for conv1d operations, so causal-conv1d is not required.

You can install fla with pip:

1

pip install flash-linear-attention

As fla is actively developed now, for the latest features and updates, an alternative way is to install the package from source. Note that installing from git uses the default mode, so you need to uninstall both fla-core and flash-linear-attention first:

1
2


# uninstall both packages first to ensure a successful upgrade
pip uninstall fla-core flash-linear-attention -y && pip install -U git+https://github.com/fla-org/flash-linear-attention

or manage fla with submodules

1
2


git submodule add https://github.com/fla-org/flash-linear-attention.git 3rdparty/flash-linear-attention
ln -s 3rdparty/flash-linear-attention/fla fla

If you have installed triton-nightly and torch pre version, please use the following command:

1
2
3


pip install einops ninja datasets transformers numpy
# uninstall both packages first to ensure a successful upgrade
pip uninstall fla-core flash-linear-attention -y && pip install -U --no-use-pep517 git+https://github.com/fla-org/flash-linear-attention --no-deps

Usage

Token Mixing

We provide ``token mixing’’ linear attention layers in fla.layers for you to use. You can replace the standard multihead attention layer in your model with other linear attention layers. Example usage is as follows:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19


>>> import torch
>>> from fla.layers import MultiScaleRetention
>>> batch_size, num_heads, seq_len, hidden_size = 32, 4, 2048, 1024
>>> device, dtype = 'cuda:0', torch.bfloat16
>>> retnet = MultiScaleRetention(hidden_size=hidden_size, num_heads=num_heads).to(device=device, dtype=dtype)
>>> retnet
MultiScaleRetention(
  (q_proj): Linear(in_features=1024, out_features=1024, bias=False)
  (k_proj): Linear(in_features=1024, out_features=1024, bias=False)
  (v_proj): Linear(in_features=1024, out_features=2048, bias=False)
  (g_proj): Linear(in_features=1024, out_features=2048, bias=False)
  (o_proj): Linear(in_features=2048, out_features=1024, bias=False)
  (g_norm_swish_gate): FusedRMSNormGated(512, eps=1e-05, activation=swish)
  (rotary): RotaryEmbedding(dim=256, base=10000.0, interleaved=False, pos_idx_in_fp32=True)
)
>>> x = torch.randn(batch_size, seq_len, hidden_size).to(device=device, dtype=dtype)
>>> y, *_ = retnet(x)
>>> y.shape
torch.Size([32, 2048, 1024])

We provide the implementations of models that are compatible with 🤗 Transformers library. Here’s an example of how to initialize a GLA model from the default configs in fla:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71


>>> from fla.models import GLAConfig
>>> from transformers import AutoModelForCausalLM
>>> config = GLAConfig()
>>> config
GLAConfig {
  "attn": null,
  "attn_mode": "chunk",
  "bos_token_id": 1,
  "clamp_min": null,
  "conv_size": 4,
  "elementwise_affine": true,
  "eos_token_id": 2,
  "expand_k": 0.5,
  "expand_v": 1,
  "feature_map": null,
  "fuse_cross_entropy": true,
  "fuse_norm": true,
  "fuse_swiglu": true,
  "hidden_act": "swish",
  "hidden_ratio": 4,
  "hidden_size": 2048,
  "initializer_range": 0.006,
  "intermediate_size": null,
  "max_position_embeddings": 2048,
  "model_type": "gla",
  "norm_eps": 1e-06,
  "num_heads": 4,
  "num_hidden_layers": 24,
  "num_kv_heads": null,
  "tie_word_embeddings": false,
  "transformers_version": "4.50.1",
  "use_cache": true,
  "use_gk": true,
  "use_gv": false,
  "use_output_gate": true,
  "use_short_conv": false,
  "vocab_size": 32000
}

>>> AutoModelForCausalLM.from_config(config)
GLAForCausalLM(
  (model): GLAModel(
    (embeddings): Embedding(32000, 2048)
    (layers): ModuleList(
      (0-23): 24 x GLABlock(
        (attn_norm): RMSNorm(2048, eps=1e-06)
        (attn): GatedLinearAttention(
          (q_proj): Linear(in_features=2048, out_features=1024, bias=False)
          (k_proj): Linear(in_features=2048, out_features=1024, bias=False)
          (v_proj): Linear(in_features=2048, out_features=2048, bias=False)
          (g_proj): Linear(in_features=2048, out_features=2048, bias=False)
          (gk_proj): Sequential(
            (0): Linear(in_features=2048, out_features=16, bias=False)
            (1): Linear(in_features=16, out_features=1024, bias=True)
          )
          (o_proj): Linear(in_features=2048, out_features=2048, bias=False)
          (g_norm_swish_gate): FusedRMSNormGated(512, eps=1e-06, activation=swish)
        )
        (mlp_norm): RMSNorm(2048, eps=1e-06)
        (mlp): GatedMLP(
          (gate_proj): Linear(in_features=2048, out_features=5632, bias=False)
          (up_proj): Linear(in_features=2048, out_features=5632, bias=False)
          (down_proj): Linear(in_features=5632, out_features=2048, bias=False)
          (swiglu_linear): SwiGLULinear()
        )
      )
    )
    (norm): RMSNorm(2048, eps=1e-06)
  )
  (lm_head): Linear(in_features=2048, out_features=32000, bias=False)
)

Fused Modules

We offer a collection of fused modules in fla.modules to facilitate faster training:

Rotary Embedding: rotary positional embeddings as adopted by the Llama architecture, a.k.a., Transformer++.
Norm Layers:
- RMSNorm, LayerNorm and GroupNorm
- RMSNormLinear, LayerNormLinear and GroupNormLinear to reduce memory usage of intermediate tensors for improved memory efficiency.
Norm Layers with Gating: combine norm layers with element-wise sigmoid or swish gating, as used by RetNet/GLA.
Cross Entropy: faster Triton implementation of cross entropy loss.
Linear Cross Entropy: fused linear layer and cross entropy loss to avoid the materialization of large logits tensors. Also refer to implementations by mgmalek and Liger-Kernel.
Linear KL Divergence: fused linear layer and KL divergence loss in a similar vein as CE loss.

[!IMPORTANT] You can control using fuse_linear_cross_entropy in the model configuration to enable/disable the fused linear cross entropy loss.

This fused implementation is more memory-efficient but may reduce numerical precision. Due to this trade-off, it is disabled by default. If you enable this feature and encounter training instability (e.g., loss divergence), we recommend disabling it to see if the issue is resolved.

Generation

Upon successfully pretraining a model, it becomes accessible for generating text using the 🤗 text generation APIs. In the following, we give a generation example:

1
2
3
4
5
6
7
8
9


>>> import fla
>>> from transformers import AutoModelForCausalLM, AutoTokenizer
>>> name = 'fla-hub/gla-1.3B-100B'
>>> tokenizer = AutoTokenizer.from_pretrained(name)
>>> model = AutoModelForCausalLM.from_pretrained(name).cuda()
>>> input_prompt = "Power goes with permanence. Impermanence is impotence. And rotation is castration."
>>> input_ids = tokenizer(input_prompt, return_tensors="pt").input_ids.cuda()
>>> outputs = model.generate(input_ids, max_length=64)
>>> tokenizer.batch_decode(outputs, skip_special_tokens=True)[0]

We also provide a simple script here for benchmarking the generation speed. Simply run it by:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13


$ python -m benchmarks.benchmark_generation \
  --path 'fla-hub/gla-1.3B-100B' \
  --repetition_penalty 2. \
  --prompt="Hello everyone, I'm Songlin Yang"

Prompt:
Hello everyone, I'm Songlin Yang
Generated:
Hello everyone, I'm Songlin Yang.
I am a 20 year old girl from China who is currently studying in the United States of America for my Master degree and also working as an English teacher at school here on campus since last summer (1st semester). My main goal to be able do well with this course so that we can have

Prompt length: 10, generation length: 64
Total prompt processing + decoding time: 4593ms

All of the pretrained models currently available can be found in fla-hub.

1
2


>>> from huggingface_hub import list_models
>>> for model in list_models(author='fla-hub'): print(model.id)

Hybrid Models

fla provides a flexible method to incorporate standard attention layers into existing linear attention models. This is easily achieved by specifying the attn argument in the model configuration.

For example, to create a 2-layer Samba model with interleaved Mamba and local attention layers, using a sliding window size of 2048:

  1
  2
  3
  4
  5
  6
  7
  8
  9
 10
 11
 12
 13
 14
 15
 16
 17
 18
 19
 20
 21
 22
 23
 24
 25
 26
 27
 28
 29
 30
 31
 32
 33
 34
 35
 36
 37
 38
 39
 40
 41
 42
 43
 44
 45
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101


>>> from fla.models import SambaConfig
>>> from transformers import AutoModelForCausalLM
>>> config = SambaConfig(num_hidden_layers=2)
>>> config.attn = {
  'layers': [1],
  'num_heads': 18,
  'num_kv_heads': 18,
  'qkv_bias': False,
  'rope_theta': 10000.,
  'window_size': 2048
}
>>> config
SambaConfig {
  "attn": {
    "layers": [
      1
    ],
    "num_heads": 18,
    "num_kv_heads": 18,
    "qkv_bias": false,
    "rope_theta": 10000.0,
    "window_size": 2048
  },
  "bos_token_id": 1,
  "conv_kernel": 4,
  "eos_token_id": 2,
  "expand": 2,
  "fuse_cross_entropy": true,
  "fuse_norm": true,
  "fuse_swiglu": true,
  "hidden_act": "swish",
  "hidden_ratio": 4,
  "hidden_size": 2304,
  "initializer_range": 0.02,
  "intermediate_size": 4608,
  "max_position_embeddings": 2048,
  "model_type": "samba",
  "norm_eps": 1e-05,
  "num_hidden_layers": 2,
  "pad_token_id": 0,
  "rescale_prenorm_residual": false,
  "residual_in_fp32": false,
  "state_size": 16,
  "tie_word_embeddings": false,
  "time_step_floor": 0.0001,
  "time_step_init_scheme": "random",
  "time_step_max": 0.1,
  "time_step_min": 0.001,
  "time_step_rank": 144,
  "time_step_scale": 1.0,
  "transformers_version": "4.50.1",
  "use_bias": false,
  "use_cache": true,
  "use_conv_bias": true,
  "vocab_size": 32000
}

>>> AutoModelForCausalLM.from_config(config)
SambaForCausalLM(
  (backbone): SambaModel(
    (embeddings): Embedding(32000, 2304)
    (layers): ModuleList(
      (0): SambaBlock(
        (mixer_norm): RMSNorm(2304, eps=1e-05)
        (mixer): Mamba(
          (conv1d): Conv1d(4608, 4608, kernel_size=(4,), stride=(1,), padding=(3,), groups=4608)
          (in_proj): Linear(in_features=2304, out_features=9216, bias=False)
          (x_proj): Linear(in_features=4608, out_features=176, bias=False)
          (dt_proj): Linear(in_features=144, out_features=4608, bias=True)
          (out_proj): Linear(in_features=4608, out_features=2304, bias=False)
        )
        (mlp_norm): RMSNorm(2304, eps=1e-05)
        (mlp): GatedMLP(
          (gate_proj): Linear(in_features=2304, out_features=6144, bias=False)
          (up_proj): Linear(in_features=2304, out_features=6144, bias=False)
          (down_proj): Linear(in_features=6144, out_features=2304, bias=False)
          (swiglu_linear): SwiGLULinear()
        )
      )
      (1): SambaBlock(
        (mixer_norm): RMSNorm(2304, eps=1e-05)
        (mixer): Attention(
          (q_proj): Linear(in_features=2304, out_features=2304, bias=False)
          (k_proj): Linear(in_features=2304, out_features=2304, bias=False)
          (v_proj): Linear(in_features=2304, out_features=2304, bias=False)
          (o_proj): Linear(in_features=2304, out_features=2304, bias=False)
          (rotary): RotaryEmbedding(dim=128, base=10000.0, interleaved=False, pos_idx_in_fp32=True)
        )
        (mlp_norm): RMSNorm(2304, eps=1e-05)
        (mlp): GatedMLP(
          (gate_proj): Linear(in_features=2304, out_features=6144, bias=False)
          (up_proj): Linear(in_features=2304, out_features=6144, bias=False)
          (down_proj): Linear(in_features=6144, out_features=2304, bias=False)
          (swiglu_linear): SwiGLULinear()
        )
      )
    )
    (norm_f): RMSNorm(2304, eps=1e-05)
  )
  (lm_head): Linear(in_features=2304, out_features=32000, bias=False)
)

During inference, you DO NOT need to revise anything for generation! The model will produce output as-is, without any need for additional configurations or modifications.

Training

We provide a minimal framework called 🔥 flame built on top of torchtitan, for efficient training of fla models.

Checkout the GLA example for more details.

Evaluation

The lm-evaluation-harness library allows you to easily perform (zero-shot) model evaluations. Follow the steps below to use this library:

Install lm_eval following their instructions.
Run evaluation with:

1
2
3
4
5
6
7
8


$ MODEL='fla-hub/gla-1.3B-100B'
$ python -m evals.harness --model hf \
    --model_args pretrained=$MODEL,dtype=bfloat16 \
    --tasks wikitext,lambada_openai,piqa,hellaswag,winogrande,arc_easy,arc_challenge,boolq,sciq,copa,openbookqa \
    --batch_size 64 \
    --num_fewshot 0 \
    --device cuda \
    --show_config

We’ve made fla compatible with hf-style evaluations, you can call evals.harness to finish the evaluations. Running the command above will provide the task results reported in the GLA paper.

Multi-GPU Evaluation with Hugging Face accelerate 🚀

To perform data-parallel evaluation (where each GPU loads a separate full copy of the model), we leverage the accelerate launcher as follows:

1
2
3
4
5
6
7
8
9


$ MODEL='fla-hub/gla-1.3B-100B'
$ accelerate launch -m evals.harness --model hf  \
    --model_args pretrained=$MODEL,dtype=bfloat16,trust_remote_code=True  \
    --tasks wikitext,lambada_openai,piqa,hellaswag,winogrande,arc_easy,arc_challenge,boolq,sciq,copa,openbookqa \
    --batch_size 64  \
    --num_fewshot 0  \
    --device cuda  \
    --show_config  \
    --trust_remote_code

📏 RULER Benchmark suite

The RULER benchmarks are commonly used for evaluating model performance on long-context tasks. You can evaluate fla models on RULER directly using lm-evaluation-harness. RULER is only available in a relatively recent version of lm-evaluation-harness, so make sure you have the latest version installed.

1
2
3


git clone --depth 1 https://github.com/EleutherAI/lm-evaluation-harness
cd lm-evaluation-harness
pip install -e .

Then, install the necessary dependencies for RULER:

1

pip install lm_eval["ruler"]

and run evaluation by (e.g., 32k contexts):

1
2
3
4
5
6
7
8


$ accelerate launch -m evals.harness \
    --output_path $OUTPUT \
    --tasks niah_single_1,niah_single_2,niah_single_3,niah_multikey_1,niah_multikey_2,niah_multikey_3,niah_multiquery,niah_multivalue,ruler_vt,ruler_cwe,ruler_fwe,ruler_qa_hotpot,ruler_qa_squad \
    --model_args pretrained=$MODEL,dtype=bfloat16,max_length=32768,trust_remote_code=True \
    --metadata='{"max_seq_lengths":[4096,8192,16384,32768]}' \
    --batch_size 2 \
    --show_config  \
    --trust_remote_code

If a GPU can’t load a full copy of the model, please refer to this link for FSDP settings.

[!Tip] If you are using lm-evaluation-harness as an external library and can’t find (almost) any tasks available, before calling lm_eval.evaluate() or lm_eval.simple_evaluate(), simply run the following to load the library’s stock tasks!

1

>>> from lm_eval.tasks import TaskManager; TaskManager().initialize_tasks()

Benchmarks

We compared our Triton-based RetNet implementation with CUDA-based FlashAttention2, using a batch size of 8, 32 heads, and a head dimension of 128, across different sequence lengths. These tests were conducted on a single H100 80GB GPU, as illustrated in the following graph

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12


# you might have to first install `fla` to enable its import via `pip install -e .`
$ python benchmark_retention.py
Performance:
         T  chunk_fwd  parallel_fwd  flash_fwd  chunk_fwdbwd  parallel_fwdbwd  flash_fwdbwd
0    128.0   0.264032      0.243536   0.083488      1.301856         1.166784      0.320704
1    256.0   0.273472      0.252848   0.094304      1.345872         1.300608      0.807936
2    512.0   0.303600      0.278896   0.098112      1.503168         1.433184      0.857216
3   1024.0   0.357248      0.367360   0.156528      1.773552         2.303424      1.160864
4   2048.0   0.454624      0.605616   0.340928      2.283728         4.483360      1.955936
5   4096.0   0.638960      1.378016   1.004992      3.374720        12.271215      4.813776
6   8192.0   1.012352      4.201344   3.625008      5.581808        40.833618     15.023697
7  16384.0   1.748512     14.489664  13.710080     10.191552       153.093765     54.336864

Citation

If you find this repository helpful, please cite our work:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42


@software{yang2024fla,
  title  = {FLA: A Triton-Based Library for Hardware-Efficient Implementations of Linear Attention Mechanism},
  author = {Yang, Songlin and Zhang, Yu},
  url    = {https://github.com/fla-org/flash-linear-attention},
  month  = jan,
  year   = {2024}
}

@inproceedings{yang2024gdn,
  title     = {Gated Delta Networks: Improving Mamba2 with Delta Rule},
  author    = {Songlin Yang and Jan Kautz and Ali Hatamizadeh},
  booktitle = {Proceedings of ICLR},
  year      = {2025}
}

@inproceedings{yang2024deltanet,
  title     = {Parallelizing Linear Transformers with the Delta Rule over Sequence Length},
  author    = {Yang, Songlin and Wang, Bailin and Zhang, Yu and Shen, Yikang and Kim, Yoon},
  booktitle = {Proceedings of NeurIPS},
  year      = {2024}
}

@inproceedings{zhang2024gsa,
  title     = {Gated Slot Attention for Efficient Linear-Time Sequence Modeling},
  author    = {Zhang, Yu and Yang, Songlin and Zhu, Ruijie and Zhang, Yue and Cui, Leyang and Wang, Yiqiao and Wang, Bolun and Shi, Freda and Wang, Bailin and Bi, Wei and Zhou, Peng and Fu, Guohong},
  booktitle = {Proceedings of NeurIPS},
  year      = {2024}
}

@inproceedings{qin2024hgrn2,
  title     = {HGRN2: Gated Linear RNNs with State Expansion},
  author    = {Qin, Zhen and Yang, Songlin and Sun, Weixuan and Shen, Xuyang and Li, Dong and Sun, Weigao and Zhong, Yiran},
  booktitle = {Proceedings of COLM},
  year      = {2024}
}

@inproceedings{yang2024gla,
  title     = {Gated Linear Attention Transformers with Hardware-Efficient Training},
  author    = {Yang, Songlin and Wang, Bailin and Shen, Yikang and Panda, Rameswar and Kim, Yoon},
  booktitle = {Proceedings of ICML},
  year      = {2024}
}

Star History

Acknowledgments

We extend our gratitude to Bitdeer for providing CI server resources that power our infrastructure.