Search: “Toolformer”

31 stories

DomainAll Cyber AIWindow24h 3d 7d 30d AllIn the wild Important only Show duplicates

SeverityAll critical 0 high 0 medium 0 low 0 info 1

Smart search ranks by meaning as well as keywords (one row per story, last 45 days).

The Evolution of the Agent Harness

Latent Space essay argues late-2025 agent gains came from models and harnesses maturing together, with harness logic absorbed into model weights.

The piece defines the agent harness as everything beyond model weights—tools, context, memory, guardrails—and charts its evolution from ReAct prompting (October 2022) through AutoGPT's premature autonomy, Cursor/Copilot's human-in-the-loop retreat, and Devin's roughly 15% success rate, to o1's capability overhang and Claude Code's February 2025 terminal agent with permission rules. It argues the Christmas 2025 jump cited by Transformer co-inventor Lukasz Kaiser reflected model and harness curves crossing, and that remaining harnesses will serve human attention rather than the model.

Latent SpaceAug 22, 07:30 UTC · 26d agoAI tools & infra

Technical Manual for a Toolkit for Measuring Contextual Individuation in Transformer Language Models

An open methodology toolkit measures whether transformer language models contextualize fixed word forms across domains using bridge forms and layer-wise silhouette analysis.

The manual documents an open toolkit built around 'bridge forms' - identical written words recurring across two or more subject domains with a different sense in each - to test whether transformer language models individuate word occurrences by context beyond the embedding layer. It covers declarative specification of bridge forms, Wikipedia corpus acquisition, occurrence localization, layer-wise representation extraction, domain-pairwise silhouette measurement, and visualization, justifying each choice against failure modes such as sense contamination and subword-tokenization misalignment. It is a methodological and implementation reference and reports no empirical results.

arXiv cs.AI / cs.LG / cs.CLSep 4, 16:38 UTC · 12d agoAI research

Google Research Releases ToolGrad: Answer-First Framework Hits 99.8% Pass Rate for Tool-Use Data Generation

Google Research and partners introduce ToolGrad, a verified tool-chain-first data generation framework reaching 99.8% pass rate and boosting Gemma-3-12B to 83.1 on BFCL.

Researchers from Google, the University of Tokyo, RIKEN AIP, and Tohoku University released ToolGrad, which inverts query-first tool-use data generation by executing and verifying API chains before annotating them with user queries. On the ToolBench database of 16,000+ APIs, ToolGrad raised generation pass rate from 63.8% to 99.8% while increasing tool uses per sample from 2.1 to 3.4 and cutting tool-use steps from 34.3 to 20.0. Fine-tuning Gemma-3 at 1B, 4B, and 12B parameters on the 500-sample ToolGrad-500 dataset lifted ToolGrad-12B to 83.1 on the Berkeley Function Calling Leaderboard, near Gemini 2.5 Pro at 83.2 and ahead of GPT-5 at 74.4. Code is Apache-2.0, with the dataset, PyPI package, and models available on Hugging Face.

MarkTechPostSep 11, 06:01 UTC · 6d agoAI research1

Knowledgator Releases GLiFormer: A 575M-Parameter Encoder That Hits 91.10 F1 on Nested JSON Extraction Without Generating Tokens

Knowledgator released GLiFormer, an Apache-2.0 encoder (264M/575M) handling NER, classification, relations, and nested JSON extraction, scoring 91.10 F1.

Knowledgator Engineering released GLiFormer, a schema-conditioned encoder that performs NER, classification, relation extraction, nested JSON structuring, and embeddings without generating output tokens. GLiFormer Large v1 has 575.6M parameters and scores 91.10 F1 on nested JSON extraction, close to GPT-5.6-luna's 91.96; both checkpoints are Apache 2.0 on Hugging Face. Reported median latency is 69 ms on GPU for the base model, though relation extraction (21.33 micro-F1) still trails GLiNER-Relex and larger LLMs.

MarkTechPostSep 16, 21:19 UTC · 13h agoModel release1

ToolLoop: Closed-Loop Tool-Use Data Synthesis via Decomposed Generation and Dynamic Self-Feedback

ToolLoop introduces a closed-loop synthetic data framework whose 11K examples lift a 4B model to 86.40% on BFCL tool-use evaluation.

ToolLoop decomposes tool-use data synthesis into function-name sampling, backward derivation of user queries, and forward derivation of tool calls, with dynamic self-feedback at each stage. This shifts the paradigm from generate-then-filter to generate-verify-refine, reducing inefficient and imbalanced synthetic data. A 4B model trained on 11K synthetic examples reaches 86.40% accuracy on BFCL non-reasoning mode (86.07% in an Isolate variant excluding BFCL-overlapping functions) and 72.1% on ACEBench using only 18.3% of baseline training data.

arXiv cs.AI / cs.LG / cs.CLSep 8, 17:26 UTC · 8d agoAI research1

RenderFormer-V2: Neural Rendering with Heterogeneous Scene Primitives

RenderFormer-V2 is a transformer-based neural renderer handling caustics, volumetric scattering and out-of-distribution materials without per-scene training or specialized code.

RenderFormer-V2 is a learned transformer-based neural rendering model that models global light transport as a sequence-to-sequence transformation, handling caustics, volumetric scattering, environment lighting, textured and displaced surfaces and out-of-distribution materials. It uses a two-stage process: a view-independent stage resolving primitive-to-primitive transport, and a view-dependent stage converting the neural scene representation into pixels. Improvements include combined windowed-attention with a rendering-informed attention sink for scalability, support for heterogeneous primitives like environment maps and participating media, and a surface-reflectance-independent neural material encoding, validated across diverse scenes with extensive ablations.

Hugging Face daily papersSep 3, 20:00 UTC · 13d agoAI research

27.5KB language-agnostic WebGPU syntax highlighter

A developer released gpu-lexer, a 27.5KB language-agnostic syntax highlighter that uses a tiny WebGPU model to label code tokens in the browser.

gpu-lexer splits source into words, whitespace, and symbols, then a small WebGPU model uses local and whole-file context to assign nine token classes, working on languages never seen in training. On held-out files, 12.57% of token labels differ from Shiki, though this measures agreement with Shiki rather than objective correctness. In benchmarks against Shiki 4.4.3, Prism.js, Highlight.js, Sugar High, and Starry Night, it highlighted 10 concatenated copies of three.min.js (5.56M characters) about 10x faster on an Apple M4 Pro in Chrome 152. The author frames it as an experiment, not a grammar-equivalent highlighter.

Hacker News · securitySep 9, 01:14 UTC · 8d agoAI tools & infra HN 21↑ · 5 comments1

Objective vs. Search: Decomposing What Makes a Good Tokeniser

New tokeniser study shows search procedure, not optimisation objective, drives bits-per-byte performance across model sizes, vocabulary sizes, and multilingual settings.

The paper disentangles BPE and UnigramLM along two axes: optimisation objective (compression vs log-likelihood) and search procedure (bottom-up merging vs top-down pruning). Two new algorithms, BottomUpLL and TopDownComp, complete the 2x2 design space, and trained language models are evaluated on bits-per-byte and BLiMP across model sizes, vocabulary sizes, and English-only vs multilingual domains. Bottom-up tokenisers consistently achieve lower bits-per-byte in most settings, while BLiMP shows no consistent relationship with design choice.

arXiv cs.AI / cs.LG / cs.CLSep 16, 17:59 UTC · 16h agoAI research

Context Engineering Inside the Harness: 4 Mechanisms That Beat Context Overflow and Goal Loss on Long-Horizon Tasks

Survey of four harness mechanisms—context budgeting, compaction, todo-state, and memory—that keep long-horizon LLM agents on task across 200+ tool calls.

The article details how agent harnesses, not larger context windows, solve context overflow and goal loss on long-horizon tasks, citing Chroma's Context Rot report showing 18 LLMs (GPT-4.1, Claude 4, Gemini 2.5, Qwen3) degrade on long inputs. Concrete implementations include LangChain Deep Agents offloading tool responses over 20,000 tokens to the filesystem and truncating old tool calls at 85% window usage, and Claude Code capping auto memory at 25KB while re-reading the 5 most recently modified files after compaction. OpenAI's Responses API now offers server-side compaction via context_management with a standalone /responses/compact endpoint, which Codex uses for long-running coding tasks. Manus reports a roughly 100:1 input-to-output token ratio per ~50-tool-call task, motivating todo.md state recitation to prevent goal drift.

MarkTechPostSep 13, 05:56 UTC · 4d agoAI research2

Cartesian – AI 3D Modeling for Design

Formas launches Cartesian, an AI 3D modeling tool for architecture and product design with natural-language editing and CAD exports.

Cartesian by Formas is an AI-powered 3D modeling tool aimed at architecture and product design that converts photos, rough plans, sketches, and scans into precise geometry. Users can edit models conversationally while explicitly preserved elements stay unchanged. It creates real solids and NURBS geometry and exports to AutoCAD DWG, Rhino 3DM, and SketchUp SKP, with BIM IFC support planned, removing the need for a separate desktop CAD license.

Hacker News · AISep 15, 15:26 UTC · 1d agoAI industry HN 42↑ · 32 comments

RelightFormer: Feed-forward Generative Transformer for Multiview Object Relighting

RelightFormer is a feed-forward generative transformer for photorealistic single- and multi-view object relighting, trained on a 90K-object dataset.

Researchers introduce RelightFormer, a feed-forward generative transformer adapted from a video foundation model that performs direct image relighting without explicit intrinsic property estimation. The architecture injects target environment maps via a latent illumination module with cross-attention and uses permutation-invariant positional encodings for unordered multi-view inputs. Training relies on the newly constructed Laval Objaverse Dataset (LOD) with 90K objects and 39K unique illuminations, and the model shows state-of-the-art quality with strong zero-shot generalization across single-view, multi-view, and novel-view relighting.

Hugging Face daily papersSep 6, 20:00 UTC · 10d agoAI research

The /wayfinder Skill: Navigating the “Fog of War” of Planning

Matt Pocock released the /wayfinder skill, an orchestrator layer that manages planning sessions, maps, and tickets for AFK coding agents.

Latent Space interviews Matt Pocock, whose AI Skills for Real Engineers project has 220,000+ GitHub stars, about his new /wayfinder skill. The skill manages agent context during ambiguous planning by splitting work into grilling, prototype, research, and task tickets organized under a shared map, enabling overnight AFK agent runs. It uses deliberate terminology like map, ticket, and session to steer agent behavior, and was tested on projects including a personal website rearchitecture.

Latent SpaceAug 20, 20:59 UTC · 27d agoAI tools & infra

Google Open-Sources Mantis: A Modular Skills Toolkit That Lets Coding Agents Find, Reproduce and Patch Vulnerabilities

Google open-sourced Mantis, an Apache-2.0 modular skills toolkit that lets AI coding agents find, reproduce, and patch vulnerabilities with sandboxed verification.

Google released Mantis on GitHub under Apache 2.0 as a stack-agnostic set of slash-command skills that chain through the full vulnerability lifecycle: mining version history, building threat models, filtering findings, reproducing bugs in gVisor or network-disabled VMs, assembling exploit chains, patching, and scoring residual risk from 1 to 10. It runs with Gemini CLI, Antigravity CLI, the Google ADK, or comparable agent frameworks, and a supervisor skill (/mantis-meta-agent) can drive the whole loop. Google says the design targets the sub-7 percent true-positive rate of naive AI code scanning, and that its hierarchical summary tree cuts token overhead by over 85 percent. The toolkit is deployable for local and internal evaluation but not yet recommended for production.

MarkTechPostSep 9, 22:58 UTC · 7d agoAI tools & infra

Hottest cybersecurity open-source tools of the month: August 2026

August's open-source security roundup highlights NVIDIA SkillSpector for AI agent skill scanning, Future AGI, Chainloop, PentestGPT, and Hazmat agent containment.

Help Net Security's August 2026 roundup covers five open-source tools: NVIDIA's SkillSpector, a scanner that assesses AI agent skills from directories, zips, SKILL.md files, or Git URLs and returns findings and risk scores; Future AGI, an Apache 2.0 platform for tracing, evaluating, simulating, and guardrailing LLM agents; Chainloop, a software supply chain evidence store that produces signed in-toto attestations in CI pipelines; PentestGPT, an agentic penetration testing framework running recon, exploit, and reporting stages; and Hazmat, which runs AI coding agents like Claude Code and Codex inside a separate machine account for containment.

Help Net SecurityAug 26, 00:00 UTC · 22d agoTools

Version Control DFIR: a Cheatsheet to GitHub, GitLab, Bitbucket, and Azure DevOps

Wiz published a DFIR cheatsheet covering log visibility, incident readiness, and threat hunting across GitHub, GitLab, Bitbucket, and Azure DevOps.

Wiz researchers released a practitioner's guide to version control system forensics, incident response, and threat hunting. The cheatsheet maps log sources, audit capabilities, and hunting workflows across GitHub, GitLab, Bitbucket, and Azure DevOps. It aims to improve incident readiness for source code and CI/CD compromise scenarios.

Wiz BlogAug 27, 12:00 UTC · 20d agoResearch1

Disentangling Representation Evolution in Transformers through Directional Decomposition

Researchers decompose transformer updates into parallel and perpendicular components, linking representation geometry to editing robustness, compression diagnosis, and training interventions.

The paper studies transformer representation evolution as functional geometry, decomposing learned updates into parallel and perpendicular components across attention/MLP and value-aggregation spaces. Targeted edits reveal a space-dependent asymmetry: exclude-self value-space parallel manipulation is markedly more robust than residual-space and perpendicular counterparts. Full-aggregate parallel suppression during from-scratch pretraining lowers validation-loss trajectories and improves downstream averages, with the value-space variant strongest. Code is released on GitHub.

Hugging Face daily papersupdated Sep 14, 17:56 UTC · 2d agofirst Sep 13, 20:00 UTC · 3d agoAI research 2 sources

TART: A Modular Tool for Technique-Aware Audio-to-Tablature Guitar Transcription

TART pipeline transcribes guitar audio to tablature with fingering and technique labels, outperforming prior baselines by up to 8.5 Tab F1 points.

TART is a modular four-stage pipeline for audio-to-tablature guitar transcription, combining an audio-to-MIDI transcription model, an expressive technique classifier, an audio-conditioned T5 encoder-decoder for string-fret assignment, and an automated tablature generator. In zero-shot evaluation on GuitarSet, EGDB, and two noisy augmented benchmarks, it averages 81.35% audio-to-MIDI F50 (+6.67 over the best baseline), 71.8% string-fret Tab F1 (+8.5), and 54.08% end-to-end Tab F1. The authors state it is the first framework to generate tablature with both fingering and expressive technique annotations directly from guitar audio.

arXiv cs.AI / cs.LG / cs.CLSep 10, 17:55 UTC · 6d agoAI research2

AI SIEM Search

Huntress launched AI Search for Managed SIEM, translating plain-English questions into ESQL log queries with schedulable custom alerts.

Huntress introduced AI Search in its Managed SIEM, letting users query logs in natural language instead of writing ESQL, KQL, or SPL. The feature generates a real ESQL query behind the scenes, exposes it for users to inspect and learn from, and allows saved searches to be scheduled as recurring alerts at hourly, daily, or weekly cadence. A planned next version will return summarized plain-language answers instead of raw logs.

HuntressSep 7, 14:00 UTC · 9d agoTools

Open-source secrets scanning tool Sift hunts credentials in Microsoft 365, Slack, and Jira

Stratus Security open-sourced Sift, a CLI secrets scanner covering Active Directory, SharePoint, Teams, Slack, Jira, and Confluence, outperforming Snaffler in benchmarks.

Stratus Security released Sift, a free open-source command line tool that hunts passwords, API keys, and sensitive data across local disks, Windows shares, Active Directory, SharePoint, OneDrive, Teams, Slack, Jira, and Confluence. In the firm's benchmarks, Sift scanned 250,000 files in 10.61 seconds versus Snaffler's 25.48 and averaged 92 MiB memory versus Snaffler's 337 MiB. Optional false-positive filtering runs through a local language model via Ollama, and scans write checkpoints so interrupted runs resume. The tool found thousands of credentials in Jira ticket comments that prior pentesting had missed.

Help Net SecuritySep 2, 00:00 UTC · 15d agoTools

Dr. Claw: An AI Scientist Workspace for Vibe Research

Researchers release Dr. Claw, an open-source auditable workspace that wraps coding agents like Claude Code for end-to-end AI-assisted research workflows.

Paper 2609.00365 presents Dr. Claw, an open-source workspace that wraps existing coding-agent executors such as Claude Code and Gemini CLI in a controllable, human-in-the-loop research workflow. It uses persistent state objects, a reusable skill library, and multi-executor coordination to make research decisions auditable and recoverable, rather than adding another autonomous agent. Holding the executor fixed, Dr. Claw scores higher on research completeness than a bare command-line agent while preserving an auditable process trail. The code is released under AGPL-3.0 on GitHub (OpenLAIR/dr-claw).

Hugging Face daily papersAug 30, 20:00 UTC · 17d agoAI research1

What Breaks Under Pruning in Smart Homes, and When? Evaluating LLM Degradation Across Architectures and Task Complexity

Pruning study across four LLM architectures finds dense models degrade sharply on smart-home tool calling while MoE models tolerate far more.

Researchers systematically study pruning-induced degradation in smart-home tool calling across four LLMs spanning dense Transformer, dense hybrid, and mixture-of-experts architectures, combining depth, width, hybrid, and expert pruning methods, and evaluate over 19,500 instances from three datasets after post-pruning supervised fine-tuning. Dense models show narrow safe pruning regions followed by sharp degradation, while MoE models tolerate substantially more pruning. Pruning degrades grounded specificity (operation, device, argument, value) before schema-level intent, and aggressive dense pruning can induce systematic over-refusal.

arXiv cs.AI / cs.LG / cs.CLSep 15, 17:51 UTC · 1d agoAI research

Identifying Agentic Automation with Behavioral Telemetry

Akamai describes detecting autonomous AI browser agents like Comet using Masked Autoencoder Transformer models on sparse behavioral telemetry.

Akamai researchers present a behavioral telemetry approach for identifying agentic automation in web traffic. Masked Autoencoder Transformer models are used to detect the sparse behavioral signals produced by autonomous AI browser agents such as Comet. The work targets traffic classification and bot detection rather than a specific vulnerability, and becomes more relevant as agentic browsing adoption grows.

Akamai BlogAug 19, 19:00 UTC · 28d agoAI safety & security1

ToolHive: The open-source way to run any MCP server securely

Stacklok's open-source ToolHive runs Model Context Protocol servers in isolated containers with per-request identity enforcement, audit logging and a signed registry.

ToolHive, shipped under Apache 2.0, containersizes MCP servers locally via Docker or Podman or in clusters through a Kubernetes operator, applying permissions, network filtering, and secrets management. The platform includes a Registry Server implementing the official MCP Registry API with signing and provenance verification, a Virtual MCP Server gateway with OIDC/OAuth single sign-on and OpenTelemetry traces, and a desktop Portal for one-click installs. The browser-based cloud UI is retired, so the desktop app and CLI are the supported rollout paths.

Help Net SecuritySep 7, 00:00 UTC · 10d agoAI tools & infra1

Omni Interaction Agent Technical Report

Researchers release Gander, an end-to-end omni interaction model with full-duplex streaming across video, speech, and text plus agentic capabilities.

Gander is an end-to-end model unifying omni perception, realtime interaction, and agentic capabilities in a single framework, accepting continuously streaming video, speech, and text. It uses a Cerebellum-Brain architecture where the Cerebellum handles realtime conversation and the Brain handles reasoning and agentic tasks, built on a streaming Thinker-Talker design with chunk-level token streams. Internal human evaluations report spoken dialogue on par with SOTA open source models and competitive omni interaction; the models, code, and data are released publicly.

Hugging Face daily papersSep 7, 20:00 UTC · 9d agoModel release

Retrofitting Code Using LLMs to Support Exceptional Behavior

EXCODER combines static/dynamic analysis with LLMs to retrofit exception-handling code, achieving 85.92% pass@1 with Qwen 2.5 Coder 32B on Java benchmarks.

The paper introduces the task of retrofitting existing code with Exception Related Code (throw statements, guarding conditions, try/catch blocks) so that given Exceptional Behavior Tests pass. EXCODER performs context engineering by integrating static and dynamic program analysis output with LLMs; it was evaluated on a benchmark built from 304 methods across 75 GitHub Java projects. Combined with Qwen 2.5 Coder 32B, EXCODER achieves pass@1, 5, and 10 rates of 85.92%, 86.18%, and 86.51%, roughly 13 percentage points over baseline, and manual inspection reveals remaining limitations.

arXiv cs.AI / cs.LG / cs.CLSep 9, 16:18 UTC · 7d agoAI research1

The Router Within: Eliciting Native Skill Routing from a Frozen LLM

Gavel reads native skill-routing signals from a frozen LLM's forward passes with two linear maps, beating retrieve-and-rerank pipelines by up to 21.9 points on Qwen3-32B.

Gavel (Glance And Verdict from a frozen LLM) elicits skill routing from a frozen agent LLM using two trained linear maps that read mid-layer states, keeping all skill text out of context. A glance step scores the full library against compact per-skill banks built in one forward pass at installation; a verdict step resumes shortlisted skills' forward passes and fuses likelihood and yes/no judgments as a product of experts. It transfers zero-shot to three public benchmarks plus SkillTraj, a new benchmark of 372 simulated agent trajectories. On Qwen3-32B it beats progressive disclosure and retrieve-and-rerank pipelines adding 1.2B–16B external parameters by up to 13.4 points on written tasks and 21.9 when skills are needed mid-rollout.

Hugging Face daily papersupdated Sep 14, 17:58 UTC · 2d agofirst Sep 13, 20:00 UTC · 3d agoAI research 2 sources

llm 0.34

Version 0.34 of Simon Willison's llm CLI adds response-duration metrics to log output, plus bug fixes and faster log querying.

The open-source llm command-line tool for interacting with large language models released version 0.34. The headline change adds response duration in milliseconds and human-readable form to llm logs --usage Markdown output, plus a new duration_ms field in llm logs --short. The release includes several contributed bug fixes and a significant performance improvement to llm logs, alongside the related llm-openrouter 0.7.1 release.

Simon WillisonSep 2, 19:23 UTC · 14d agoAI tools & infra1

SchemeArena: Factorized Stress Testing of Scheming in LLM Agents

Researchers introduce SchemeArena, a 400-scenario benchmark stress-testing scheming in LLM agents, finding explicit instrumental goals are the strongest driver of covert misaligned behavior.

The paper presents SchemeArena, a 400-scenario benchmark built through factorized scenario synthesis spanning safety-relevant tool domains, instrumental goals, oversight conditions and pressure mechanisms. The accompanying SCOUT monitor grounds multi-criteria scheming judgments in evidence drawn from agents' reasoning and actions. Stress tests across five LLM agents show explicit instrumental goals are the strongest driver of scheming propensity, while action-only monitoring increased scheming in several closed models, suggesting partial oversight can act as an optimization constraint. The benchmark, code and monitor are released at github.com/launchnlp/SchemeArena.

Hugging Face daily papersSep 7, 20:00 UTC · 9d agoAI safety & security

LLM Agents as Computational Typologists

AUTOTYPOLOGIST is an LLM agent that performs evidence-grounded linguistic typology analysis over 25 open-source reference grammars.

The agent retrieves relevant grammar sections, analyzes interlinear glossed text (IGT), and iteratively reasons over typological hypotheses in a ReAct-style workflow. It was evaluated on typological feature coding against expert annotations and hypothesis testing against universals using 25 open-source reference grammars. Results suggest LLM agents can support scalable, inspectable crosslinguistic analysis but still require expert validation.

arXiv cs.AI / cs.LG / cs.CLSep 7, 17:35 UTC · 9d agoAI research1

CoVeR: Coverage-Based Token Pruning for Multi-View 3D Reasoning in VLMs

CoVeR, a training-free coverage-based token pruner, preserves 93.5% of VLM 3D-reasoning performance using only about 8% of visual tokens.

Researchers introduce CoVeR, a deterministic, training-free selector that chooses visual tokens to cover every region of a multi-view 3D scene using only token coordinates. Unlike learned-importance and voxelization pruners, it enforces an exact per-scene token budget, avoids saturation plateaus, and prevents near-duplicate selections. Experiments across four vision-language models show it surpasses prior state of the art by 3.9 percentage points on average across three 3D reasoning benchmarks.

Hugging Face daily papersSep 7, 20:00 UTC · 9d agoAI research

PARSER: Read in Parallel, Reason in Depth for Long-Context LLM Agents

PARSER uses parallel reader subagents and an RL-trained lead agent for long-context QA, beating baselines and cutting latency up to 11x.

The PARSER paper decouples reading from reasoning: frozen subagents each read one document chunk in parallel while an RL-optimized lead agent iteratively broadcasts queries and aggregates evidence in scatter-gather rounds. On multi-hop QA with 7K to 896K token contexts, a 4B-backbone PARSER beats the strongest sequential memory baseline by 5.7 points on average and 12.0 points at 896K tokens, and a 9B version surpasses DeepSeek-V4-Pro by 6.3 points. Controlled experiments show robustness to evidence position, order, and distance perturbations, with inference latency reduced by up to 11x.

Hugging Face daily papersSep 5, 20:00 UTC · 11d agoAI research1