FlashMorph formulates hybrid layer selection as budget-constrained optimization, trains per-layer gates on synthetic retrieval data with linearization regularization, then discretizes and distills to produce efficient hybrid architectures.
hub Canonical reference
InternLM2 Technical Report
Canonical reference. 77% of citing Pith papers cite this work as background.
50
Pith papers citing it
27
external citations · Pith
Background
77% of classified citations
abstract
The evolution of Large Language Models (LLMs) like ChatGPT and GPT-4 has sparked discussions on the advent of Artificial General Intelligence (AGI). However, replicating such advancements in open-source models has been challenging. This paper introduces InternLM2, an open-source LLM that outperforms its predecessors in comprehensive evaluations across 6 dimensions and 30 benchmarks, long-context modeling, and open-ended subjective evaluations through innovative pre-training and optimization techniques. The pre-training process of InternLM2 is meticulously detailed, highlighting the preparation of diverse data types including text, code, and long-context data. InternLM2 efficiently captures long-term dependencies, initially trained on 4k tokens before advancing to 32k tokens in pre-training and fine-tuning stages, exhibiting remarkable performance on the 200k ``Needle-in-a-Haystack" test. InternLM2 is further aligned using Supervised Fine-Tuning (SFT) and a novel Conditional Online Reinforcement Learning from Human Feedback (COOL RLHF) strategy that addresses conflicting human preferences and reward hacking. By releasing InternLM2 models in different training stages and model sizes, we provide the community with insights into the model's evolution.
hub tools
citation-role summary
background 11
baseline 1
method 1
citation-polarity summary
representative citing papers
Representational convergence across 16 LLMs on 800 reasoning problems is stronger for failed tasks and pre-decision stages but shows minimal causal influence on predictions, pointing to shared processing constraints over shared reasoning.
Language representations serve as the asymptotic attractor for convergence in independently trained multimodal neural networks due to feature density asymmetry.
VTAgent uses a question-guided agent to anchor keyframes for evidence-aware Video TextVQA, delivering up to +12 accuracy and new SOTA results via training-free operation plus SFT and RL.
StoryReward, trained on a new 100k story preference dataset, sets state-of-the-art performance on the introduced StoryRMB benchmark for aligning LLM stories with human preferences.
CompliBench uses simulation and adversarial flaw injection to create labeled dialogue data showing that top proprietary LLMs perform poorly at spotting guideline violations while fine-tuned smaller models outperform them and generalize to new domains.
TaxPraBen is a new benchmark with 14 datasets and a structured evaluation method for measuring LLM performance on Chinese real-world tax tasks and scenarios.
Visual-ERM is a new multimodal reward model that supplies fine-grained visual feedback for training vision-language models on chart-to-code, table, and SVG tasks, yielding measurable gains over prior rewards.
UniGeoSeg releases the first million-scale dataset for instruction-driven remote sensing segmentation and a unified model that achieves state-of-the-art results with strong zero-shot generalization.
TOPS formulates visual token pruning as constructing Token Optimal Preservation Sets using three information-theoretic principles and demonstrates superior performance on MLLM benchmarks.
Position bias in on-policy distillation degrades later-token supervision; IW-OPD weights tokens by accumulated discrepancy, yielding faster convergence and up to 6.9 point gains on AIME-2025.
EventDrive supplies a multi-task benchmark and EventDrive-VLM architecture that fuses event data, RGB, and language supervision, reporting gains in temporal precision and motion awareness for driving intelligence.
LyraV uses FDTC and SToP for per-frame incremental decoding to reach 98.29% video synchrony at 3.89 FPS while preserving general understanding.
Introduces SRTD and SRTD-lite to symmetrize topological divergences for neural representations and NTS as a rank-correlation-based metric bounded in [-1,1] for cross-scenario benchmarking.
HARVE removes the component of the reward-head vector aligned with a multi-directional hacking subspace from residual streams using a small set of contrastive examples, improving robustness on RewardHackBench across eight models without fine-tuning while preserving general capability.
Smaller models provide temporally correlated policy-level diversity that serves as structured exploration for training larger models in GRPO, yielding accuracy gains such as +8.8% on AIME 24 with reduced compute via the S2L-PO framework.
MegaScale-Omni delivers 1.27x-7.57x higher throughput for dynamic multimodal LLM training by decoupling encoder and LLM parallelism, using unified colocation, and applying adaptive workload balancing.
Decompose and Recompose decomposes seen robotic demonstrations into skill-action alignments and recomposes them via visual-semantic retrieval and planning to enable zero-shot cross-task generalization.
Delta-LLaVA adds Change-Enhanced Attention, Change-SEG with prior embeddings, and Local Causal Attention to MLLMs to overcome temporal blindness, outperforming general models on a new unified benchmark for bi- and tri-temporal remote sensing tasks.
BadSkill poisons embedded models in agent skills to achieve up to 99.5% attack success rate on triggered tasks with only 3% poison rate while preserving normal behavior on non-trigger inputs.
HAWK is a training-free method that prunes over 80% of visual tokens in MLLMs while retaining 96% accuracy by using head importance weights and text-guided attention to select task-relevant tokens.
A new personalization benchmark shows reward models are only ~76% accurate at identifying user-specific preference, and its scores track downstream BoN/PPO quality better than prior benchmarks.
A minimal agentic system achieves competitive performance in automated theorem proving with a simpler design and lower cost than state-of-the-art methods.
Generalized on-policy distillation with reward scaling above one (ExOPD) lets student models surpass teacher performance when merging domain experts on math and code tasks.
citing papers explorer
-
Morphing into Hybrid Attention Models
FlashMorph formulates hybrid layer selection as budget-constrained optimization, trains per-layer gates on synthetic retrieval data with linearization regularization, then discretizes and distills to produce efficient hybrid architectures.
-
Convergence Without Understanding: When Language Models Agree on Representations but Disagree on Reasoning
Representational convergence across 16 LLMs on 800 reasoning problems is stronger for failed tasks and pre-decision stages but shows minimal causal influence on predictions, pointing to shared processing constraints over shared reasoning.
-
The Wittgensteinian Representation Hypothesis: Is Language the Attractor of Multimodal Convergence?
Language representations serve as the asymptotic attractor for convergence in independently trained multimodal neural networks due to feature density asymmetry.
-
VTAgent: Agentic Keyframe Anchoring for Evidence-Aware Video TextVQA
VTAgent uses a question-guided agent to anchor keyframes for evidence-aware Video TextVQA, delivering up to +12 accuracy and new SOTA results via training-free operation plus SFT and RL.
-
StoryAlign: Evaluating and Training Reward Models for Story Generation
StoryReward, trained on a new 100k story preference dataset, sets state-of-the-art performance on the introduced StoryRMB benchmark for aligning LLM stories with human preferences.
-
CompliBench: Benchmarking LLM Judges for Compliance Violation Detection in Dialogue Systems
CompliBench uses simulation and adversarial flaw injection to create labeled dialogue data showing that top proprietary LLMs perform poorly at spotting guideline violations while fine-tuned smaller models outperform them and generalize to new domains.
-
TaxPraBen: A Scalable Benchmark for Structured Evaluation of LLMs in Chinese Real-World Tax Practice
TaxPraBen is a new benchmark with 14 datasets and a structured evaluation method for measuring LLM performance on Chinese real-world tax tasks and scenarios.
-
Visual-ERM: Reward Modeling for Visual Equivalence
Visual-ERM is a new multimodal reward model that supplies fine-grained visual feedback for training vision-language models on chart-to-code, table, and SVG tasks, yielding measurable gains over prior rewards.
-
UniGeoSeg: Towards Unified Open-World Segmentation for Geospatial Scenes
UniGeoSeg releases the first million-scale dataset for instruction-driven remote sensing segmentation and a unified model that achieves state-of-the-art results with strong zero-shot generalization.
-
TOPS: First-Principles Visual Token Pruning via Constructing Token Optimal Preservation Sets for Efficient MLLM Inference
TOPS formulates visual token pruning as constructing Token Optimal Preservation Sets using three information-theoretic principles and demonstrates superior performance on MLLM benchmarks.
-
On the Position Bias of On-Policy Distillation
Position bias in on-policy distillation degrades later-token supervision; IW-OPD weights tokens by accumulated discrepancy, yielding faster convergence and up to 6.9 point gains on AIME-2025.
-
EventDrive: Event Cameras for Vision-Language Driving Intelligence
EventDrive supplies a multi-task benchmark and EventDrive-VLM architecture that fuses event data, RGB, and language supervision, reporting gains in temporal precision and motion awareness for driving intelligence.
-
Don't Pause: Streaming Video-Language Synchrony for Online Video Understanding
LyraV uses FDTC and SToP for per-frame incremental decoding to reach 98.29% video synchrony at 3.89 FPS while preserving general understanding.
-
Symmetric Divergence and Normalized Similarity: A Unified Topological Framework for Representation Analysis
Introduces SRTD and SRTD-lite to symmetrize topological divergences for neural representations and NTS as a rank-correlation-based metric bounded in [-1,1] for cross-scenario benchmarking.
-
HARVE: Hacking-Aware Reward-Head Vector Editing for Robust Reward Models
HARVE removes the component of the reward-head vector aligned with a multi-directional hacking subspace from residual streams using a small set of contrastive examples, improving robustness on RewardHackBench across eight models without fine-tuning while preserving general capability.
-
Smaller Models are Natural Explorers for Policy-Level Diversity in GRPO
Smaller models provide temporally correlated policy-level diversity that serves as structured exploration for training larger models in GRPO, yielding accuracy gains such as +8.8% on AIME 24 with reduced compute via the S2L-PO framework.
-
MegaScale-Omni: A Hyper-Scale, Workload-Resilient System for MultiModal LLM Training in Production
MegaScale-Omni delivers 1.27x-7.57x higher throughput for dynamic multimodal LLM training by decoupling encoder and LLM parallelism, using unified colocation, and applying adaptive workload balancing.
-
Decompose and Recompose: Reasoning New Skills from Existing Abilities for Cross-Task Robotic Manipulation
Decompose and Recompose decomposes seen robotic demonstrations into skill-action alignments and recomposes them via visual-semantic retrieval and planning to enable zero-shot cross-task generalization.
-
Decoding the Delta: Unifying Remote Sensing Change Detection and Understanding with Multimodal Large Language Models
Delta-LLaVA adds Change-Enhanced Attention, Change-SEG with prior embeddings, and Local Causal Attention to MLLMs to overcome temporal blindness, outperforming general models on a new unified benchmark for bi- and tri-temporal remote sensing tasks.
-
BadSkill: Backdoor Attacks on Agent Skills via Model-in-Skill Poisoning
BadSkill poisons embedded models in agent skills to achieve up to 99.5% attack success rate on triggered tasks with only 3% poison rate while preserving normal behavior on non-trigger inputs.
-
HAWK: Head Importance-Aware Visual Token Pruning in Multimodal Models
HAWK is a training-free method that prunes over 80% of visual tokens in MLLMs while retaining 96% accuracy by using head importance weights and text-guided attention to select task-relevant tokens.
-
Personalized RewardBench: Evaluating Reward Models with Human Aligned Personalization
A new personalization benchmark shows reward models are only ~76% accurate at identifying user-specific preference, and its scores track downstream BoN/PPO quality better than prior benchmarks.
-
A Minimal Agent for Automated Theorem Proving
A minimal agentic system achieves competitive performance in automated theorem proving with a simpler design and lower cost than state-of-the-art methods.
-
Learning beyond Teacher: Generalized On-Policy Distillation with Reward Extrapolation
Generalized on-policy distillation with reward scaling above one (ExOPD) lets student models surpass teacher performance when merging domain experts on math and code tasks.
-
Fine-R1: Make Multi-modal LLMs Excel in Fine-Grained Visual Recognition by Chain-of-Thought Reasoning
Fine-R1 uses chain-of-thought supervised fine-tuning on a structured FGVR reasoning dataset plus triplet augmented policy optimization to outperform general MLLMs and CLIP models on seen and unseen fine-grained categories with 4-shot training.
-
InternVL3.5: Advancing Open-Source Multimodal Models in Versatility, Reasoning, and Efficiency
InternVL3.5 advances open-source multimodal models with Cascade RL for +16% reasoning gains and ViR for 4x inference speedup, with the 241B model reaching SOTA among open-source MLLMs on multimodal, reasoning, and agentic tasks.
-
GraspVLA: a Grasping Foundation Model Pre-trained on Billion-scale Synthetic Action Data
GraspVLA shows that pretraining a grasping model on a billion synthetic action frames enables zero-shot open-vocabulary performance and sim-to-real transfer.
-
Visual-RFT: Visual Reinforcement Fine-Tuning
Visual-RFT applies reinforcement learning with verifiable perception rewards to improve large vision-language models on fine-grained classification, few-shot detection, and grounding tasks.
-
Expanding Performance Boundaries of Open-Source Multimodal Models with Model, Data, and Test-Time Scaling
InternVL 2.5 is the first open-source MLLM to surpass 70% on the MMMU benchmark via model, data, and test-time scaling, with a 3.7-point gain from chain-of-thought reasoning.
-
Enhancing the Reasoning Ability of Multimodal Large Language Models via Mixed Preference Optimization
Mixed Preference Optimization with the MMPR dataset boosts multimodal CoT reasoning, lifting InternVL2-8B to 67.0 accuracy on MathVista (+8.7 points) and matching the 76B model.
-
From Crowdsourced Data to High-Quality Benchmarks: Arena-Hard and BenchBuilder Pipeline
BenchBuilder automates extraction of 500 challenging prompts from crowd-sourced data into Arena-Hard-Auto, delivering 3x model separation versus MT-Bench and 98.6% human-preference correlation at $20 cost.
-
LASER: A Corrective Lens for LVLMs via Visual Attention Preservation and Sink Suppression
LASER uses Visual Grounding Reward and Sink Suppression Reward to preserve visual attention trajectories and suppress sink tokens, reducing visual forgetting in LVLMs.
-
LiveStarPro: Proactive Streaming Video Understanding with Hierarchical Memory for Long-Horizon Streams
LiveStarPro uses SVeD for response timing via perplexity, SCAM for incremental alignment, and TSHM for event-chain memory to achieve 28.9% better semantic correctness and 1.58x speedup on long video streams.
-
MOSAIC: Efficient Mixture-of-Agent Scheduling via Adaptive Aggregation and Inference Concurrency
MOSAIC uses an Integer Linear Program scheduler for expert placement and prompt assignment plus adaptive aggregation to achieve 1.7-2.3x end-to-end speedup on 4-GPU MoA workloads while keeping accuracy within 0.1pp.
-
Grounded 3D-Aware Spatial Vision-Language Modeling
GR3D is a VLM that combines explicit 2D, implicit 2D, and monocular 3D grounding mechanisms to improve performance on spatial understanding benchmarks.
-
Why Do Vision Language Models Struggle To Recognize Human Emotions?
VLMs fail at video emotion recognition because they collapse rare emotions into common ones and process frames as an unordered bag; summarizing skipped frames in text partially recovers temporal cues.
-
A Scalable Multi-LLM Collaboration System with Retrieval-based Selection and Exploration-Exploitation-Driven Enhancement
SMCS coordinates 15 open-source LLMs via retrieval-based prior selection and exploration-exploitation posterior enhancement, outperforming GPT-4.1 by 5.36% and GPT-o3-mini by 5.28% on eight benchmarks.
-
Towards Reasoning Era: A Survey of Long Chain-of-Thought for Reasoning Large Language Models
The paper unifies perspectives on Long CoT in reasoning LLMs by introducing a taxonomy, detailing characteristics of deep reasoning and reflection, and discussing emergence phenomena and future directions.
-
InternVideo2.5: Empowering Video MLLMs with Long and Rich Context Modeling
InternVideo2.5 improves video MLLMs by incorporating dense vision task annotations via direct preference optimization and compact spatiotemporal representations via adaptive hierarchical token compression, yielding better benchmark performance, 6x longer video memory, and new capabilities likeobject
-
InternLM-XComposer-2.5: A Versatile Large Vision Language Model Supporting Long-Contextual Input and Output
InternLM-XComposer-2.5 is a 7B vision-language model supporting up to 96K context that reaches GPT-4V-level performance on image, video, and multi-turn tasks and adds LoRA-driven text-image composition capabilities.
-
InternVideo3: Agentify Foundation Models with Multimodal Contextual Reasoning
InternVideo3 introduces Multimodal Contextual Reasoning and M^2LA attention to enable closed-loop evidence accumulation in long-video understanding and agentic tool use, reporting strong benchmark results.
-
Exploring Lightweight Large Language Models for Court View Generation
Lightweight LLMs are benchmarked for court view generation and charge prediction across architectures, sizes, DNN comparisons, and task ordering on three datasets using the new CVGEvalKit framework.
-
From Memorization to Creation: Evaluating the Cognitive Depth of LLM-Generated Educational Questions
Evaluation of LLMs shows that specific prompting can increase higher-order questions by 11.53% and reduce repetitiveness by 24.45% in question generation for education.
-
Skywork-Reward: Bag of Tricks for Reward Modeling in LLMs
Data-centric filtering yields an 80K preference dataset and reward models that lead RewardBench while boosting other top entries.
-
MinerU: An Open-Source Solution for Precise Document Content Extraction
MinerU delivers an open-source pipeline for high-precision document content extraction by integrating specialized models with tuned preprocessing and postprocessing rules.
-
How Far Are We to GPT-4V? Closing the Gap to Commercial Multimodal Models with Open-Source Suites
InternVL 1.5 narrows the performance gap to proprietary multimodal models via a stronger transferable vision encoder, dynamic high-resolution tiling, and curated English-Chinese training data.
-
Enhancing Fitness Intelligence through Domain-Specific LLM Post-Training
FitOne-8B/32B models improve average scores on ACSM-EP and NSCA-CSCS certification exams by up to 12.73% over base Qwen3 while retaining general capabilities.
-
Data-Centric Foundation Models in Computational Healthcare: A Survey
The paper surveys data-centric strategies for foundation models in computational healthcare and supplies a curated list of related models and datasets.
- MMLA: How Memory Lets the Past Shape the Future
- CAR: Query-Guided Confidence-Aware Reranking for Retrieval-Augmented Generation