Pith. sign in

REVIEW 3 major objections 4 minor 1 cited by

Memorization and Knowledge Injection in Gated LLMs

T0 review · 3 major / 4 minor · reviewed 2026-08-16 · deepseek-v4-flash

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

Pith's one-line read MEGa stores each new memory in dedicated gated low-rank weights, letting an LLM recall and answer questions about sequentially injected events while largely avoiding catastrophic forgetting.

desk verdict A solid empirical paper on gated per-memory LoRA adapters; the residual forgetting is plausibly gating-limited, but the authors need to separate argmax errors from softmax dilution with an oracle-gating control. read the letter →

arxiv 2504.21239 v1 pith:DONZPDT3 submitted 2025-04-30 cs.CL cs.AI

Xu Pan , Ely Hahami , Zechen Zhang , Haim Sompolinsky This is my paper · ORCID
classification cs.CLcs.AI
keywords continuallearningcatastrophicforgettingknowledgeinjectiongatedlow-rankadaptationepisodicmemorylargelanguagemodelsretrieval-augmentedgeneration
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 sets out to establish that a large language model can absorb new event memories one at a time, keep them available for later recall, and answer questions about them, all without overwriting its prior knowledge, provided each memory is stored in its own dedicated gated low-rank weight set. The proposed framework, MEGa, trains one low-rank adapter (LoRA) per incoming paragraph and stores the paragraph's context key as the average of the base model's internal activations near the final MLP layer. At inference, a softmax gate compares the query's embedding with all stored keys and activates only the relevant adapter, so later memories do not erase earlier ones. On two datasets—fictional character stories and 2024 Wikipedia events—MEGa recalls stories at 0.90–0.92 cosine similarity, answers questions at 72.5% and 78.0% accuracy, and keeps MMLU within about one point of the untrained base model, while sequential fine-tuning baselines collapse. This matters because it offers a weight-based route to continual knowledge injection that does not require an external database, and the paper frames it as a step toward biologically plausible episodic memory in LLMs.

What carries the argument

The load-bearing mechanism is a per-memory gated LoRA module. A rank-128 LoRA adapter is trained for each sample on the MLP layers, and each sample is associated with a context key $K_i=f(D_i)$, the average of the model's internal activations at the input to the final MLP layer. During inference the model applies global gating weights $g_i=\mathrm{softmax}(\beta f(q)^\top K)_i$ to the sum of adapters, so the effective weight matrix in each layer is the pretrained weight plus a weighted combination of memory adapters. The same embedding function $f$ is used both to store memory keys and to embed user queries, which keeps key generation and retrieval inside one model and makes the gate the sole routing decision. Because the gate is computed once per query and shared across all layers, it selects one dominant memory per question while still allowing a soft mixture of two memories for compositional questions.

What would settle it

To test whether gating or adapter interference is the real limit, construct 50 memories with the same protagonist, location, and date that differ only in the event described, train MEGa on all 50, and ask a question about one specific event. If correct-gate rate collapses toward chance while recall with the correct gate forced stays near the paper's reported ~0.9 cosine similarity, the embedding is the bottleneck and the central claim stands as stated; if recall degrades even with the correct gate forced, the adapters themselves interfere and the no-catastrophic-forgetting claim fails.

Watch

Extended reading notes

Core claim

MEGa's discovery is that catastrophic forgetting in LLM knowledge injection can be avoided by dedicating a separate low-rank weight adapter to each memory and routing each query to the right adapter through a semantic similarity computed inside the model itself. Each memory's key is $K_i = f(D_i)$, where $f$ averages the base model's internal activations at the input to the final MLP layer; each adapter $\{A_i^\ell, B_i^\ell\}$ is fine-tuned on that memory's paragraph with an instruction-style prompt, and inference uses the merged weights $\Theta^\ell_{\mathrm{infer}} = W^\ell_{\mathrm{PT}} + \sum_i g_i A_i^\ell B_i^\ell$ with $g_i = \mathrm{softmax}(\beta f(q)^\top K)_i$. The stored memories are not merged into the shared weights, so later injections cannot overwrite earlier ones, yet the model can still reconstruct a full paragraph from a question cue, answer fresh questions about it, and even mix two adapters to answer questions that require information from two memories. The paper reports correct gate selection of 85.0% and 87.8% on its two datasets and attributes the remaining performance decline to softmax gating failures, not to interference between the adapters.

Load-bearing premise

The load-bearing premise is that the embedding function $f$, defined as the average of the base model's internal activations at the input to the final MLP layer, ranks the correct memory's stored key above all other keys for an arbitrary user question; if the ranking fails, the gate activates the wrong adapter and the memory is not retrieved even though its stored weights remain intact.

Editorial extensions

If this is right

  • Sequential one-by-one injection of 50 event memories into Llama-3.1-8B-Instruct yields QA accuracy of 72.5% (fictional stories) and 78.0% (2024 Wikipedia events), within roughly ten points of RAG using the same embedding model and top-1 retrieval.
  • Earlier memories survive later ones: recall of the first stored sample stays near its initial quality as the memory sequence grows, and MEGa's forgetting curves are nearly flat, whereas un-gated fine-tuning quickly defaults to reciting only the most recent memory.
  • General knowledge is preserved: MMLU accuracy after injection stays at 61.75% and 61.99%, compared with 62.56% for the base model, while the continual fine-tuning baselines fall to the mid-40s to mid-50s range.
  • Because gates are soft, mixing two adapters in one forward pass answers compositional questions requiring two separate memories at 49.6% and 70.4%, far above other continual baselines and comparable to full-parameter batch learning.
  • A two-stage internal-RAG variant, in which the model first recalls the relevant story and then answers from that recalled text, reaches 80.7% and 84.7% QA accuracy, nearly matching external RAG while keeping all memory inside the weights.

Reading between the lines

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

  • The authors' own diagnostic—that gating failures, not adapter interference, drive the residual forgetting—suggests a cheap scaling path they did not pursue: replace the average-activation key with a contrastively trained or multi-vector retriever, which would raise the correct-gate rate without retraining any LoRA adapter. Their Figure 8 already shows a stronger embedding model cuts forgetting, so
  • A stress test the paper does not run: store two memories that share the same protagonist, place, and date but differ in the outcome, then ask about the differing fact. Softmax gating cannot return two keys, so the model would either blend the adapters and hallucinate a mixture or choose only one memory; a winner-take-all or hierarchical gate would be needed for this kind of same-context, different
  • If the gating analysis generalizes beyond 50 memories, MEGa's real bottleneck becomes retrieval, not storage: the adapter count and parameter count grow linearly with each memory, so at thousands of memories the softmax over all keys becomes a nearest-neighbor search problem. Treating the context keys as an index and the adapters as a separate engram store would make MEGa a hybrid of RAG and weigh
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 4 minor

Summary. The paper introduces MEGa (Memory Embedded in Gated LLMs), a continual learning method that stores each new episodic memory in a dedicated LoRA adapter and, at inference, mixes adapters via softmax gating computed from the similarity between the query embedding and stored memory embeddings. The embedding function f is the average of the base model's internal activations at the input to the final MLP layer. Experiments on a synthetic fictional-character dataset and on Wikipedia 2024 events use 50 memories per partition over 20 partitions, with memory recall (cosine similarity), QA accuracy (GPT judge), and MMLU retention as the main metrics. MEGa is compared to full fine-tuning, LoRA, L2 and EWC regularization, batch learning, and RAG. The central claims are that MEGa mitigates catastrophic forgetting, achieves QA accuracy close to RAG, preserves general knowledge, and can answer compositional questions requiring knowledge from two memories.

Significance. If the results hold, MEGa is a useful and conceptually clean architecture-based approach to sequential knowledge injection: it requires no external database, uses a single unified embedding function, and the gating design is biologically motivated. The paper ships code and data, uses 20 dataset partitions and standard baselines, and reports means and standard deviations, which is good experimental practice. The strongest empirical results are the large gaps between MEGa and the regularization-based CL baselines on recall and QA, and the preservation of MMLU accuracy. However, the central mechanistic claim about gating is not backed by the measurements reported, and two design choices—hyperparameter selection for baselines and the construction of evaluation questions—require scrutiny before the comparison is fully convincing.

major comments (3)
  1. [Section 4.2, Eq. (3)-(4)] The paper attributes the residual decline in performance with an increasing number of stored memories to 'failures in the softmax-based gating selection mechanism' (Section 4.2) and supports this with correct-gate rates of 85.0% and 87.8% (Section 4.1). However, the correct-gate rate is defined only as argmax correctness; because inference weights in Eq. (3) are a softmax-weighted sum over all 50 adapters, even when the argmax is correct the other adapters contribute nonzero updates. The paper does not report QA accuracy or recall cosine conditioned on the top-1 gate being correct, nor the average probability mass assigned to the correct gate as the number of memories grows. Without such conditioning, the residual decline could be caused by softmax dilution or adapter interference, and the attribution to the gating selection mechanism is not established. I recommend an oracle-gating experiment that hard-selects the ground-truth adapter to separate these causes.
  2. [Appendix C (Table 4, Figure 6)] The main experiments choose L2 coefficient 0.1 even though the hyperparameter search in Table 4 shows that L2 coefficient 1.0 yields substantially better continual learning performance at the end of 50 samples (e.g., FULL L2 QA accuracy of 29.50% vs 27.73% on Fictional Character and 45.43% vs 33.20% on Wiki Events). The stated reason is that L2=1 fails to inject a single fictional character story due to refusal behavior. This means the baseline numbers in Table 1 are not the best achievable for L2-regularized baselines on the reported final metric, and the main comparison may overstate MEGa's advantage. Please either report the L2=1 results in Table 1 or justify why single-story refusal should exclude that hyperparameter while MEGa's own hyperparameters (beta, rank, alpha) are selected for best final performance.
  3. [Appendix E.1.3 and E.2] The evaluation questions are generated with an explicit instruction to 'provide AMPLE context in the question, including information from the original passage as context' (Appendix E.1.3), and compositional questions are similarly required to 'provide ample context from the content of the passages' (Appendix E.2). Since gating in Eq. (4) is based on the similarity between the query embedding and stored memory embeddings, this question construction likely inflates the reported correct-gate rates (85.0%/87.8%) relative to what a user would naturally ask. Please report retrieval accuracy and QA accuracy on questions that do not recapitulate the passage content, or at least quantify how gate accuracy depends on the amount of context in the question.
minor comments (4)
  1. [Table 1 caption] The RAG recall metric is defined as 'hit rate · 1 + (1 − hit rate) · c' without defining c in the caption; the text later defines c as the average cosine similarity between randomly selected training samples, but this should appear at the first use, and the hit rate should be reported separately.
  2. [Section 2.2 / References] One reference, 'Wang, Y., Liu, X., Chen, X., O'Brien, S., Wu, J., and McAuley, J. Self-updatable large language models by integrating context into model parameters. In The Thirteenth International Conference on Learning Representations.', lacks a year, unlike all other entries in the reference list.
  3. [Section 3.4.2] The GPT judge is run with temperature 1.0, which introduces stochasticity into the reported QA accuracies; the paper should either run the judge multiple times and report the variance, or use a lower temperature to make the evaluation deterministic.
  4. [Section 5 (Discussion)] The text contains a typo: 'Our preesent ''autobiographic'' stories' should be 'Our present ''autobiographic'' stories'.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: MEGa is evaluated against external baselines and MMLU; Eq. (4) is a design choice, not a fitted derivation, and the self-citations are motivational rather than load-bearing.

full rationale

The paper's only formal relation, Eq. (4) (g = softmax(β f(q)^T K)), is a proposed inference mechanism rather than a derived prediction; MEGa's claims are evaluated against external baselines (RAG, full/LoRA continual fine-tuning, batch learning) and an external benchmark (MMLU) in Table 1. The paper does not fit a parameter and then present a closely related quantity as a prediction: β is stated as a design choice (β = 1 for main experiments, β = 0.1 for compositional questions), and L2/EWC coefficients are swept and reported as hyperparameters in Appendix C. The embedding f is used both for stored keys and query gating, but key-query similarity is an empirical property tested by the reported correct-gate rates (85.0% and 87.8%); it is not guaranteed by construction, so this is a potential weakness (gating errors limit scaling) rather than circularity. Self-citations (Shan et al. 2024; Li & Sompolinsky 2022; Kanter & Sompolinsky 1987) provide background motivation or prior theory and are not used to force the MEGa design or its evaluation. The manuscript also flags its own limitations (linear parameter growth, small memory scale, need for better embeddings), and none of those limitations amounts to a self-referential derivation. Overall, no load-bearing step reduces to its inputs.

Assumptions & free parameters 4 free parameters · 4 assumptions · 0 invented entities

The MEGa pipeline depends on a small set of assumptions and tuned values: the gating temperature, LoRA rank and scale, regularization coefficients, and the embedding quality of the chosen internal layer. None are derived from first principles; they are selected by ablations and sweeps, and the paper itself identifies embedding quality as the main bottleneck. No new physical or conceptual entities are introduced.

free parameters (4)
  • gating softmax temperature beta = 1 for main experiments, 0.1 for compositional questions
    Controls how peaked the gating distribution is; the value was changed between task types after observing that beta=1 prevents mixing of two adapters needed for compositional questions (Section 3.3.2).
  • LoRA rank r and scaling factor alpha = r=128, alpha=128
    Selected from rank ablations (Appendix D); higher rank increases per-memory parameter cost and capacity. This is a standard hyperparameter, not a fitted physical constant.
  • L2 regularization coefficient lambda_L2 = 0.1
    The hyperparameter sweep favored lambda=1.0 for continual metrics, but the authors chose 0.1 because lambda=1.0 caused the model to refuse to inject a single fictional story (Appendix C, Table 4 and text). This is a hand-selected value after observing failure behavior.
  • EWC coefficient lambda_EWC = 1.0
    Chosen from hyperparameter sweep (Appendix C, Table 4).
assumptions (4)
  • domain assumption The average of the base model's internal activations at the input to the final MLP layer is a reliable semantic embedding for matching user queries to stored memory paragraphs.
    This is assumed throughout Section 3.3 and validated only empirically in Appendix Figures 7-9; the paper's own results show gate accuracy of 85-88% with Llama embeddings, so the assumption is imperfect.
  • domain assumption Independently fine-tuned per-sample LoRA adapters, when combined by softmax gating, produce minimal interference with each other and with the base model's general knowledge.
    The method relies on the additive LoRA formulation in Eq. (3) without formal interference bounds; empirical MMLU stability supports it, but it is not derived.
  • domain assumption The GPT-4.5-generated paraphrases and QA pairs, and the GPT-o3-mini judge, accurately reflect memory content and answer correctness.
    Used for fine-tuning and evaluation (Sections 3.2, 3.4, Appendix E). The judge is a language model, not a human, and temperature 1.0 makes the judge stochastic.
  • domain assumption The softmax gating in Eq. (4) with temperature beta can be set to allow compositional mixing of two memories.
    For compositional questions, beta is lowered to 0.1 so that two relevant adapters both receive nonzero weight; this is not derived and depends on embedding geometry.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Memorization and Knowledge Injection in Gated LLMs." pith.science (2026). https://pith.science/paper/DONZPDT3

@misc{pith2026250421239,
  author       = {Pith},
  title        = {Pith review of: Memorization and Knowledge Injection in Gated LLMs},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/DONZPDT3}},
  note         = {Machine review of arXiv:2504.21239}
}
read the original abstract

Large Language Models (LLMs) currently struggle to sequentially add new memories and integrate new knowledge. These limitations contrast with the human ability to continuously learn from new experiences and acquire knowledge throughout life. Most existing approaches add memories either through large context windows or external memory buffers (e.g., Retrieval-Augmented Generation), and studies on knowledge injection rarely test scenarios resembling everyday life events. In this work, we introduce a continual learning framework, Memory Embedded in Gated LLMs (MEGa), which injects event memories directly into the weights of LLMs. Each memory is stored in a dedicated set of gated low-rank weights. During inference, a gating mechanism activates relevant memory weights by matching query embeddings to stored memory embeddings. This enables the model to both recall entire memories and answer related questions. On two datasets - fictional characters and Wikipedia events - MEGa outperforms baseline approaches in mitigating catastrophic forgetting. Our model draws inspiration from the complementary memory system of the human brain.

Figures

Figures reproduced from arXiv: 2504.21239 by the authors.

Figure 1
Figure 1. Illustration of our fine-tuning and inference pipeline. We store the embedding of each sample as its context key and initialize a LoRA adapter for each sample (i.e., raw text paragraphs). LoRA adapters are fine-tuned individually using a constructed user prompt. During inference, the base model processes the query to generate its embedding, which is then used to compute the LoRA gating weights. Questions can be augm… view at source ↗
Figure 2
Figure 2. Recall forgetting curve comparing MEGa and other ap￾proaches. The model is prompted to recall the fine-tuned sample associated with a given question. Cosine similarity is computed between the recalled text and the original fine-tuning sample. The “Random pairs” baseline represents the average cosine similarity between embeddings of two randomly selected samples (0.465 for the Fictional Character dataset and 0.226 fo… view at source ↗
Figure 3
Figure 3. QA accuracy forgetting curve comparing MEGa and other approaches. The curve shows the QA accuracy for ques￾tions related to the first sample in the sequence, measured as the model is sequentially trained on additional samples. As more sam￾ples are introduced, accuracy on the first sample declines. For the Fictional Character dataset, the base model’s QA accuracy is 0.13%, overlapping with the x-axis [PITH_FULL_IMAG… view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: MMLU accuracy during continual fine-tuning. 4.5. Composite Knowledge MEGa can mix different LoRA adapters by applying softmax-weighted gating, enabling the combination of knowledge from multiple memories. To test this capability, we evaluate whether such a mixture can …
Figure 5
Figure 5. Figure 5: We found that the effectiveness of injecting new knowledge into instruct model depends on the “fine-tuning prompt”. The category “fact” represents the prompts (n=15) that indicate the finetuning sample is a fact, for example “Please tell me a story that you memorized:”…
Figure 6
Figure 6. Figure 6: Comparison between L2 coefficient 0.1 and 1.0. L2 coefficient 1.0 elicits an interesting increasing trend in the QA accuracy forgetting curve. The model refuses to answer the question after finetuning on the first sample. The refusal behavior is removed after finetunin…
Figure 7
Figure 7. Figure 7: Analysis on the embedding quality in Llama intermediate layers. Correct gate rate is computed as the rate of question embeddings which have the maximum inner product with the relevant passage embedding. It is computed for each dataset partition, which has 50 passage sa…
Figure 8
Figure 8. Figure 8: Due to the modular design, MEGa’s performance can be instantly boosted by replacing Llama embedding by other embedding models. We compare the performance of MEGa with Llama embeddings and OpenAI text-embedding-3-large embeddings [PITH_FULL_IMAGE:figures/full_fig_p019_8.png]
Figure 9
Figure 9. Figure 9: Comparing Llama embedding with OpenAI text-embedding-3-large embedding. The datasets used in the left plot are 20 dataset partitions, each with 50 samples (i.e. stories), as in other experiments. In the right plot, to test how does the correct gate rate scaling with nu…

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

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

  1. Diffusion-Inspired Masked Fine-Tuning for Knowledge Injection in Autoregressive LLMs

    cs.CL 2025-10 unverdicted novelty 6.0 of 10

    Masked fine-tuning enables autoregressive LLMs to inject new factual knowledge without paraphrases and with reversal-curse resistance, matching diffusion LLM advantages on QA tasks.

Reference graph

Works this paper leans on

97 extracted references · 19 canonical work pages · cited by 1 Pith paper

  1. [1]

    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 format.date year duplicate empty "emp...

  2. [2]

    Longbench: A bilingual, multitask benchmark for long context understanding

    Bai, Y., Lv, X., Zhang, J., Lyu, H., Tang, J., Huang, Z., Du, Z., Liu, X., Zeng, A., Hou, L., et al. Longbench: A bilingual, multitask benchmark for long context understanding. arXiv preprint arXiv:2308.14508, 2023

    arXiv 2023

  3. [3]

    a is b" fail to learn

    Berglund, L., Tong, M., Kaufmann, M., Balesni, M., Stickland, A. C., Korbak, T., and Evans, O. The reversal curse: Llms trained on" a is b" fail to learn" b is a". arXiv preprint arXiv:2309.12288, 2023

    arXiv 2023

  4. [4]

    Biderman, D., Portes, J., Ortiz, J. J. G., Paul, M., Greengard, P., Jennings, C., King, D., Havens, S., Chiley, V., Frankle, J., et al. Lora learns less and forgets less. arXiv preprint arXiv:2405.09673, 2024

    arXiv 2024

  5. [5]

    and Schulz, E

    Binz, M. and Schulz, E. Turning large language models into cognitive models. In The Twelfth International Conference on Learning Representations, 2024. URL https://openreview.net/forum?id=eiC4BKypf1

    2024

  6. [6]

    Concepts as semantic pointers: A framework and computational model

    Blouw, P., Solodkin, E., Thagard, P., and Eliasmith, C. Concepts as semantic pointers: A framework and computational model. Cognitive science, 40 0 (5): 0 1128--1162, 2016

    2016

  7. [7]

    D., Dhariwal, P., Neelakantan, A., Shyam, P., Sastry, G., Askell, A., et al

    Brown, T., Mann, B., Ryder, N., Subbiah, M., Kaplan, J. D., Dhariwal, P., Neelakantan, A., Shyam, P., Sastry, G., Askell, A., et al. Language models are few-shot learners. Advances in neural information processing systems, 33: 0 1877--1901, 2020

    1901

  8. [8]

    Buehler, E. L. and Buehler, M. J. X-lora: Mixture of low-rank adapter experts, a flexible framework for large language models with applications in protein mechanics and molecular design. APL Machine Learning, 2 0 (2), 2024

    2024

Show all 97 references
  1. [9]

    J., Chen, C.-T., Cheng, J.-H., and Huang, H.-H

    Chan, B. J., Chen, C.-T., Cheng, J.-H., and Huang, H.-H. Don't do rag: When cache-augmented generation is all you need for knowledge tasks. arXiv preprint arXiv:2412.15605, 2024

    2024 arXiv

  2. [10]

    How is chatgpt's behavior changing over time? arXiv preprint arXiv:2307.09009, 2023

    Chen, L., Zaharia, M., and Zou, J. How is chatgpt's behavior changing over time? arXiv preprint arXiv:2307.09009, 2023

    2023 arXiv

  3. [11]

    Recall and learn: Fine-tuning deep pretrained language models with less forgetting

    Chen, S., Hou, Y., Cui, Y., Che, W., Liu, T., and Yu, X. Recall and learn: Fine-tuning deep pretrained language models with less forgetting. arXiv preprint arXiv:2004.12651, 2020

    2004 arXiv

  4. [12]

    Empirical evaluation of gated recurrent neural networks on sequence modeling

    Chung, J., Gulcehre, C., Cho, K., and Bengio, Y. Empirical evaluation of gated recurrent neural networks on sequence modeling. arXiv preprint arXiv:1412.3555, 2014

    2014 arXiv

  5. [13]

    X., and Schulz, E

    Coda-Forno, J., Binz, M., Wang, J. X., and Schulz, E. Cogbench: a large language model walks into a psychology lab. In Forty-first International Conference on Machine Learning, 2024. URL https://openreview.net/forum?id=Q3104y8djk

    2024

  6. [14]

    Larimar: Large language models with episodic memory control

    Das, P., Chaudhury, S., Nelson, E., Melnyk, I., Swaminathan, S., Dai, S., Lozano, A., Kollias, G., Chenthamarakshan, V., Dan, S., et al. Larimar: Large language models with episodic memory control. arXiv preprint arXiv:2403.11901, 2024

    2024 arXiv

  7. [15]

    Research on the online update method for retrieval-augmented generation (rag) model with incremental learning

    Fan, Y., Wang, Y., Liu, L., Tang, X., Sun, N., and Yu, Z. Research on the online update method for retrieval-augmented generation (rag) model with incremental learning. arXiv preprint arXiv:2501.07063, 2025

    2025 arXiv

  8. [16]

    Switch transformers: Scaling to trillion parameter models with simple and efficient sparsity.(2021)

    Fedus, W., Zoph, B., and Shazeer, N. Switch transformers: Scaling to trillion parameter models with simple and efficient sparsity.(2021). arXiv preprint cs.LG/2101.03961, 2021

    2021 arXiv

  9. [17]

    A., Oomerjee, A., Christopoulou, F., Lampouras, G., Bou-Ammar, H., and Wang, J

    Fountas, Z., Benfeghoul, M. A., Oomerjee, A., Christopoulou, F., Lampouras, G., Bou-Ammar, H., and Wang, J. Human-like episodic memory for infinite context llms. arXiv preprint arXiv:2407.09450, 2024

    2024

  10. [18]

    French, R. M. Catastrophic forgetting in connectionist networks. Trends in cognitive sciences, 3 0 (4): 0 128--135, 1999

    1999

  11. [19]

    and Stratos, K

    Gangadhar, G. and Stratos, K. Model editing by standard fine-tuning. arXiv preprint arXiv:2402.11078, 2024

    2024 arXiv

  12. [20]

    and Derrida, B

    Gardner, E. and Derrida, B. Optimal storage properties of neural network models. Journal of Physics A: Mathematical and general, 21 0 (1): 0 271, 1988

    1988

  13. [21]

    Does fine-tuning llms on new knowledge encourage hallucinations? arXiv preprint arXiv:2405.05904, 2024

    Gekhman, Z., Yona, G., Aharoni, R., Eyal, M., Feder, A., Reichart, R., and Herzig, J. Does fine-tuning llms on new knowledge encourage hallucinations? arXiv preprint arXiv:2405.05904, 2024

    2024 arXiv

  14. [22]

    J., Fiete, I., and Irie, K

    Gershman, S. J., Fiete, I., and Irie, K. Key-value memory in the brain. arXiv preprint arXiv:2501.02950, 2025

    2025 arXiv

  15. [23]

    Model editing at scale leads to gradual and catastrophic forgetting

    Gupta, A., Rao, A., and Anumanchipalli, G. Model editing at scale leads to gradual and catastrophic forgetting. arXiv preprint arXiv:2401.07453, 2024 a

    2024 arXiv

  16. [24]

    K., and Soares, G

    Gupta, N., Kirtania, S., Gupta, P., Kariya, K., Gulwani, S., Iyer, A., Parthasarathy, S., Radhakrishna, A., Rajamani, S. K., and Soares, G. Stackfeed: Structured textual actor-critic knowledge base editing with feedback. arXiv preprint arXiv:2410.10584, 2024 b

  17. [25]

    J., Shu, Y., Gu, Y., Yasunaga, M., and Su, Y

    Guti \'e rrez, B. J., Shu, Y., Gu, Y., Yasunaga, M., and Su, Y. Hipporag: Neurobiologically inspired long-term memory for large language models. In The Thirty-eighth Annual Conference on Neural Information Processing Systems, 2024

    2024

  18. [26]

    Machine psychology: Investigating emergent capabilities and behavior in large language models using psychological methods

    Hagendorff, T. Machine psychology: Investigating emergent capabilities and behavior in large language models using psychological methods. arXiv preprint arXiv:2303.13988, 1, 2023

    2023 arXiv

  19. [27]

    A., Mukherjee, S., Tang, X., et al

    Han, H., Wang, Y., Shomer, H., Guo, K., Ding, J., Lei, Y., Halappanavar, M., Rossi, R. A., Mukherjee, S., Tang, X., et al. Retrieval-augmented generation with graphs (graphrag). arXiv preprint arXiv:2501.00309, 2024

    2024 arXiv

  20. [28]

    Aging with grace: Lifelong model editing with discrete key-value adaptors

    Hartvigsen, T., Sankaranarayanan, S., Palangi, H., Kim, Y., and Ghassemi, M. Aging with grace: Lifelong model editing with discrete key-value adaptors. Advances in Neural Information Processing Systems, 36, 2024

    2024

  21. [29]

    and Schmidhuber, J

    Hochreiter, S. and Schmidhuber, J. Long short-term memory. Neural computation, 9 0 (8): 0 1735--1780, 1997

    1997

  22. [30]

    Hopfield, J. J. Neural networks and physical systems with emergent collective computational abilities. Proceedings of the national academy of sciences, 79 0 (8): 0 2554--2558, 1982

    1982

  23. [31]

    K.-M., Lin, T.-H., Liao, C.-W., Fang, H.-C., Huang, C.-W., and Chen, Y.-N

    Hsueh, C.-H., Huang, P. K.-M., Lin, T.-H., Liao, C.-W., Fang, H.-C., Huang, C.-W., and Chen, Y.-N. Editing the mind of giants: An in-depth exploration of pitfalls of knowledge editing in large language models. arXiv preprint arXiv:2406.01436, 2024

    2024 arXiv

  24. [32]

    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. Lora: Low-rank adaptation of large language models. arXiv preprint arXiv:2106.09685, 2021

    2021 arXiv

  25. [33]

    Mitigating catastrophic forgetting in large language models with self-synthesized rehearsal

    Huang, J., Cui, L., Wang, A., Yang, C., Liao, X., Song, L., Yao, J., and Su, J. Mitigating catastrophic forgetting in large language models with self-synthesized rehearsal. arXiv preprint arXiv:2403.01244, 2024

    2024 arXiv

  26. [34]

    Janik, R. A. Aspects of human memory and large language models. arXiv preprint arXiv:2311.03839, 2023

    2023 arXiv

  27. [35]

    Q., Sablayrolles, A., Roux, A., Mensch, A., Savary, B., Bamford, C., Chaplot, D

    Jiang, A. Q., Sablayrolles, A., Roux, A., Mensch, A., Savary, B., Bamford, C., Chaplot, D. S., Casas, D. d. l., Hanna, E. B., Bressand, F., et al. Mixtral of experts. arXiv preprint arXiv:2401.04088, 2024

    2024 arXiv

  28. [36]

    Jung, M. J. and Kim, J. Pmoe: Progressive mixture of experts with asymmetric transformer for continual learning. arXiv preprint arXiv:2407.21571, 2024

    2024 arXiv

  29. [37]

    A rank stabilization scaling factor for fine-tuning with lora

    Kalajdzievski, D. A rank stabilization scaling factor for fine-tuning with lora. arXiv preprint arXiv:2312.03732, 2023

    2023 arXiv

  30. [38]

    Unfamiliar finetuning examples control how language models hallucinate

    Kang, K., Wallace, E., Tomlin, C., Kumar, A., and Levine, S. Unfamiliar finetuning examples control how language models hallucinate. arXiv preprint arXiv:2403.05612, 2024

    2024 arXiv

  31. [39]

    and Sompolinsky, H

    Kanter, I. and Sompolinsky, H. Associative recall of memory without errors. Physical Review A, 35 0 (1): 0 380, 1987

    1987

  32. [40]

    A., Milan, K., Quan, J., Ramalho, T., Grabska-Barwinska, A., et al

    Kirkpatrick, J., Pascanu, R., Rabinowitz, N., Veness, J., Desjardins, G., Rusu, A. A., Milan, K., Quan, J., Ramalho, T., Grabska-Barwinska, A., et al. Overcoming catastrophic forgetting in neural networks. Proceedings of the national academy of sciences, 114 0 (13): 0 3521--3526, 2017

    2017

  33. [41]

    and Sompolinsky, H

    Kleinfeld, D. and Sompolinsky, H. Associative neural network model for the generation of temporal patterns. theory and application to central pattern generators. Biophysical journal, 54 0 (6): 0 1039--1051, 1988

    1988

  34. [42]

    M., and Raghunathan, A

    Kotha, S., Springer, J. M., and Raghunathan, A. Understanding catastrophic forgetting in language models via implicit inference. arXiv preprint arXiv:2309.10105, 2023

    2023 arXiv

  35. [43]

    Knowledge injection via prompt distillation

    Kujanp \"a \"a , K., Valpola, H., and Ilin, A. Knowledge injection via prompt distillation. arXiv preprint arXiv:2412.14964, 2024

    2024 arXiv

  36. [44]

    Mixout: Effective regularization to finetune large-scale pretrained language models

    Lee, C., Cho, K., and Kang, W. Mixout: Effective regularization to finetune large-scale pretrained language models. arXiv preprint arXiv:1909.11299, 2019

    1909 arXiv

  37. [45]

    Zero-shot relation extraction via reading comprehension

    Levy, O., Seo, M., Choi, E., and Zettlemoyer, L. Zero-shot relation extraction via reading comprehension. arXiv preprint arXiv:1706.04115, 2017

    2017 arXiv

  38. [46]

    u ttler, H., Lewis, M., Yih, W.-t., Rockt \

    Lewis, P., Perez, E., Piktus, A., Petroni, F., Karpukhin, V., Goyal, N., K \"u ttler, H., Lewis, M., Yih, W.-t., Rockt \"a schel, T., et al. Retrieval-augmented generation for knowledge-intensive nlp tasks. Advances in Neural Information Processing Systems, 33: 0 9459--9474, 2020

    2020

  39. [47]

    Z., Liu, E., Ross, A., Zeitoun, A., Neubig, G., and Andreas, J

    Li, B. Z., Liu, E., Ross, A., Zeitoun, A., Neubig, G., and Andreas, J. Language modeling with editable external knowledge. arXiv preprint arXiv:2406.11830, 2024

    2024 arXiv

  40. [48]

    and Sompolinsky, H

    Li, Q. and Sompolinsky, H. Globally gated deep linear networks. Advances in Neural Information Processing Systems, 35: 0 34789--34801, 2022

    2022

  41. [49]

    On the storage capacity of hopfield models with correlated patterns

    L \"o we, M. On the storage capacity of hopfield models with correlated patterns. The Annals of Applied Probability, 8 0 (4): 0 1216--1250, 1998

    1998

  42. [50]

    Moelora: Contrastive learning guided mixture of experts on parameter-efficient fine-tuning for large language models

    Luo, T., Lei, J., Lei, F., Liu, W., He, S., Zhao, J., and Liu, K. Moelora: Contrastive learning guided mixture of experts on parameter-efficient fine-tuning for large language models. arXiv preprint arXiv:2402.12851, 2024

    2024 arXiv

  43. [51]

    An empirical study of catastrophic forgetting in large language models during continual fine-tuning

    Luo, Y., Yang, Z., Meng, F., Li, Y., Zhou, J., and Zhang, Y. An empirical study of catastrophic forgetting in large language models during continual fine-tuning. arXiv preprint arXiv:2308.08747, 2023

    2023 arXiv

  44. [52]

    L., McNaughton, B

    McClelland, J. L., McNaughton, B. L., and Lampinen, A. K. Integration of new information in memory: new insights from a complementary learning systems perspective. Philosophical Transactions of the Royal Society B, 375 0 (1799): 0 20190637, 2020

    2020

  45. [53]

    Mecklenburg, N., Lin, Y., Li, X., Holstein, D., Nunes, L., Malvar, S., Silva, B., Chandra, R., Aski, V., Yannam, P. K. R., et al. Injecting new knowledge into large language models via supervised fine-tuning. arXiv preprint arXiv:2404.00213, 2024

    2024 arXiv

  46. [54]

    Locating and editing factual associations in gpt

    Meng, K., Bau, D., Andonian, A., and Belinkov, Y. Locating and editing factual associations in gpt. Advances in Neural Information Processing Systems, 35: 0 17359--17372, 2022 a

    2022

  47. [55]

    S., Andonian, A., Belinkov, Y., and Bau, D

    Meng, K., Sharma, A. S., Andonian, A., Belinkov, Y., and Bau, D. Mass-editing memory in a transformer. arXiv preprint arXiv:2210.07229, 2022 b

    2022 arXiv

  48. [56]

    Mitchell, E., Lin, C., Bosselut, A., Finn, C., and Manning, C. D. Fast model editing at scale. arXiv preprint arXiv:2110.11309, 2021

    2021 arXiv

  49. [57]

    o ksal, A., Imani, A., Fayyaz, M., and Sch \

    Modarressi, A., K \"o ksal, A., Imani, A., Fayyaz, M., and Sch \"u tze, H. Memllm: Finetuning llms to use an explicit read-write memory. arXiv preprint arXiv:2404.11672, 2024

    2024 arXiv

  50. [58]

    Few-shot fine-tuning vs

    Mosbach, M., Pimentel, T., Ravfogel, S., Klakow, D., and Elazar, Y. Few-shot fine-tuning vs. in-context learning: A fair comparison and evaluation. arXiv preprint arXiv:2305.16938, 2023

    2023 arXiv

  51. [59]

    Sgpt: Gpt sentence embeddings for semantic search

    Muennighoff, N. Sgpt: Gpt sentence embeddings for semantic search. arXiv preprint arXiv:2202.08904, 2022

    2022 arXiv

  52. [60]

    Short-term memory and long-term memory are still different

    Norris, D. Short-term memory and long-term memory are still different. Psychological bulletin, 143 0 (9): 0 992, 2017

    2017

  53. [61]

    Fine-tuning or retrieval? comparing knowledge injection in llms

    Ovadia, O., Brief, M., Mishaeli, M., and Elisha, O. Fine-tuning or retrieval? comparing knowledge injection in llms. arXiv preprint arXiv:2312.05934, 2023

    2023 arXiv

  54. [62]

    Propagating knowledge updates to lms through distillation

    Padmanabhan, S., Onoe, Y., Zhang, M., Durrett, G., and Choi, E. Propagating knowledge updates to lms through distillation. Advances in Neural Information Processing Systems, 36, 2024

    2024

  55. [63]

    In-context editing: Learning knowledge from self-induced distributions

    Qi, S., Yang, B., Jiang, K., Wang, X., Li, J., Zhong, Y., Yang, Y., and Zheng, Z. In-context editing: Learning knowledge from self-induced distributions. arXiv preprint arXiv:2406.11194, 2024

    2024 arXiv

  56. [64]

    Memorag: Boosting long context processing with global memory-enhanced retrieval augmentation

    Qian, H., Liu, Z., Zhang, P., Mao, K., Lian, D., Dou, Z., and Huang, T. Memorag: Boosting long context processing with global memory-enhanced retrieval augmentation. 2025

    2025

  57. [65]

    L., Poeppel, D., and Vo, V

    Raccah, O., Chen, P., Willke, T. L., Poeppel, D., and Vo, V. A. Memory in humans and deep language models: Linking hypotheses for model augmentation. arXiv preprint arXiv:2210.01869, 2022

    2022 arXiv

  58. [66]

    Saxe, A., Sodhani, S., and Lewallen, S. J. The neural race reduction: Dynamics of abstraction in gated networks. In International Conference on Machine Learning, pp.\ 19287--19309. PMLR, 2022

    2022

  59. [67]

    Efficiency of local learning rules in threshold-linear associative networks

    Sch \"o nsberg, F., Roudi, Y., and Treves, A. Efficiency of local learning rules in threshold-linear associative networks. Physical Review Letters, 126 0 (1): 0 018301, 2021

    2021

  60. [68]

    A rapid and efficient learning rule for biological neural circuits

    Sezener, E., Grabska-Barwi \'n ska, A., Kostadinov, D., Beau, M., Krishnagopal, S., Budden, D., Hutter, M., Veness, J., Botvinick, M., Clopath, C., et al. A rapid and efficient learning rule for biological neural circuits. BioRxiv, pp.\ 2021--03, 2021

    2021

  61. [69]

    Stochastic consolidation of lifelong memory

    Shaham, N., Chandra, J., Kreiman, G., and Sompolinsky, H. Stochastic consolidation of lifelong memory. Scientific Reports, 12 0 (1): 0 13107, 2022

    2022

  62. [70]

    Order parameters and phase transitions of continual learning in deep neural networks

    Shan, H., Li, Q., and Sompolinsky, H. Order parameters and phase transitions of continual learning in deep neural networks. arXiv preprint arXiv:2407.10315, 2024

    2024 arXiv

  63. [71]

    Outrageously large neural networks: The sparsely-gated mixture-of-experts layer

    Shazeer, N., Mirhoseini, A., Maziarz, K., Davis, A., Le, Q., Hinton, G., and Dean, J. Outrageously large neural networks: The sparsely-gated mixture-of-experts layer. arXiv preprint arXiv:1701.06538, 2017

    2017 arXiv

  64. [72]

    Continual learning of large language models: A comprehensive survey

    Shi, H., Xu, Z., Wang, H., Qin, W., Wang, W., Wang, Y., Wang, Z., Ebrahimi, S., and Wang, H. Continual learning of large language models: A comprehensive survey. arXiv preprint arXiv:2404.16789, 2024

    2024 arXiv

  65. [73]

    and Sutton, R

    Silver, D. and Sutton, R. S. Welcome to the era of experience. Preprint of a chapter to appear in Designing an Intelligence, edited by George Konidaris, MIT Press (forthcoming)., April 2025

    2025

  66. [74]

    How to complete domain tuning while keeping general ability in llm: Adaptive layer-wise and element-wise regularization

    Song, S., Xu, H., Ma, J., Li, S., Peng, L., Wan, Q., Liu, X., and Yu, J. How to complete domain tuning while keeping general ability in llm: Adaptive layer-wise and element-wise regularization. arXiv preprint arXiv:2501.13669, 2025

    2025 arXiv

  67. [75]

    Lamol: Language modeling for lifelong language learning

    Sun, F.-K., Ho, C.-H., and Lee, H.-Y. Lamol: Language modeling for lifelong language learning. arXiv preprint arXiv:1909.03329, 2019

    1909 arXiv

  68. [76]

    Teyler, T. J. and DiScenna, P. The hippocampal memory indexing theory. Behavioral neuroscience, 100 0 (2): 0 147, 1986

    1986

  69. [77]

    Understanding the limits of lifelong knowledge editing in llms

    Thede, L., Roth, K., Bethge, M., Akata, Z., and Hartvigsen, T. Understanding the limits of lifelong knowledge editing in llms. arXiv preprint arXiv:2503.05683, 2025

    2025

  70. [78]

    Gated linear networks

    Veness, J., Lattimore, T., Budden, D., Bhoopchand, A., Mattern, C., Grabska-Barwinska, A., Sezener, E., Wang, J., Toth, P., Schmitt, S., et al. Gated linear networks. In Proceedings of the AAAI conference on artificial intelligence, volume 35, pp.\ 10015--10023, 2021

    2021

  71. [79]

    Wise: Rethinking the knowledge memory for lifelong model editing of large language models

    Wang, P., Li, Z., Zhang, N., Xu, Z., Yao, Y., Jiang, Y., Xie, P., Huang, F., and Chen, H. Wise: Rethinking the knowledge memory for lifelong model editing of large language models. Advances in Neural Information Processing Systems, 37: 0 53764--53797, 2024 a

    2024

  72. [80]

    Self-updatable large language models by integrating context into model parameters

    Wang, Y., Liu, X., Chen, X., O'Brien, S., Wu, J., and McAuley, J. Self-updatable large language models by integrating context into model parameters. In The Thirteenth International Conference on Learning Representations

  73. [81]

    Towards lifespan cognitive systems

    Wang, Y., Han, C., Wu, T., He, X., Zhou, W., Sadeq, N., Chen, X., He, Z., Wang, W., Haffari, G., et al. Towards lifespan cognitive systems. arXiv preprint arXiv:2409.13265, 2024 b

    2024 arXiv

  74. [82]

    Infllm: Training-free long-context extrapolation for llms with an efficient context memory

    Xiao, C., Zhang, P., Han, X., Xiao, G., Lin, Y., Zhang, Z., Liu, Z., and Sun, M. Infllm: Training-free long-context extrapolation for llms with an efficient context memory. arXiv preprint arXiv:2402.04617, 2024

    2024 arXiv

  75. [83]

    Meteora: Multiple-tasks embedded lora for large language models

    Xu, J., Lai, J., and Huang, Y. Meteora: Multiple-tasks embedded lora for large language models. arXiv preprint arXiv:2405.13053, 2024 a

    2024 arXiv

  76. [84]

    Xu, Z., Jiang, F., Niu, L., Deng, Y., Poovendran, R., Choi, Y., and Lin, B. Y. Magpie: Alignment data synthesis from scratch by prompting aligned llms with nothing. arXiv preprint arXiv:2406.08464, 2024 b

    2024 arXiv

  77. [85]

    Model merging in llms, mllms, and beyond: Methods, theories, applications and opportunities

    Yang, E., Shen, L., Guo, G., Wang, X., Cao, X., Zhang, J., and Tao, D. Model merging in llms, mllms, and beyond: Methods, theories, applications and opportunities. arXiv preprint arXiv:2408.07666, 2024 a

    2024 arXiv

  78. [86]

    Memory3: Language modeling with explicit memory

    Yang, H., Lin, Z., Wang, W., Wu, H., Li, Z., Tang, B., Wei, W., Wang, J., Tang, Z., Song, S., et al. Memory3: Language modeling with explicit memory. arXiv preprint arXiv:2407.01178, 2024 b

    2024 arXiv

  79. [87]

    A., and Lam, W

    Yang, H., Zhang, Y., Xu, J., Lu, H., Heng, P. A., and Lam, W. Unveiling the generalization power of fine-tuned large language models. arXiv preprint arXiv:2403.09162, 2024 c

    2024 arXiv

  80. [88]

    The butterfly effect of model editing: Few edits can trigger large language models collapse

    Yang, W., Sun, F., Ma, X., Liu, X., Yin, D., and Cheng, X. The butterfly effect of model editing: Few edits can trigger large language models collapse. arXiv preprint arXiv:2402.09656, 2024 d

    2024 arXiv

  81. [89]

    Melo: Enhancing model editing with neuron-indexed dynamic lora

    Yu, L., Chen, Q., Zhou, J., and He, L. Melo: Enhancing model editing with neuron-indexed dynamic lora. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 38, pp.\ 19449--19457, 2024

    2024

  82. [90]

    J., and Ma, Y

    Zhai, Y., Tong, S., Li, X., Cai, M., Qu, Q., Lee, Y. J., and Ma, Y. Investigating the catastrophic forgetting in multimodal large language model fine-tuning. In Conference on Parsimony and Learning, pp.\ 202--227. PMLR, 2024

    2024

  83. [91]

    Dh-rag: A dynamic historical context-powered retrieval-augmented generation method for multi-turn dialogue

    Zhang, F., Zhu, D., Ming, J., Jin, Y., Chai, D., Yang, L., Tian, H., Fan, Z., and Chen, K. Dh-rag: A dynamic historical context-powered retrieval-augmented generation method for multi-turn dialogue. arXiv preprint arXiv:2502.13847, 2025

    2025 arXiv

  84. [92]

    A comprehensive study of knowledge editing for large language models

    Zhang, N., Yao, Y., Tian, B., Wang, P., Deng, S., Wang, M., Xi, Z., Mao, S., Zhang, J., Ni, Y., et al. A comprehensive study of knowledge editing for large language models. arXiv preprint arXiv:2401.01286, 2024

    2024 arXiv

  85. [93]

    Q., and Artzi, Y

    Zhang, T., Wu, F., Katiyar, A., Weinberger, K. Q., and Artzi, Y. Revisiting few-sample bert fine-tuning. arXiv preprint arXiv:2006.05987, 2020

    2006 arXiv

  86. [94]

    and Wu, J

    Zhang, X. and Wu, J. Dissecting learning and forgetting in language model finetuning. In The Twelfth International Conference on Learning Representations, 2024

    2024

  87. [95]

    Loraretriever: Input-aware lora retrieval and composition for mixed tasks in the wild

    Zhao, Z., Gan, L., Wang, G., Zhou, W., Yang, H., Kuang, K., and Wu, F. Loraretriever: Input-aware lora retrieval and composition for mixed tasks in the wild. arXiv preprint arXiv:2402.09997, 2024

    2024 arXiv

  88. [96]

    Model tailor: Mitigating catastrophic forgetting in multi-modal large language models

    Zhu, D., Sun, Z., Li, Z., Shen, T., Yan, K., Ding, S., Kuang, K., and Wu, C. Model tailor: Mitigating catastrophic forgetting in multi-modal large language models. arXiv preprint arXiv:2402.12048, 2024 a

    2024 arXiv

  89. [97]

    Initializing and retrofitting key-value adaptors for traceable model editing

    Zhu, H., Lan, Y., Li, X., and Qian, W. Initializing and retrofitting key-value adaptors for traceable model editing. 2024 b

    2024

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.