Pith. sign in

REVIEW 5 major objections 5 minor 2 cited by

AutoLoRA: Automatic LoRA Retrieval and Fine-Grained Gated Fusion for Text-to-Image Generation

T0 review · 5 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read AutoLoRA improves text-to-image generation by retrieving relevant LoRA adapters from their weight matrices alone and fusing them with a fine-grained gating mechanism, boosting aesthetic quality and text alignment over the base FLUX model.

desk verdict AutoLoRA has a genuinely new weight-encoding retriever and a sensible fusion mechanism, but the evaluation never measures retrieval accuracy and is partly circular; worth refereeing if the authors can add non-circular benchmarks. read the letter →

arxiv 2508.02107 v1 pith:LQW3B5WH submitted 2025-08-04 cs.CV

Zhiwen Li , Zhongjie Duan , Die Chen , Cen Chen , Daoyuan Chen , Yaliang Li , Yingda Chen This is my paper
classification cs.CV
keywords LoRAtext-to-imagegenerationadapterretrievalfusiongatedcontrastivelearningFLUXdiffusionmodels
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper proposes AutoLoRA, a way to make large text-to-image models better by automatically finding and combining small community-trained adapters called LoRAs. Its retriever reads only the LoRA weight matrices, not any training data, and matches them to a text prompt by encoding both into a shared space through contrastive learning. A learned gating mechanism then fuses the retrieved LoRAs dynamically, with different weights per feature dimension at each layer and diffusion step. The authors show on 162 real FLUX LoRAs that this improves aesthetic scores and text alignment over the base FLUX model, and that fusing more LoRAs keeps improving quality. If correct, the approach would let anyone tap into the growing pool of open-source adapters without documentation or training sets.

What carries the argument

The load-bearing mechanism is the weight-encoding LoRA retriever: each LoRA layer is turned into a token embedding by $v_i = B_i A_i q_i \hat{W}_i$, then a transformer encoder outputs a global LoRA vector that is aligned with CLIP text embeddings via the two-way contrastive loss of Eq. (3). The second mechanism is the fine-grained gated fusion module of Eq. (7), which uses three learned gates—base feature gate, LoRA-specific gate, cross-interaction gate—to produce a sigmoid gating matrix $G \in \mathbb{R}^{k \times d}$, followed by amplitude calibration with a fusion-scaling parameter $w_o$ in Eq. (8). Training uses interference-resistant sampling: a target and an interference LoRA are active but only the target's image-text pairs supervise, via the flow-matching loss of Eq. (10).

What would settle it

Give the retriever a held-out LoRA whose weights have been randomly shuffled, breaking the weight–semantics link while keeping its example renderings and captions unchanged; if the retriever still places it near its original cluster, retrieval is riding on artifacts rather than weight semantics. Likewise, replace the VLM captions in the test set with human-written prompts; if top-1 retrieval accuracy drops to chance, the measured gains are an artifact of the shared caption-generation pipeline.

Watch

Extended reading notes

Core claim

AutoLoRA's central claim is that the semantic content of a LoRA—what style or subject it adds—is readable from its weight matrices alone. The paper builds a LoRA encoder that treats each linear layer's pair of low-rank matrices $(B_i, A_i)$ as a token, maps it to a token embedding via a learned projection, and aggregates tokens with transformer blocks into a single embedding. Trained with a contrastive loss against a frozen CLIP text encoder, this embedding is compared with CLIP text embeddings to retrieve top-k relevant LoRAs. Retrieved LoRAs are then combined by a fine-grained gated fusion module that, at every linear layer and diffusion timestep, computes per-dimension weights from the original hidden state and the LoRA outputs using three gating terms, plus a global LoRA obtained by SVD of the summed weights. On a pool of 162 FLUX.1-dev LoRAs, this pipeline improves MPS, HPS, aesthetic score, and VQAScore over the base model, and the improvement holds on out-of-distribution DiffusionDB prompts.

Load-bearing premise

The whole framework rests on the assumption that a LoRA's semantic content can be learned from its weight matrices using only VLM-generated captions of a few example images as labels; if weights carry too little signal or those captions are not faithful, the retriever cannot generalize and the reported gains would not hold up.

Editorial extensions

If this is right

  • Any newly added community LoRA can be retrieved zero-shot from its weights alone, since no training data or metadata is needed.
  • Fusing top-3 retrieved LoRAs improves MPS, HPS, aesthetic score, and VQAScore over base FLUX on both synthetic and DiffusionDB prompts.
  • The gated fusion mechanism remains stable when fusing two or three LoRAs, where direct linear addition and DARE degrade or fail completely.
  • The global LoRA, built by SVD of the summed weights, reduces interference and improves multi-LoRA fusion performance.
  • The framework scales to arbitrarily many retrieved LoRAs, avoiding the fixed-expert limitation of MoE-based fusion methods.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • If weight-based semantic encoding transfers across architectures, the same retriever could index other adapter formats or base diffusion models, such as Stable Diffusion 3, without retraining on their LoRA pools.
  • Because the same VL model generates both the training captions and the synthetic test prompts, reported retrieval gains could be partly inflated by caption-style overlap; a strict test would replace test captions with human-written prompts.
  • The per-dimension gating weights might serve as an interpretability signal, revealing which LoRA contributes which visual attribute to the final image, and could enable user-controlled mixing.
  • The pool can grow incrementally at registration time: a new LoRA only needs its weights encoded once, after which retrieval and fusion proceed without retraining the framework.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

5 major / 5 minor

Summary. The paper proposes AutoLoRA, a framework for retrieving task-relevant LoRA adapters from a growing pool and fusing them into a base text-to-image diffusion model. The retriever maps LoRA weight matrices into a shared embedding space with a CLIP-style text encoder via contrastive learning, using VLM-generated captions of example renderings as supervision. The fusion module computes per-dimension gating weights for the retrieved LoRAs at each layer, augmented by a synthetic 'global LoRA' obtained by SVD-decomposing the sum of the selected LoRAs' delta matrices. Experiments on a pool of 162 FLUX.1-dev LoRAs report improvements in MPS, HPS, Aesthetic, and VQA scores over the base model on a synthetic prompt set and on DiffusionDB prompts, together with comparisons against Direct, K-LoRA, and DARE fusion baselines.

Significance. The central idea is timely and potentially useful: encoding LoRA weights directly into a semantic space would enable zero-shot retrieval without access to training data, addressing a real bottleneck in the open-source adapter ecosystem. The interference-resistant training objective and the layer/dimension-wise gating design are reasonable building blocks. However, the paper's strongest claimed contribution is retrieval-driven improvement, and that claim is not directly supported by the current experiments. The evaluation is end-to-end only, with no retrieval accuracy measurement, no random-LoRA control, small absolute gains without significance testing, and a synthetic test set that shares the caption-generation pipeline used for retriever training. If the authors add a rigorous retrieval benchmark and isolate the contribution of retrieval from fusion, the framework could be a solid systems contribution; as it stands, the evidence is suggestive but not conclusive.

major comments (5)
  1. [Effectiveness of LoRA Encoder; Experimental Setup; Tables 1, 3, 4] Retrieval quality is never measured directly. There is no precision@k, recall@k, or any ground-truth relevance judgment for the 162-LoRA pool, and no baseline such as CLIP text-only matching, random selection, or a popularity heuristic. Because the fusion module and the global LoRA of Eq. (9) can alter outputs even when the retrieved LoRAs are irrelevant, the end-to-end gains in Tables 1, 3, and 4 do not establish that retrieval contributed. A control where the same gated fusion module receives k randomly selected or least-similar LoRAs is necessary to support the abstract's claim that AutoLoRA improves generation 'by retrieving relevant LoRA models.'
  2. [Experimental Setup (synthetic prompt set), Weight Encoding-base LoRA Retriever] The synthetic prompt set appears to be circular with retriever training. The retriever is trained on Qwen VL captions of 1-5 example renderings per LoRA, and the synthetic prompts are also generated by Qwen VL from those same renderings. On this test set, top-k retrieval can succeed by recognizing which LoRA's caption produced the prompt, rather than by generalizable semantic retrieval from LoRA weights. The authors need a held-out benchmark with independent labels, for example human-written prompts, captions from a different VLM or pipeline, and LoRAs that were not seen during retriever training, to support the zero-shot generalization claim.
  3. [Tables 1, 3, 4, 5; Fine-grained Dynamic Gated LoRA Fusion] None of the quantitative results include variance, confidence intervals, or significance tests, and the reported gains are small in absolute terms (e.g., Table 1: HPS 0.324 to 0.340; VQA 0.916 to 0.922 on the synthetic set). Without multiple seeds, the number of evaluation prompts, and per-prompt variability, it is impossible to tell whether these differences are meaningful. The DiffusionDB results are also averaged over 1,000 rewritten prompts, but no breakdown by prompt type is given. Please report standard errors or confidence intervals and, ideally, paired significance tests against the base model and the random-LoRA control.
  4. [Eq. (7), Fine-grained Dynamic Gated LoRA Fusion] Equation (7) is dimensionally inconsistent as written. The inputs are x ∈ R^{l×d} and L ∈ R^{k×l×d}, but the additive terms have shapes R^{l×d}, R^{k×l×d}, and R^{k×l×d}, and the resulting G is claimed to be R^{k×d}. The mechanism for reducing the per-token (or per-position) computation to a single k×d gate is not specified. This is central to reproducing the fusion module; please define the exact tensor operations, including any pooling or broadcasting over the sequence dimension l.
  5. [Effectiveness of LoRA Encoder (Figure 4)] The only direct evidence for the LoRA encoder's retrieval ability is a pairwise similarity heatmap over 21 hand-selected LoRAs grouped into six themes. This is a qualitative sanity check, not a retrieval evaluation. It does not quantify retrieval accuracy on the 162-LoRA pool, does not include near-duplicate or confusable themes, and cannot support the claim of 'highly precise LoRA retrieval in practical applications.' A quantitative retrieval metric with a meaningful relevance definition is required.
minor comments (5)
  1. [Throughout] There are numerous typos and stylistic issues: 'Itroduction' in the section heading, 'perfermance' in the abstract, 'strategies for multi-LoRA fusion strategies' (redundant), and inconsistent capitalization of 'LoRA' and 'loRA'. Please copyedit carefully.
  2. [Eq. (1)] The notation in Eq. (1) is under-specified: q_i is called a trainable parameter but is used as a vector in R^k, and the role of the matrix product B_i A_i q_i is not explained. Please state explicitly that q_i is a learned column-selection vector and specify all dimensions.
  3. [Eq. (10)] In the flow-matching loss, x_1 is never defined, and 'xt' is described as 'the targer image in laten space.' Please define x_0, x_1, and x_t consistently and correct the typos.
  4. [Related Works] The related-work section refers to 'MoLE' twice with different citations: once as (Wu, Huang, and Wei 2024) in Adapter Retrieval and once as (Zhao et al. 2024) in Adapter Fusion. Please disambiguate these methods and align the citations with the correct references.
  5. [Experimental Setup] The paper does not state how many evaluation prompts were used for Tables 3, 4, and 5, how many images per prompt were generated, or whether the same random seeds were used across methods. Please include these details for reproducibility.

Circularity Check

1 steps flagged · score 6.0 of 10

Synthetic retrieval set is built from the same Qwen VL captions and renderings used to train the retriever, so the synthetic-prompt gains in Table 1 reduce to a fit to the training labels; the DiffusionDB and fusion results are not circular but lack a random-LoRA control.

  1. fitted input called prediction [Training Object (Eq. 3) and Experimental Setup (Synthetic retrieval test set)]
    "First, use VLM to convert each LoRA rendering into text, and use these texts as the labels corresponding to each LoRA. ... We then utilized Qwen VL to generate descriptive captions for each image. For the LoRA retrieval experiments, we randomly selected 1–3 images from the dataset and employed Qwen VL to create prompts that simultaneously capture features from multiple images, ultimately constructing a synthetic prompt set comprising 900 prompts in total."

    The test prompts are generated by the same Qwen VL captioner and from the same 1-5 example renderings that supplied the training labels. A synthetic prompt is therefore a paraphrase or merge of the exact caption used as the positive text for that LoRA in the contrastive loss of Eq. (3). Top-k retrieval on this set reduces to retrieving the LoRA whose embedding is nearest to its own Qwen caption in the trained space, i.e., measuring training-set fit rather than zero-shot retrieval. The paper then cites these synthetic-prompt results to conclude that suitable LoRAs were successfully retrieved, making the synthetic half of Table 1 a self-consistent fit rather than an independent prediction.

full rationale

The retriever's only labeled supervision is Qwen VL's text for 1-5 example renderings per LoRA, used in the contrastive objective of Eq. (3). The synthetic retrieval test set is then produced by the same Qwen VL model from the same renderings. A prompt built from a LoRA's example image is effectively a perturbation of the caption used to train that LoRA's embedding, so top-k retrieval on this set is a fitted-label match rather than a zero-shot generalization test. The paper nonetheless uses these synthetic-prompt results to conclude that suitable LoRAs have been successfully retrieved, which is the circular component of the evaluation. The DiffusionDB portion is not circular because it uses an external prompt source, but it is not a retrieval-accuracy test: there is no ground-truth association between prompts and LoRAs, and Tables 1, 3, and 4 never compare the same fusion module with randomly selected or least-similar LoRAs, so aggregate aesthetic and text-alignment gains do not by themselves prove that the retriever selected the right adapters. The fusion module itself is evaluated independently in Tables 2-4 and is not circular. Overall, the circularity is partial and concentrated in the synthetic retrieval benchmark; the central retrieval claim therefore rests in part on a construction that reduces to the training labels.

Assumptions & free parameters 5 free parameters · 5 assumptions · 1 invented entities

The central system has two learned stages. The retriever rests on a strong assumption that raw LoRA weights encode semantics; the fusion rests on gates trained only on pairs generalizing to arbitrary sets. Both are plausible but unproven, and the evaluation shares data-generation machinery with training.

free parameters (5)
  • LoRA encoder projection matrix W_hat_i and transformer parameters
    Fitted by the contrastive loss in Eq. (3) to map LoRA layers into a shared embedding space; these parameters are the core of the retriever.
  • Gating weights wx, wl, wc, bias b, and fusion scaling wo
    Fitted by the interference-resistant flow matching loss in Eq. (10); these determine how retrieved LoRAs are combined.
  • Global LoRA decomposition rank = 4 (typical)
    Chosen by hand in Eq. (9) for computational efficiency; the paper does not study sensitivity to this rank.
  • Number of retrieved LoRAs k = 1, 2, 3
    Evaluation uses top-1, top-2, and top-3; the framework itself does not specify how to select k for a given prompt.
  • LoRA encoder transformer depth n
    The paper shows n transformer blocks in Fig. 1 but never states the value or whether it was tuned.
assumptions (5)
  • domain assumption LoRA weight matrices contain semantic information that can be projected and read by a transformer encoder.
    Eqs. (1)-(3) assume the weight-based embedding is sufficient for text-to-LoRA matching; no theoretical or empirical proof is given beyond the in-pool heatmap in Fig. 4.
  • domain assumption Qwen VL captions of 1-5 renderings per LoRA are accurate semantic labels for the LoRA.
    Training and synthetic evaluation both rely on these captions; incorrect or unrepresentative captions would break retrieval and fusion training.
  • ad hoc to paper A sum of LoRA delta matrices, decomposed to rank 4 by SVD, captures cross-contextual information useful for fusion.
    Eq. (9) introduces the global LoRA; the only support is the ablation in Table 5, which is conducted on the same pool.
  • domain assumption Flow matching loss with one target and one random interference LoRA generalizes to arbitrary k-way combinations.
    Training uses pairs in Eq. (10), but evaluation fuses 2-3 LoRAs; no proof is provided that pair-trained gates scale to more adapters.
  • domain assumption Pre-trained CLIP text encoder provides a valid semantic space for matching text prompts to LoRA weights.
    The LoRA encoder is trained against frozen CLIP text embeddings, so any CLIP limitation transfers directly to retrieval.
invented entities (1)
  • Global general LoRA L_g
    purpose: A synthetic LoRA built by SVD of the sum of retrieved LoRA matrices, intended to capture shared context and reduce fusion conflict.
    Only supported by the in-paper ablation in Table 5 on the same LoRA pool; there is no external validation of this construct.

how reviews work

0 comments
Cite this review

Pith. "Pith review of AutoLoRA: Automatic LoRA Retrieval and Fine-Grained Gated Fusion for Text-to-Image Generation." pith.science (2026). https://pith.science/paper/LQW3B5WH

@misc{pith2026250802107,
  author       = {Pith},
  title        = {Pith review of: AutoLoRA: Automatic LoRA Retrieval and Fine-Grained Gated Fusion for Text-to-Image Generation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/LQW3B5WH}},
  note         = {Machine review of arXiv:2508.02107}
}
read the original abstract

Despite recent advances in photorealistic image generation through large-scale models like FLUX and Stable Diffusion v3, the practical deployment of these architectures remains constrained by their inherent intractability to parameter fine-tuning. While low-rank adaptation (LoRA) have demonstrated efficacy in enabling model customization with minimal parameter overhead, the effective utilization of distributed open-source LoRA modules faces three critical challenges: sparse metadata annotation, the requirement for zero-shot adaptation capabilities, and suboptimal fusion strategies for multi-LoRA fusion strategies. To address these limitations, we introduce a novel framework that enables semantic-driven LoRA retrieval and dynamic aggregation through two key components: (1) weight encoding-base LoRA retriever that establishes a shared semantic space between LoRA parameter matrices and text prompts, eliminating dependence on original training data, and (2) fine-grained gated fusion mechanism that computes context-specific fusion weights across network layers and diffusion timesteps to optimally integrate multiple LoRA modules during generation. Our approach achieves significant improvement in image generation perfermance, thereby facilitating scalable and data-efficient enhancement of foundational models. This work establishes a critical bridge between the fragmented landscape of community-developed LoRAs and practical deployment requirements, enabling collaborative model evolution through standardized adapter integration.

Figures

Figures reproduced from arXiv: 2508.02107 by the authors.

Figure 1
Figure 1. Illustration of the AutoLoRA Framework. It comprises two key components: a) [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Qualitative comparison between AutoLoRA and original FLUX.1-dev. Top-1, Top-2, and Top-3 represent images [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Qualitative comparison of object-style LoRA fusion. Compared with other baseline methods, our fine-grained gating [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: LoRA embedding similarity heatmap. LoRAs [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]

Discussion (0). Sign in to comment.

Forward citations

Cited by 2 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Parametric Memory Decoding for Zero-Shot Routing in LoRA-Based External Parametric Memory

    cs.LG 2026-07 conditional novelty 6.0 of 10

    PMDRouter selects LoRAs zero-shot by decoding scale-normalized linear response energy from one adapter-free backbone prefill, and leads most internal-signal baselines on a new multi-granularity EPM bench.

  2. Training-Free Multi-Concept LoRA Composition with Prompt-Aware Weighting

    cs.CV 2026-06 unverdicted novelty 6.0 of 10

    Prompt-aware weighting strategies W-Switch and W-Composite improve multi-concept LoRA composition in diffusion models without training.

Reference graph

Works this paper leans on

29 extracted references · 22 canonical work pages · cited by 2 Pith papers

  1. [1]

    , " * write output.state after.block = add.period write newline

    ENTRY address archivePrefix author booktitle chapter edition editor eid eprint howpublished institution isbn journal key month note number organization pages publisher school series title type volume year label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block FUNCTION init.state.consts #0 'before.a...

  2. [2]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in capitalize " " * FUNCT...

  3. [3]

    X.; Gao, H.; Chen, D.; Li, J.; Zeng, W.; Yu, X.; Wu, Y.; Xie, Z.; Li, Y

    Dai, D.; Deng, C.; Zhao, C.; Xu, R. X.; Gao, H.; Chen, D.; Li, J.; Zeng, W.; Yu, X.; Wu, Y.; Xie, Z.; Li, Y. K.; Huang, P.; Luo, F.; Ruan, C.; Sui, Z.; and Liang, W. 2024. DeepSeekMoE: Towards Ultimate Expert Specialization in Mixture-of-Experts Language Models. In Ku, L.; Martins, A.; and Srikumar, V., eds., Proceedings of the 62nd Annual Meeting of the ...

    work page 2024

  4. [4]

    H.; and Khan, F

    Dong, J.; Liang, W.; Li, H.; Zhang, D.; Cao, M.; Ding, H.; Khan, S. H.; and Khan, F. S. 2024. How to Continually Adapt Text-to-Image Diffusion Models for Flexible Customization? In Globersons, A.; Mackey, L.; Belgrave, D.; Fan, A.; Paquet, U.; Tomczak, J. M.; and Zhang, C., eds., Advances in Neural Information Processing Systems 38: Annual Conference on N...

    work page 2024

  5. [5]

    Esser, P.; Kulal, S.; Blattmann, A.; Entezari, R.; M \" u ller, J.; Saini, H.; Levi, Y.; Lorenz, D.; Sauer, A.; Boesel, F.; Podell, D.; Dockhorn, T.; English, Z.; and Rombach, R. 2024. Scaling Rectified Flow Transformers for High-Resolution Image Synthesis. In Forty-first International Conference on Machine Learning, ICML 2024, Vienna, Austria, July 21-27...

    work page 2024

  6. [6]

    Fedus, W.; Zoph, B.; and Shazeer, N. 2022. Switch Transformers: Scaling to Trillion Parameter Models with Simple and Efficient Sparsity. J. Mach. Learn. Res., 23: 120:1--120:39

    work page 2022

  7. [7]

    Gao, Y.; Xiong, Y.; Gao, X.; Jia, K.; Pan, J.; Bi, Y.; Dai, Y.; Sun, J.; Guo, Q.; Wang, M.; and Wang, H. 2023. Retrieval-Augmented Generation for Large Language Models: A Survey. CoRR, abs/2312.10997

    arXiv 2023

  8. [8]

    Z.; Shi, Y.; Chen, Y.; Fan, Z.; Xiao, W.; Zhao, R.; Chang, S.; Wu, W.; Ge, Y.; Shan, Y.; and Shou, M

    Gu, Y.; Wang, X.; Wu, J. Z.; Shi, Y.; Chen, Y.; Fan, Z.; Xiao, W.; Zhao, R.; Chang, S.; Wu, W.; Ge, Y.; Shan, Y.; and Shou, M. Z. 2023. Mix-of-Show: Decentralized Low-Rank Adaptation for Multi-Concept Customization of Diffusion Models. In Oh, A.; Naumann, T.; Globerson, A.; Saenko, K.; Hardt, M.; and Levine, S., eds., Advances in Neural Information Proces...

    work page 2023

Show all 29 references
  1. [9]

    J.; Shen, Y.; Wallis, P.; Allen - Zhu, Z.; Li, Y.; Wang, S.; Wang, L.; and Chen, W

    Hu, E. J.; Shen, Y.; Wallis, P.; Allen - Zhu, Z.; Li, Y.; Wang, S.; Wang, L.; and Chen, W. 2022. LoRA: Low-Rank Adaptation of Large Language Models. In The Tenth International Conference on Learning Representations, ICLR 2022, Virtual Event, April 25-29, 2022 . OpenReview.net

    2022

  2. [10]

    Labs, B. F. 2024. FLUX

    2024

  3. [11]

    Lepikhin, D.; Lee, H.; Xu, Y.; Chen, D.; Firat, O.; Huang, Y.; Krikun, M.; Shazeer, N.; and Chen, Z. 2021. GShard: Scaling Giant Models with Conditional Computation and Automatic Sharding. In 9th International Conference on Learning Representations, ICLR 2021, Virtual Event, A...

    2021

  4. [12]

    u ttler, H.; Lewis, M.; Yih, W.; Rockt \

    Lewis, P.; Perez, E.; Piktus, A.; Petroni, F.; Karpukhin, V.; Goyal, N.; K \" u ttler, H.; Lewis, M.; Yih, W.; Rockt \" a schel, T.; Riedel, S.; and Kiela, D. 2020. Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks. In Larochelle, H.; Ranzato, M.; Hadsell, R.; B...

    2020

  5. [13]

    Lin, Z.; Pathak, D.; Li, B.; Li, J.; Xia, X.; Neubig, G.; Zhang, P.; and Ramanan, D. 2024. Evaluating Text-to-Visual Generation with Image-to-Text Generation. In Leonardis, A.; Ricci, E.; Roth, S.; Russakovsky, O.; Sattler, T.; and Varol, G., eds., Computer Vision - ECCV 2024 ...

    2024

  6. [14]

    Muqeeth, M.; Liu, H.; Liu, Y.; and Raffel, C. 2024. Learning to Route Among Specialized Experts for Zero-Shot Generalization. In Forty-first International Conference on Machine Learning, ICML 2024, Vienna, Austria, July 21-27, 2024 . OpenReview.net

    2024

  7. [15]

    Muqeeth, M.; Liu, H.; and Raffel, C. 2024. Soft Merging of Experts with Adaptive Routing. Trans. Mach. Learn. Res., 2024

    2024

  8. [16]

    Ouyang, Z.; Li, Z.; and Hou, Q. 2025. K-LoRA: Unlocking Training-Free Fusion of Any Subject and Style LoRAs. In IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2025, Nashville, TN, USA, June 11-15, 2025 , 13041--13050. Computer Vision Foundation / IEEE

    2025

  9. [17]

    B.; H \" a renstam - Nielsen, L.; Seg \` u , M.; Dovesi, P

    Qorbani, R.; Villani, G.; Panagiotakopoulos, T.; Colomer, M. B.; H \" a renstam - Nielsen, L.; Seg \` u , M.; Dovesi, P. L.; Karlgren, J.; Cremers, D.; Tombari, F.; and Poggi, M. 2025. Semantic Library Adaptation: LoRA Retrieval and Fusion for Open-Vocabulary Semantic Segmenta...

    2025

  10. [18]

    W.; Hallacy, C.; Ramesh, A.; Goh, G.; Agarwal, S.; Sastry, G.; Askell, A.; Mishkin, P.; Clark, J.; Krueger, G.; and Sutskever, I

    Radford, A.; Kim, J. W.; Hallacy, C.; Ramesh, A.; Goh, G.; Agarwal, S.; Sastry, G.; Askell, A.; Mishkin, P.; Clark, J.; Krueger, G.; and Sutskever, I. 2021. Learning Transferable Visual Models From Natural Language Supervision. In Meila, M.; and Zhang, T., eds., Proceedings of...

    2021

  11. [19]

    Rombach, R.; Blattmann, A.; Lorenz, D.; Esser, P.; and Ommer, B. 2022. High-Resolution Image Synthesis with Latent Diffusion Models. In IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2022, New Orleans, LA, USA, June 18-24, 2022 , 10674--10685. IEEE

    2022

  12. [20]

    Schuhmann, C.; Beaumont, R.; Vencu, R.; Gordon, C.; Wightman, R.; Cherti, M.; Coombes, T.; Katta, A.; Mullis, C.; Wortsman, M.; Schramowski, P.; Kundurthy, S.; Crowson, K.; Schmidt, L.; Kaczmarczyk, R.; and Jitsev, J. 2022. LAION-5B: An open large-scale dataset for training ne...

    2022

  13. [21]

    Shah, V.; Ruiz, N.; Cole, F.; Lu, E.; Lazebnik, S.; Li, Y.; and Jampani, V. 2024. ZipLoRA: Any Subject in Any Style by Effectively Merging LoRAs. In Leonardis, A.; Ricci, E.; Roth, S.; Russakovsky, O.; Sattler, T.; and Varol, G., eds., Computer Vision - ECCV 2024 - 18th Europe...

    2024

  14. [22]

    Simsar, E.; Hofmann, T.; Tombari, F.; and Yanardag, P. 2025. LoRACLR: Contrastive Adaptation for Customization of Diffusion Models. In IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2025, Nashville, TN, USA, June 11-15, 2025 , 13189--13198. Computer Visio...

    2025

  15. [23]

    J.; Montoya, E.; Munechika, D.; Yang, H.; Hoover, B.; and Chau, D

    Wang, Z. J.; Montoya, E.; Munechika, D.; Yang, H.; Hoover, B.; and Chau, D. H. 2023. DiffusionDB: A Large-scale Prompt Gallery Dataset for Text-to-Image Generative Models. In Rogers, A.; Boyd - Graber, J. L.; and Okazaki, N., eds., Proceedings of the 61st Annual Meeting of the...

    2023

  16. [24]

    Wu, X.; Hao, Y.; Sun, K.; Chen, Y.; Zhu, F.; Zhao, R.; and Li, H. 2023. Human Preference Score v2: A Solid Benchmark for Evaluating Human Preferences of Text-to-Image Synthesis. CoRR, abs/2306.09341

    2023 arXiv

  17. [25]

    Wu, X.; Huang, S.; and Wei, F. 2024. Mixture of LoRA Experts. In The Twelfth International Conference on Learning Representations, ICLR 2024, Vienna, Austria, May 7-11, 2024 . OpenReview.net

    2024

  18. [26]

    Yang, Y.; Wang, W.; Peng, L.; Song, C.; Chen, Y.; Li, H.; Yang, X.; Lu, Q.; Cai, D.; Wu, B.; and Liu, W. 2024. LoRA-Composer: Leveraging Low-Rank Adaptation for Multi-Concept Customization in Training-Free Diffusion Models. CoRR, abs/2403.11627

    2024 arXiv

  19. [27]

    Yu, L.; Yu, B.; Yu, H.; Huang, F.; and Li, Y. 2024. Language Models are Super Mario: Absorbing Abilities from Homologous Models as a Free Lunch. In Forty-first International Conference on Machine Learning, ICML 2024, Vienna, Austria, July 21-27, 2024 . OpenReview.net

    2024

  20. [28]

    Zhang, S.; Wang, B.; Wu, J.; Li, Y.; Gao, T.; Zhang, D.; and Wang, Z. 2024. Learning Multi-Dimensional Human Preference for Text-to-Image Generation. In IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2024, Seattle, WA, USA, June 16-22, 2024 , 8018--8027. IEEE

    2024

  21. [29]

    Zhao, Z.; Gan, L.; Wang, G.; Hu, Y.; Shen, T.; Yang, H.; Kuang, K.; and Wu, F. 2024. Retrieval-Augmented Mixture of LoRA Experts for Uploadable Machine Learning. CoRR, abs/2406.16989

    2024 arXiv

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.