- Rating
- 3(+1)
- Confidence
- 2
- Informativeness
- 2
- Presentation
- 2
- Soundness
- 2
Read review
Paper summary
This paper proposes Parameters as Memory (PAM), a methodology that internalizes stable long-term agent memories into parameter-efficient adapters instead of repeatedly retrieving and inserting historical text into the prompt. The authors instantiate PAM through Reflective Memory Optimization (REMO), which divides long conversations into overlapping windows, normalizes entities and temporal information, and rewrites each window using multiple complementary views, including direct and inverse factual expressions, cross-turn reasoning chains, question-answer representations, and structured entity or timeline summaries. The resulting corpus is used to train LoRA adapters while keeping the backbone model frozen. At inference time, the model can recall historical information from the adapter without receiving the complete conversation history. Experiments on LoCoMo, LongMemEval, and MemoryAgentBench demonstrate improvements over the base model and naive parameter-memory training, particularly for temporal reasoning, personalization, summarization, and conflict resolution. However, explicit retrieval remains stronger for exact lookup and rapidly changing information, suggesting that parameter memory should complement rather than replace external memory.
Critical strengths
- A practical alternative to external memory. PAM introduces a clear framework for storing stable, frequently reused memories in lightweight LoRA adapters, reducing the need to repeatedly retrieve and insert long histories into the prompt.
- Well-designed memory rewriting pipeline. REMO constructs multiple access paths to the same memory through forward and inverse expressions, temporal reasoning, cross-turn composition, QA encoding, and entity-level organization. The controlled ablations show that the gains cannot be explained solely by QA-format alignment.
- Comprehensive empirical and efficiency evaluation. The paper evaluates two backbone models across LoCoMo, LongMemEval, and MemoryAgentBench, includes component and LoRA-rank ablations, and reports practical serving metrics such as prompt length, latency, KV-cache usage, storage, and offline training cost.
Critical weaknesses
Limited methodological novelty and unclear gain attribution. PAM primarily combines established techniques—sliding-window preprocessing, LLM-based data augmentation, and standard LoRA supervised fine-tuning—rather than introducing a new learning algorithm. Moreover, REMO produces substantially more training data than the NaiveSFT baseline, but the paper does not provide a data-volume-matched comparison using simpler paraphrasing or generic QA augmentation. Therefore, it is difficult to determine whether the gains arise from the proposed reflective strategy taxonomy or simply from additional synthetic supervision.
Limited practicality for dynamic, large-scale memory. PAM requires generating training data and updating a dedicated adapter before new information becomes usable. The reported adapter occupies approximately 1.34 GB, and the system does not implement automatic routing between parametric and external memory, reliable deletion, or explicit unlearning of outdated facts. These limitations make the approach less suitable for rapidly changing, privacy-sensitive, or exact-match information; indeed, RAG substantially outperforms PAM on LongMemEval accuracy.
Questions for authors
see weakness
5 comments
Response to Reviewer cbNL [2/3]
Response to Question 2
How practical is PAM for dynamic, large-scale, or privacy-sensitive memory, given data-generation and adapter-update costs, adapter storage, missing routing/deletion mechanisms, and the LongMemEval gap?
Response:
PAM internalizes reusable historical experience in parameters, exchanging offline writing for less historical context processing at query time. We evaluate continual updates, storage and query costs, and integration with external memory.
1. Continual updates and memory retention. PAM maintains an independent adapter for each user or memory unit, with a shared frozen backbone. New history is written by continuing to train the corresponding adapter.
We divided each of the first three LoCoMo histories chronologically into three stages and trained its adapter sequentially. Questions are grouped by when their supporting evidence first becomes available. The table reports F1 averaged equally across the three histories.
| Adapter state | Stage 1 questions | Stage 2 questions | Stage 3 questions |
|---|---|---|---|
| Base model | 7.75 | 5.51 | 5.47 |
| After Stage 1 | 31.33 | 16.03 | 17.30 |
| After Stage 1 → 2 | 22.12 | 46.04 | 18.02 |
| After Stage 1 → 2 → 3 | 22.45 | 29.87 | 42.68 |
Writing Stage 2 and Stage 3 improves F1 on their corresponding questions by 30.00 and 24.66 points, respectively, showing that an existing adapter can continually absorb new information. Continued writing also affects earlier memories. We therefore compared replay and cumulative-history rebuilding on LoCoMo Item 1 under the same total training budget.
LoCoMo Item 1: F1 after the Stage 3 update.
| Maintenance strategy | Stage 1 questions | Stage 2 questions | Stage 3 questions |
|---|---|---|---|
| Continued training without replay | 30.37 | 30.20 | 40.10 |
| Continued training with 20% replay | 46.05 | 35.30 | 37.50 |
| Adapter rebuilt from cumulative history | 48.36 | 46.02 | 38.37 |
Using 20% replay raises Stage 1 / 2 F1 to 46.05 / 35.30; cumulative-history rebuilding further raises these scores to 48.36 / 46.02. Under the fixed budget, replay allocates part of training to older data, shifting Stage 3 F1 from 40.10 to 37.50 and illustrating the trade-off between new learning and retention.
Continued training, replay, and cumulative-history rebuilding thus provide different maintenance options. The current implementation does not provide automatic routing or reliable fact-level deletion; information requiring precise deletion should remain in a managed external memory store.
Response to Reviewer cbNL [3/3]
2. Construction, storage, and query costs. We ran the complete 20-strategy REMO pipeline with Qwen2.5-7B-Instruct and trained rank-128 LoRA adapters on H200 GPUs. Construction times below are in minutes and include interrupted attempts.
| Dataset | Data generation | Adapter training | Total construction time |
|---|---|---|---|
| LoCoMo Item 1 | 74.84 | 15.66 | 90.50 |
| LoCoMo Item 2 | 51.31 | 17.63 | 68.94 |
| LoCoMo Item 3 | 100.19 | 16.96 | 117.16 |
LoCoMo Item 1: storage and loading costs.
| Method | Storage size | Loading time (s) |
|---|---|---|
| PAM (REMO) | 1.292 GB | 2.385 |
| BM25, including source text | 0.177 MB | 0.41 |
PAM loading measures attachment to an already resident backbone. At this adapter size, 1,000 users require approximately 1.292 TB of adapter storage. Storage and loading therefore remain part of the maintenance cost at scale.
For query measurements, we selected 128 fixed questions from Item 1 and alternated paired measurements over three rounds on the same shared H200 GPU, with batch size 1. Models and adapters remained resident; PAM used unmerged FP32 LoRA. Each call generated exactly 32 tokens, with prefix-KV reuse disabled. Values are medians. Generation time includes prefill and excludes retrieval, tokenization, loading, and network costs.
| RAG input condition | PAM generation time (s) | RAG generation time (s) | PAM prompt KV (MiB) | RAG prompt KV (MiB) |
|---|---|---|---|---|
| Actual BM25 top-5, 1,592.5 tokens | 0.910 | 0.463 | 1.04 | 87.09 |
| 8,192 tokens | 0.909 | 0.662 | 1.04 | 448.00 |
| 16,384 tokens | 0.909 | 0.954 | 1.04 | 896.00 |
| 24,576 tokens | 0.910 | 1.297 | 1.04 | 1,344.00 |
| 32,736 tokens | 0.910 | 1.656 | 1.04 | 1,790.25 |
PAM receives only the question, with a median input length of 19 tokens. Fixed-length inputs concatenate the same user's source text in BM25 relevance order, cycling it when needed to measure the computational cost of long inputs. RAG generates faster with short contexts. Under this configuration, latencies are close around 16k input tokens, after which PAM's advantage grows with input length. At 32,736 tokens, PAM achieves a paired speedup of 1.82×.
Counting parameters and prompt KV together, the memory crossover occurs when RAG uses 22,528 more input tokens than PAM, offsetting PAM's additional LoRA parameters. At 32,736 tokens, PAM saves approximately 557 MiB across these two components.
These measurements quantify the parametric-memory trade-off: offline construction and adapter maintenance reduce the context, computation, and KV costs of repeatedly processing history in subsequent queries.
3. LongMemEval and complementarity with external memory. In the original LongMemEval experiment, PAM achieves 25.8% accuracy and RAG 52.5%. The benchmark requires recovering specific facts, temporal evidence, and cross-session details. RAG supplies source evidence at query time, while PAM recovers information from parameters. Explicit retrieval directly supports such precise access.
We evaluated their combination on LongMemEval-S using Qwen2.5-7B-Instruct. Each question's full candidate history was used to train an independent rank-32 LoRA through four rewriting views: factual restatements, temporal information, user attributes, and source-derived QA. RAG and RAG+PAM used identical BM25 top-5 sessions and prompts, differing only in whether the corresponding adapter was enabled. This supplementary experiment reports F1 separately from the original accuracy results.
| Method | F1 |
|---|---|
| RAG | 15.61 |
| PAM | 13.03 |
| RAG + PAM | 26.53 |
With the same retrieval budget, RAG+PAM improves F1 from 15.61 to 26.53, a gain of 10.92 points. The original LoCoMo experiments with LLaMA-3.1-8B also demonstrate this complementarity:
| External memory system | Alone, F1 | With PAM, F1 |
|---|---|---|
| A-Mem | 23.97 | 34.63 |
| Mem0 | 26.70 | 39.20 |
PAM learns reusable historical experience, while external memory provides explicit access to specific evidence and recent information. These results support combining parametric memory and retrieval as complementary access mechanisms for different memory needs.
Response to Reviewer cbNL [1/3]
We thank the reviewer for recognizing PAM's motivation, multi-view rewriting pipeline, and evaluation. To address gain attribution and practicality, we added matched-budget controls, continual-update experiments, and construction and query cost measurements.
Response to Question 1
Response:
PAM contributes a supervision design and writing pipeline for long-term parametric memory. REMO organizes dispersed dialogue history into complementary views, including direct and inverse facts, cross-turn relations, temporal information, QA, and entity structures. Adapters internalize this information as reusable memory for subsequent generation, reducing repeated access to source history at query time.
1. Data construction under matched training budgets. We used Qwen2.5-7B-Instruct on the first three LoCoMo conversation histories, with three training conditions:
All conditions use the same backbone, rank-128 LoRA, optimizer, learning-rate schedule, and random seed, with 600 updates and 1,228,800 non-padding input tokens. The three histories contain 152, 81, and 152 evaluation questions. The metric is lexical F1, averaged equally across histories.
Under the same training-token budget, REMO outperforms Naive SFT on all three histories, improving average F1 by 30.62 points. Converting sessions into QA pairs raises average F1 from 8.98 to 34.51; REMO reaches 39.59, a further 5.08-point improvement over QA training. These results show that data construction affects memory QA performance and that REMO's gains do not arise solely from more training tokens. Full settings are provided in Supplementary Experiments S1–S2.
2. Comparisons with other memory internalization pipelines. We added task adaptations of EntiGraph, Active Reading, and on-policy self-distillation (OPSD). EntiGraph generates relational descriptions and summaries around source entities and their combinations. Active Reading plans study questions and strategies, then generates paraphrases, QA, and other exercises. In OPSD, the student samples responses and learns from a frozen teacher that additionally reads the source session.
LoCoMo results with Qwen2.5-7B-Instruct. Metric: lexical F1.
For this comparison, REMO, EntiGraph, and Active Reading use newly generated corpora from Qwen2.5-7B-Instruct, with matched corpus token volumes, adapter configurations, and training budgets. These runs are independent of those in the preceding table. OPSD uses a separate distillation setup. REMO achieves the highest F1 on all three histories in this comparison, exceeding EntiGraph and Active Reading by 25.96 and 12.97 points on average. Configurations and other evaluation results are provided in Supplementary Experiments S3–S4.
These experiments evaluate REMO against raw-history training, direct data augmentation, and alternative internalization pipelines. They support improving parametric memory through supervision design and show how training corpora and internalization methods affect memory QA performance.