Model Checkpoints

Open-source language models have a reproducibility problem. Despite releasing weights, training data, and recipes, none of them are provably reproducible due to the non-associativity of floating-point arithmetic. Deep learning frameworks often offer a deterministic execution mode, allowing reproducible operations on the same machines. Unfortunately, this determinism does not carry across hardware such that a user can verify that a released checkpoint was actually produced using the declared training recipe. This leaves room for undisclosed data, injected biases, or backdoors that existing techniques such as proof-of-learning or proof-of-training-data cannot rule out. We introduce a new tier of model transparency, fully auditable, in which every operation on every data sample during training is independently reproducible on heterogeneous commodity hardware with bitwise certainty. By imposing a definite order on the sources of training nondeterminism, GPU kernel reductions, data batch ordering across a data-parallel cluster, and inter/intra-node collective communication, we make it possible to replay any individual step of a large, distributed training run on a single piece of commodity hardware and check it against the published trajectory. Because replaying an entire run on one machine is infeasible, we support this with a collective verification scheme in which many independent auditors each certify individual steps, together covering the whole run. We release Open-1B, a model trained under this regime, together with its full pretraining dataset, every intermediate checkpoint, the training codebase, and the audit harness needed to reproduce and verify any step of its training.

An unlearning audit reads its verdict off numbers that an unlearned model and its retrained reference each publish, and both also ship batch-normalization statistics that no gradient step wrote and no release records. Refitting them on kept data at bit-identical weights moves 47 of 221 released checkpoints past the spread their own release's seeds show, several inside a method whose average does not move: what moves is the checkpoint's property, not its method's. What does the moving is not the removed data surviving in the state: exchanging kept records for removed ones inside a fixed fitting pool moves a published cell by almost nothing, while how far a checkpoint's shipped state has drifted from any refit does track it. The consequence for a published decision is real but narrow: twelve verdicts cross, four clear a measured recalibration budget, two clear it on every replicate, and a population we trained and sited near its own criterion yields none. A release should therefore name the fitting convention beside the number, on the batch-normalized vision models where this channel exists.

We study how model post-training and test-time inference design affect natural-language proof generation for hard olympiad mathematics. Starting from Nemotron 3 Ultra, we train two specialist checkpoints using supervised fine-tuning and reinforcement learning, and evaluate checkpoint choice, verification, and refinement. Based on these findings, we present an open-model test-time-compute pipeline. The system operates entirely in natural language, with no formal prover, external tools, or internet access. Three Nemotron 3 Ultra checkpoints - the general-availability model and two post-trained specialists - power an iterative search that generates, verifies, and refines candidate proofs; a separate high-compute stage then selects each final submission. The system scored 30 out of 42 points at IMO 2026, reaching the gold-medal threshold. We release the two post-trained checkpoints as well as the training data, the training and inference code, the submitted solutions, and Nemotron-IMO-Bench, a new benchmark of 200 novel olympiad-level problems.

LLM serving systems already reuse KV caches, but only when the reused text sits at the very start of the prompt. Two growing workloads break this condition: a retrieval-augmented generation server assembles a different set of retrieved chunks for every query, and a multi-agent coordinator reads reports written by other agents. Reused inside a new prompt, a cache carries the wrong positions and never attended to the other sources. The cache may also have been written by a different checkpoint of the same model family, which changes the stored values. Repair methods for such caches have appeared in three separate communities, each measured on its own terms, and existing benchmarks test only exact-prefix reuse, where nothing is lost. KVShareArena benchmarks KV-cache reuse across prompt contexts and model checkpoints on retrieved chunks and agent reports. It scores every method by the fraction of the gap it recovers between no cache and full recomputation, and charges compute, memory, and per-request latency with the cache in hand, reporting the one-time cost of building a cache separately. We find that correcting positions, which needs no recomputation, is enough until a question needs several sources at once. There, only methods that pay, by re-encoding part of the cache or by training, recover half to two thirds of the gap; unrepaired caches can be worse than no cache. Cache-compression methods that are harmless on a single prompt fall significantly behind position correction on freshly written agent reports. These patterns hold across three model boards. When a different checkpoint wrote the cache, training-free methods are barely affected, while an adapter trained on one checkpoint's caches loses quality. Harness, frozen querysets, and cost accounting ship as a pip package with an automated submission workflow and a public leaderboard.

Language models fine-tuned on customer behavior can predict outcomes and generate explanations, but these readouts are often treated as interchangeable. Holding model checkpoint and prompt content fixed, we compare probabilities obtained by scoring answer tokens with predictions generated after a written rationale. Across 13 model-domain cells covering four retail tasks in three markets, including two using fully public data and checkpoints, the scored readout ranks outcomes more accurately in 12 of 13 cells (two-sided sign test, p approximately 0.003), by 1.5 to 14.5 points in area under the receiver operating characteristic curve (AUC). Paired bootstrap confidence intervals exclude zero in every newly measured cell. The gap varies with task-specific supervision and mismatch between training and serving formats, ranging from -2.2 points for an untuned base model to +13.7 for rationale-format supervision. Analysis of approximately 9,000 rationales identifies two correlates: reduced reliance on the dominant predictive feature and convergence on stock formulations. Probability saturation does not track the gap. A third readout, eliciting a probability before any verdict, improves calibration (Brier score from 0.47 to 0.15) while ranking within noise of scoring, but only for outcome rates represented in training; it is worse than scoring when the scored head is already calibrated. We interpret these differences through the objectives matched by each readout, identify training choices that narrow the gap, and propose retaining generated rationales while sourcing ranking from the scored head.

Language-model checkpoints are commonly selected by pretraining loss or benchmark scores, assuming that the highest-scoring checkpoint will remain the best starting point for subsequent training. We show that this assumption can fail in a full 30B mixture-of-experts training pipeline. The checkpoints that perform better after the full downstream training stack also have higher solution density, i.e., retain downstream performance under local weight perturbations.

Across the full Pythia suite (160M-12B, eight checkpoints, four task families), a linear probe can read a target variable from the residual stream as early as step 1,000 at every scale -- yet steering along that same reading direction remains null-equivalent in 43 of 48 model-checkpoint cells. Internal readability systematically outruns causal efficacy, and the lag does not shrink with scale. We call this structure lagged coupling and decompose it into three dissociable tracks: (i) internal readability, saturated (AUROC >= 0.990) from the first checkpoint everywhere; (ii) behavioral readability, which develops gradually and progressively later at larger scales (12B reaches 0.909 only at the final checkpoint); (iii) causal efficacy, almost always null-equivalent, occasionally counterproductive early, with one isolated positive pulse (12B, step 8,000, z = +2.49) our grid cannot resolve. The ordering is dominantly read-before-write (11/11 units, no inversion). Representation headroom along the probe direction grows up to 57x with training and scale while causal write-in stays below 0.11% of headroom -- the variable is increasingly written into the representation and increasingly ignored by the readout. Under a fully pre-registered protocol, both single-onset hypotheses resolve INDETERMINATE (scale slope +0.24, 95% CI [-0.60, +0.87]; time vote 3:3) -- a disciplined negative explained by the three-track decomposition. A pre-registered OLMo-2 replication preserves the direction at attenuated magnitude. Our results caution against inferring steerability from probe accuracy and establish a developmental bottleneck: representation formation reliably outpaces causal readout consolidation.

Latent world-action models avoid rendering future pixels by predicting an action-relevant visual subgoal in feature space. LaWAM established this formulation, but its original presentation left the world model, multimodal backbone, and deployment checkpoint tightly coupled. We introduce AcrossWAM1.0, a modularization and scaling study of this latent world-action stack. Rather than presenting latent subgoals as a new algorithm, we make the module boundary explicit: a policy adapter produces latent-action and action-generation contexts; a retained latent world decoder grounds the predicted transition in the current scene;and a flow-matching expert generates continuous action chunks. We further separate training-only teachers from the inference graph and provide a verifiable deployment export. On 2,000 paired LIBERO episodes, replacing a Qwen3-VL-2B backbone with Qwen3.5-0.8B yields 97.45% success versus 98.00% for the 2B model (a-0.55percentage-point difference; exact McNemarp=0.266). This does not prove equivalence, but it meets a prespecified two-point retention criterion. The compact, inference-reachable checkpoint contains 1,472.6M unique parameters, 42.4% fewer than the original 2B policy, while all retained tensors are bitwise identical to the source checkpoint. Cross-family execution is additionally checked with a MiniCPM-V adapter smoke test; closed-loop cross-family transfer remains an open evaluation. AcrossWAM1.0 therefore contributes an auditable software and evaluation boundary for compact latent world-action policies, distinct from LaWAM's original latent-subgoal contribution.

Audio-visual speech enhancement (AVSE) uses visual-speech cues from a target speaker to recover that speaker's speech from noisy or overlapping speech. Many widely used protocols construct mixed signals from separately recorded audio sources and assume reliable video, leaving their performance under natural overlap and visual failure insufficiently characterized. The Real-World AVSE Challenge evaluates two related settings. Track1 comprises two scenarios: real-world mixtures recorded with two speakers speaking simultaneously, without a corresponding clean reference signal, and synthetic remixes obtained by manually mixing the separately recorded speech of two speakers, with a clean reference signal available; Track2 reuses audio but pairs it with a degraded target video and contains additional 3-m far-field recordings. The speakers in the development and test sets are disjoint. Evaluation metrics include clean-waveform fidelity, learned quality estimates, transcription accuracy, and speaker identification. In the remix task on the development set, the baseline model achieved an SI-SDR of $-4.069$dB and an STOI of $0.388$ on Track1, and an SI-SDR of $-2.851$dB and an STOI of $0.470$ on Track2. We release the AV-ConvTasNet checkpoints, the offline evaluator, and the official baseline results on the development and test sets.

While post-training improves the capabilities of large language models (LLMs), it generally lowers their output diversity and creativity, negatively impacting tasks that explicitly require creativity (e.g., story generation) as well as those that require it implicitly, e.g., reinforcement learning (RL). We instead propose CreativeInstruct, a scalable instruction-tuning method that teaches LLMs to balance creative, base-model-like generations with the quality of post-trained models, by learning to inject special [StartCreativity] spans that bias generation toward creativity. Furthermore, we introduce a structural diversity metric based on graph edit distance, which captures narrative level variation missed by purely lexical and semantic metrics. On narrative generation, CreativeInstruct matches or exceeds the diversity of both multi-model baselines and distilled variants of their outputs, without sacrificing quality or requiring multiple models at inference time. These results are mirrored in our human evaluation, where we find that annotators rate CreativeInstruct generations as more creative than the post-trained LLMs' generations in 70.3% of cases. We also show the benefits of creative models as a substrate for RL: GRPO applied to a CreativeInstruct checkpoint improves by ~4% on AMC and ~5% points on MATH over the same training applied to the post-trained checkpoint.

On-device training of deep neural networks is fundamentally constrained by the computational and memory costs of large-scale datasets. Coreset selection offers a practical solution by retaining only a compact subset of real training samples. However, existing gradient-based methods commonly rely on gradients computed at a single model snapshot and employ greedy or pursuit-based selection procedures, limiting their ability to capture evolving optimization dynamics and handle strongly correlated samples. We propose GLOBE (Gradient Local-Balanced Extraction), a trajectory-aligned coreset selection framework that formulates sample selection as a globally optimized sparse weighting problem. GLOBE represents each sample by a gradient trajectory constructed across multiple training checkpoints, thereby capturing its influence throughout different stages of optimization. To preserve the training behavior of the full dataset, we introduce a multi-order matching objective that jointly aligns the first-order mean and projected uncentered second-order moments of gradient trajectories. GLOBE further combines Group LASSO, Elastic Net regularization, and nonnegative budget constraints to induce group- and sample-level sparsity while stabilizing the weights of correlated trajectories. Finally, class-balanced Top-K selection maintains adequate category coverage under limited sampling budgets. Experiments across six benchmarks and five evaluation architectures demonstrate that GLOBE consistently outperforms existing coreset selection methods in downstream test accuracy, particularly at low retention ratios. These results highlight the effectiveness of combining dynamic gradient information, multi-order distribution matching, and structured sparsity for data-efficient learning.

Large chest radiography archives are difficult to search because most studies are paired only with free-text reports rather than structured clinical annotations. Vision-language models offer a natural interface for text-to-image retrieval, but current biomedical models are primarily optimized for report-to-image matching rather than for satisfying short clinical search queries. This creates an objective mismatch: a model may retrieve images related to words in the query while failing to satisfy the full clinical constraint, especially for conjunctions and negations such as ``atelectasis and no pneumonia.'' We introduce CXR-Retrieve, a structured benchmark for compositional chest X-ray text-to-image retrieval. The benchmark contains 5,159 test images from the official test-split of MIMIC-CXR-JPG and 145 textual queries spanning single and conjunction findings, both positive and negative. Relevance is defined by whether a retrieved image satisfies all asserted pathology constraints, rather than by whether it matches a paired report. We further propose a label-aware contrastive fine-tuning objective for clinical retrieval. Our method attracts image-text pairs with compatible asserted pathology constraints, including shared confirmed absences, while explicitly repelling contradictory pairs. Starting from the in-domain CXR-CLIP checkpoint, our method improves Precision@5 over CXR-CLIP by 8.5 percentage points on two-pathology conjunctions and by 22.0 percentage points on negation queries. These results show that reliable chest X-ray retrieval requires training objectives that model not only which findings are mentioned, but also how they are clinically asserted.

Developers judge a model checkpoint by how it behaves. After supervised fine-tuning (SFT), two checkpoints that perform about the same across relevant benchmarks are treated as interchangeable, equally ready for the next alignment stage, typically preference optimization. We ask whether this judgment misses a pretraining imprint: a difference that no post-SFT benchmark reveals, yet that decides how each checkpoint responds to further training. To find out, we run a controlled experiment on the final window of pretraining, the last data trained on before instruction tuning. Six branches fork from one partially pretrained checkpoint and differ only in this window: 500 million tokens, 0.1% to 1% of the tokens that precede it. Each branch trains its window on a single data source: generic web text, filtered web text, normative discourse, safety text, mathematical text, or synthetic educational text. SFT and post-training are then identical. After SFT the branches behave near-identically, within about one point on instruction following, refusal, and capability, yet the same post-training carries them to very different endpoints, under both a direct preference optimization update and a reinforcement learning update with a verifiable reward. We measure this deviation through refusal of harmful requests: when post-training begins the safety text branch refuses no more than the web text branch, yet by the end it has lost far less of its refusal. The other four branches gain little or no protection, so the effect is selective to what the window contained. The protection requires the safety text to arrive last rather than earlier in pretraining, and it reproduces on a second model family. What a model is pretrained on last shapes how it reacts to alignment. Therefore, a checkpoint should not be evaluated by its post-SFT behavior alone, and what it was trained on last should be reported with it.

This paper presents a safety-centered empirical evaluation of uncertainty-aware last-layer adaptation for referable diabetic retinopathy screening using RETFound, a self-supervised vision-transformer retinal foundation model used here as a frozen feature encoder, and the public APTOS 2019 and DDR diabetic retinopathy fundus image datasets. We compare a cached-feature softmax head, post-hoc temperature scaling, variational Bayesian last-layer heads, a diagonal Laplace last-layer approximation, and an SNGP-style cached-feature head. On APTOS, uncertainty-aware operating points improved sensitivity and selective-referral behavior. The strongest APTOS selective-referral result deferred approximately 20 percent of cases and reduced accepted-case false negatives to zero while preserving high accepted-case specificity. However, threshold tuning also reduced false negatives at high false-positive cost, so false-negative reduction alone was not unique to Bayesian modeling. On DDR, native Bayesian heads qualitatively reproduced the APTOS direction but with weaker tradeoffs, while the APTOS-trained SNGP checkpoint transferred poorly and failed to provide useful external selective-referral behavior. These results highlight the value of safety-centered evaluation beyond aggregate accuracy: uncertainty-aware last-layer heads can improve internal safety-oriented operating points, but trustworthy retinal screening claims require explicit safety-coverage evaluation and second-dataset validation under shift.

Small Language Models (SLMs) are increasingly being considered for deployment on edge devices such as laptops, enabling private, low-latency, and locally personalized applications. However, personalization requires models to adapt over time to evolving user- or task-specific data, placing them in a continual learning setting. This creates the risk of catastrophic forgetting, where learning new information degrades performance on previously learned tasks or broader model capabilities. Recent benchmarks such as TRACE have shown that continual fine-tuning can significantly degrade the general abilities of aligned large language models. In this work, we present a study for sequential LoRA personalization of SLMs. We save model checkpoints after each adaptation stage and evaluate them on current tasks, previously seen tasks, and a fixed reference set. This checkpoint-level protocol enables us to monitor task performance, forgetting, and reference set drift over time. We show that lightweight reference set distributional diagnostics can reveal model-specific instability patterns during sequential LoRA personalization of SLMs, including cases where task-level metrics alone hide harmful adaptation. We hope this can highlight new research avenues for monitoring stability of SLMs in a continual learning setting.

Next-token prediction is the self-supervised signal that trains language models, and every observed prompt token provides the same signal at test time. We study whether this signal can define the inner-loop objective for test-time training (TTT) in pretrained long-context language models. Many TTT architectures require models to be trained with test-time adaptation in mind, limiting their direct applicability to released LLM checkpoints. While recent in-place TTT methods make fast-weight adaptation possible for pretrained LLMs without redesigning the backbone, they leave a central question unresolved: what should each test-time write store? Existing recipes train the fast weight to match a learned local value proxy but they are not directly tied to the self-supervised next-token prediction signal. We introduce Test-Time Training with Next-Token Prediction (TTT-NTP), a drop-in fast-weight adaptation method for pretrained LLMs that instead supervises updates using the model's own next contextual hidden state. This makes each local write follow the same causal computation that supports next-token prediction: the value target is a pointwise linear projection of a single next-position contextual state. On RULER Full-13, averaged over 4k to 32k contexts, TTT-NTP is the only method that consistently improves the released backbone across four models spanning three families and a 0.6-8B size range, by 3.9 points on Llama-3.1-8B, 3.0 on Mistral-7B-v0.3, 4.1 on Qwen3-4B, and 2.9 on Qwen3-0.6B. On the real-world LongBench-v2 long-document QA benchmark, TTT-NTP improves over the base model by 5.6 points on Llama-3.1-8B and 3.7 on Mistral-7B-v0.3, while preserving commonsense and knowledge performance. Our code is publicly available at https://github.com/yancyou/TTT-NTP.

Block Attention Residuals (Block AttnRes) by replace fixed additive residuals with a learned softmax over earlier depth-source representations, surfacing cross-layer routing as an inspectable tensor in the forward pass. This is a tempting interpretability target: information flow normally inferred indirectly is now directly observable. We ask whether such exposure suffices for mechanistic interpretation. We probe two same-scale ($0.6$B) Block AttnRes checkpoints under identical routing-ablation interventions: a vanilla Qwen3 inference-wrapped through a deterministic recency-bias schedule that the codebase admits as a routing-equivalent loading path, and a Block AttnRes Qwen3 trained from scratch with routing as part of optimisation. The wrapped baseline's routing weights are content-independent and reproduce the schedule's analytic prediction. The trained AttnRes checkpoint instead exhibits three localised routing motifs: an embedding-source pathway through early-layer MLP, a current-state pathway through early-layer attention and MLP, and an older-history pathway through late-layer attention. Beyond this stratification, we find a sharp dissociation between average routing mass and causal importance: in both sublayers, the largest mass slice is not the largest causal contribution, and one source family carries appreciable mass with no detectable causal role under intervention. Architectural exposure of routing is therefore necessary but not sufficient for mechanistic interpretation: structured depth routing emerges only when routing has been part of training, and even then, descriptive routing summaries should be treated as candidate hypotheses to be tested by causal interventions, not as evidence of mechanism in their own right.

Quantized checkpoints are often screened first with quality metrics and only later, if at all, with direct safety tests. This paper audits that shortcut on a matched 51-row matrix spanning 6 models, 4 families, a 7-level GGUF ladder, and AWQ/GPTQ INT4 checkpoints. In this matrix the shortcut fails: all 36 quality-safety pairings split direction across models, and 9 hidden-danger rows plus 1 near-hidden-danger row show quality stable or improved while refusal falls by 12-68 percentage points. Seven of the 11 AWQ/GPTQ rows are hidden-danger. A four-probe mechanistic follow-up over the 17 Hugging Face-backed FP16/AWQ/GPTQ cells does not rescue it: entropy, refusal-direction, and calibration probes are weak or null separators of dangerous rows, and although probe-identified safety-associated neurons absorb 1.39$\times$ more quantization error overall ($p < 5 \times 10^{-7}$), the effect is not regime-specific. Claude Sonnet 4 relabels 11,470 items in a predefined stratified set, agrees with the primary gemma3:12b judge on 89.9% of rows ($κ= 0.873$, 95% CI [0.866, 0.881]), and changes 0/10 hidden-danger cells. A calibrated study-internal behavioral screen -- the Refusal Template Stability Index (RTSI), built from four refusal-template drift features and calibrated on this matrix -- routes 10/10 hidden- or near-hidden-danger rows to direct safety testing (Wilson 95% CI lower bound 0.72) while leaving 23 of 45 non-baseline rows in a low-risk bucket under both in-sample scoring and row-level leave-one-out validation; on the same matrix, the best single-feature baselines (unique-prefix-rate-delta, raw refusal-rate delta) recover 9/10 and 8/10 respectively at matched bucket size, and cross-stack transfer requires recalibration. For the quantized checkpoints, model families, and safety outcomes studied here, retained quality cannot waive direct safety evaluation.

As deep learning models scale, managing, inspecting, and modifying large checkpoints has become increasingly challenging. Researchers often need to alter model weights for layer restructuring, precision casting, low-rank factorization, and architectural debugging, yet these workflows often rely on fragile ad-hoc Python scripts. Here, we introduce BrainSurgery, a tool for robust and reproducible "tensor surgery" on neural network checkpoints, and provide a system demonstration covering four examples and three case studies from model upcycling to LoRA extraction. By abstracting storage formats and memory management, BrainSurgery executes complex transformations through declarative YAML plans. It supports structural modifications, mathematical transformations, and tensor reshaping through expressive regex and structural targeting, while built-in assertions validate tensor shapes, data types, and values to prevent silent errors. We envision that BrainSurgery will provide a strong foundation for future research through its reproducible and validated operations.

The standard LLM training pipeline applies reinforcement learning (RL) only after pre-training and supervised fine-tuning (SFT). We question this status quo by training a LLM from scratch and applying RL, SFT, and SFT followed by RL directly to intermediate pre-training checkpoints. We find that RL is effective very early, and often matches the full SFT$\to$RL pipeline early as well. Through experiments on harder problems, we find that targeted pre-training data composition is a strong lever for RL effectiveness, even more so than model scale. Beyond reasoning accuracy, applying RL directly to base checkpoints expands the model's distribution; the sharpening effect reported in recent work arises only when RL follows SFT. The general capabilities of the model remain essentially unchanged by RL, while they degrade following SFT. Finally, we merge RL and SFT objectives by parallel averaging, which outperforms across all other training methods discussed, across metrics, while preserving general capabilities. Together, these results suggest that LLM training might benefit from an expanded use of RL.

Smart contract decompilation aims to recover high-level source code from bytecode, but evaluating decompilers remains difficult because existing studies use narrow datasets, inconsistent metrics, and limited semantic consistency checks. This gap is increasingly important as large language models (LLMs) begin to generate source-like Solidity that may compile and appear plausible, even when its semantics diverge from the original contract. We introduce SCDBench, a dataset and benchmark methodology for LLM-based smart contract decompilation. The dataset contains 600 real-world Solidity contracts with paired bytecode inputs, ground-truth source code, and replayable semantic checkpoints. SCDBench evaluates decompiler outputs through four cumulative stages: format completeness, compilability, Application Binary Interface (ABI) recovery, and semantic consistency via differential replay. We evaluate Claude Opus 4.7, GPT-5.3-Codex, and GLM-5 in a zero-shot decompilation setting, including GLM-5 variants with and without extended reasoning and a zero-shot compilation-repair setting. The results show that frontier LLMs can often produce structured and compilable Solidity, but achieving semantic consistency remains far from solved: the best-performing frontier model perfectly decompiles only 42/600 contracts. We further show that introducing same-model compilation repair substantially improves performance at modest additional cost. SCDBench establishes a common ground for rigorous, reproducible evaluation and aims to accelerate the development of reliable smart contract decompilers for blockchain security and transparency.

Equivariant neural networks encode geometric symmetries by construction, yet they are often difficult to optimize and can underperform less constrained architectures. A growing body of work addresses this through architectural modifications such as constraint relaxation or approximate equivariance, while the role of the optimizer remains comparatively underexplored. We study this direction by comparing Muon and Adam across several equivariant and geometric architectures under pointcloud and molecular learning settings. On ModelNet40, where the comparison is clearest, Muon consistently improves over Adam across all architectures considered. We then analyze the trained ModelNet40 checkpoints through Hessian estimates, loss surface visualizations, and spectral properties of learned weights and intermediate representations. The checkpoints reached by Muon have larger Hessian curvature summaries but more regular loss surfaces, and their learned weights and representations have higher stable and effective ranks. These observations suggest that the interaction between optimizer design and geometric inductive bias deserves further attention from the community.

Neural network checkpoints have quietly become a large-scale data resource: millions of trained weight vectors now exist, each encoding task-, domain-, and architecture-specific knowledge. This position paper argues that model checkpoints should be treated as a first-class data modality, and that generative modeling in weight space should be standardized as a core machine learning primitive. Recent advances demonstrate that neural weights can be synthesized on demand, often matching fine-tuning performance while reducing adaptation cost by orders of magnitude. We contend that these results reflect an underlying structural fact: high-performing models occupy low-dimensional, highly structured regions of weight space shaped by symmetry, flatness, modularity, and shared subspaces. Building on this view, we organize existing methods into a five-stage pipeline, survey applications where the approach is already practical, and clarify current limits: adapter-scale and conditional generation are advancing rapidly, while unrestricted frontier-scale checkpoint synthesis remains open. Our goal is to shift the community's default mindset from optimizing models per task to sampling models from learned weight distributions, accelerating toward an era in which AI systems routinely improve or create other AI systems.

Recent work has fine-tuned language models on chess data and reported high benchmark scores as evidence that the resulting models can understand the rules of chess, play full chess games at a professional level, or generate human-readable explanations grounded in expert knowledge. We train KinGPT, a 25M-parameter character-level language model trained only on (position, best-move) pairs, who exceeds 3B-parameter ChessGPT on a 600-puzzle mate-in-N suite and 4B-parameter C1-4B over a 20-theme puzzle benchmark. We examine several claims made in existing literature regarding chess-trained language models and assert that their impressive benchmark performance is largely explained by pattern-matching. We also demonstrate how LLM-Modulo, a verifier-in-the-loop framework, raises RedPajama 3B's best move accuracy from 1.2% to 21.2% and move generation validity from 19.3% to 95.3% on mate-in-N chess puzzles, comparable to gains achieved from ChessGPT's fine-tuning on chess-specific web corpora at a fraction of the cost. Our results illustrate how pairing a general LLM with an external verifier offers a more flexible alternative to directly training on synthetic data for well-defined domains. We open source all training/evaluation code, datasets, puzzle samples, and KinGPT model checkpoints for reproducibility.

Conventional multi-stage cell tracking approaches rely heavily on detection or segmentation in each frame as a prerequisite, requiring substantial resources for high-quality segmentation masks and increasing the overall prediction time. To address these limitations, we propose CAP, a novel end-to-end one-stage framework that reimagines cell tracking by treating Cell as Point. Unlike traditional methods, CAP eliminates the need for explicit detection or segmentation, instead jointly tracking cells for sequences in one stage by leveraging the inherent correlations among their trajectories. This simplification reduces both labeling requirements and pipeline complexity. However, directly processing the entire sequence in one stage poses challenges related to data imbalance in capturing cell division events and long sequence inference. To solve these challenges, CAP introduces two key innovations: (1) adaptive event-guided (AEG) sampling, which prioritizes cell division events to mitigate the occurrence imbalance of cell events, and (2) the rolling-as-window (RAW) inference strategy, which ensures continuous and stable tracking of newly emerging cells over extended sequences. By removing the dependency on segmentation-based preprocessing while addressing the challenges of imbalanced occurrence of cell events and long-sequence tracking, CAP demonstrates promising cell tracking performance and is 8 to 32 times more efficient than existing methods. The code and model checkpoints are available at https://github.com/YXSong000/CAP.