REVIEW 37 cited by
Eagle 2: Building Post-Training Data Strategies from Scratch for Frontier Vision-Language Models
Not yet reviewed by Pith; the record is open.
This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.
SPECIMEN: schema-true, not a live event
T0 review · schema-true
One-sentence machine reading of the paper's core claim.
pith:XXXXXXXX · record.json · timestamp
Eagle 2: Building Post-Training Data Strategies from Scratch for Frontier Vision-Language Models
read the original abstract
Recently, promising progress has been made by open-source vision-language models (VLMs) in bringing their capabilities closer to those of proprietary frontier models. However, most open-source models only publish their final model weights, leaving the critical details of data strategies and implementation largely opaque. In this work, we address VLM post-training from a data-centric perspective, showing the key role of data strategy in developing frontier VLMs. By studying and building our post-training data strategy from scratch, we share detailed insights into the development processes, aiming to benefit the development of competitive models for the open-source community. Our introduced data strategy, together with training recipes and model design, leads to a family of performant VLMs named Eagle2. Specifically, Eagle2-9B achieves state-of-the-art results across various multimodal benchmarks, matching certain competitive models with up to 70B parameters.
Forward citations
Cited by 37 Pith papers
-
Hand-Object Interaction in the Age of Large Foundation Models:Reconstruction, Generation, and Embodied Transfer
A taxonomy of eight foundation-model priors organizes HOI reconstruction, generation, and embodied transfer, mapping what knowledge large models inject and where.
-
AMALIA-VL: A Native European Portuguese Open-Source Vision and Language Model
AMALIA-VL introduces the first open-source instruction-tuned LVLM natively optimized for European Portuguese via vision-language alignment, instruction tuning, preference optimization, and a pt-PT-centric data mix.
-
Argus-Retriever: Vision-LLM Late-Interaction Retrieval with Region-Aware Query-Conditioned MoE for Visual Document Retrieval
Argus achieves the highest reported NDCG scores among open late-interaction models on ViDoRe V1 and combined V1+V2 by introducing query-dependent document representations via a region-aware MoE on Qwen3.5-VL, trained ...
-
Being-H0.7: A Latent World-Action Model from Egocentric Videos
Being-H0.7 adds future-aware latent reasoning to direct VLA policies via dual-branch alignment on latent queries, matching world-model benefits at VLA efficiency.
-
EmbodiedMidtrain: Bridging the Gap between Vision-Language Models and Vision-Language-Action Models via Mid-training
EmbodiedMidtrain mid-trains VLMs on curated VLA-aligned data subsets to improve downstream performance on robot manipulation benchmarks.
-
High-Resolution Visual Reasoning via Multi-Turn Grounding-Based Reinforcement Learning
MGPO elicits grounding in LMMs via multi-turn RL with binary rewards, yielding 5.4% and 5.2% gains on MME-Realworld and V* Bench and surpassing GPT-4o on the latter after training on 21K samples.
-
RoboInter1.5: A Holistic Intermediate Representation Suite for Embodied World Modeling and Robotic Manipulation
Dense per-frame intermediate representations (traces, masks, grasp poses, subtasks) improve embodied VQA, VLA action generation, and world-model video prediction in the new 230k-episode RoboInter-Data suite.
-
AMALIA-VL: A Native European Portuguese Open-Source Vision and Language Model
AMALIA-VL is the first open-source LVLM natively optimized for European Portuguese via three-stage training on a pt-PT-centric data mix combining curated, translated, and novel datasets.
-
AMALIA-VL: A Native European Portuguese Open-Source Vision and Language Model
Introduces AMALIA-VL, the first open-source instruction-tuned LVLM for European Portuguese, using a high-resolution vision encoder, pt-PT language model, learned connector, and three-stage training on a custom data mix.
-
Contrastive Action-Image Pre-training for Visuomotor Control
CAIP learns action-aligned visual representations via contrastive pre-training on human hand keypoints from egocentric video, outperforming DINOv2, SigLIP, MVP, and R3M with >30% gains on real dexterous manipulation tasks.
-
Hy-Embodied-0.5-VLA: From Vision-Language-Action Models to a Real-World Robot Learning Stack
A full learning stack—10K-hour UMI data, a flow-matching VLA, preference-optimization RL, and asynchronous deployment—reports SOTA RoboTwin results and cross-embodiment transfer to four real robots.
-
RhinoVLA Technical Report
RhinoVLA matches π0.5-scale VLA task performance while reaching 11.69 Hz end-to-end inference on the Huixi R1 edge SoC via token-efficient Qwen3-VL, a 72D unified action interface, and hardware co-optimization.
-
LARA: Latent Action Representation Alignment for Vision-Language-Action Models
LARA jointly optimizes LAM and VLA models via representation alignment to improve robotic manipulation performance using human videos.
-
Cambrian-P: Pose-Grounded Video Understanding
Adding per-frame camera-pose supervision to a video MLLM improves spatial and general video question answering by 2–6% and yields SOTA streaming pose estimates on ScanNet.
-
Cambrian-P: Pose-Grounded Video Understanding
Cambrian-P adds per-frame camera pose tokens and a regression head to video MLLMs, delivering 4.5-6.5% gains on spatial benchmarks, generalization to other video QA tasks, and SOTA streaming pose estimation on ScanNet.
-
$M^2$-VLA: Boosting Vision-Language Models for Generalizable Manipulation via Layer Mixture and Meta-Skills
M²-VLA shows that generalized VLMs can serve as direct backbones for robotic manipulation by selectively extracting task-critical features via Mixture of Layers and adding Meta Skill Modules for efficient trajectory learning.
-
$M^2$-VLA: Boosting Vision-Language Models for Generalizable Manipulation via Layer Mixture and Meta-Skills
Freezing a VLM backbone and routing its layers through Mixture-of-Layers plus a Meta-Skill memory yields higher success and stronger zero-shot generalization than fine-tuned VLAs on LIBERO and real robots.
-
Unmasking the Illusion of Embodied Reasoning in Vision-Language-Action Models
State-of-the-art vision-language-action models catastrophically fail dynamic embodied reasoning due to lexical-kinematic shortcuts, behavioral inertia, and semantic feature collapse caused by architectural bottlenecks...
-
MLLM-as-a-Judge Exhibits Model Preference Bias
MLLMs show self-preference bias and family-level mutual bias when judging captions; Philautia-Eval quantifies it and Pomms ensemble reduces it.
-
Adaptive Action Chunking at Inference-time for Vision-Language-Action Models
Adaptive Action Chunking uses action entropy to dynamically adjust chunk sizes in VLA models, improving performance on simulated and real robotic manipulation tasks.
-
RefBench-PRO: Perceptual and Reasoning Oriented Benchmark for Referring Expression Comprehension
RefBench-PRO organizes REC into attribute, position, interaction, relation, commonsense, and reject tasks; no tested MLLM exceeds 72%, and Ref-R1 raises Qwen2.5-VL-7B from 57.6 to 69.4 on it.
-
FG-CLIP 2: A Bilingual Fine-grained Vision-Language Alignment Model
A two-stage bilingual CLIP-style model with region-text supervision and a new text-side contrastive loss outperforms prior open models on fine-grained vision-language tasks in English and Chinese.
-
FLARE: Robot Learning with Implicit World Modeling
FLARE integrates predictive latent world modeling into diffusion transformer policies for robots, delivering up to 26% gains on multitask manipulation benchmarks and enabling co-training with action-free human videos.
-
InternVL3: Exploring Advanced Training and Test-Time Recipes for Open-Source Multimodal Models
InternVL3-78B sets a new open-source SOTA of 72.2 on MMMU via native joint multimodal pre-training, V2PE, MPO, and test-time scaling while remaining competitive with proprietary models.
-
SmolVLM: Redefining small and efficient multimodal models
SmolVLM-256M outperforms a 300-times larger model using under 1 GB GPU memory, while the 2.2B version matches state-of-the-art VLMs at half the memory cost.
-
GR00T N1: An Open Foundation Model for Generalist Humanoid Robots
GR00T N1 is a new open VLA foundation model for humanoid robots that outperforms imitation learning baselines in simulation and shows strong performance on real-world bimanual manipulation tasks.
-
SigLIP-HD by Fine-to-Coarse Supervision
Fine-to-coarse L1 supervision lets a standard-resolution SigLIP 2 encoder produce better visual tokens for MLLMs without higher-resolution inference.
-
Hand-Object Interaction in the Age of Large Foundation Models:Reconstruction, Generation, and Embodied Transfer
Foundation-model HOI work is organized into eight geometric, semantic, and visual sub-priors that enter six reconstruction/generation tasks and three robot-transfer routes.
-
LeapBot-WA: World-Anchor Action Models via Predictive Latent Alignments
LeapBot-WA shows robot policies can be trained with latent world-model predictions instead of pixel video generation, hitting state-of-the-art for predictive action models and staying competitive with generative WAMs.
-
RhinoVLA Technical Report
RhinoVLA runs a ~2.5B-parameter robot policy end-to-end at 11.69 Hz on the Huixi R1 edge SoC with LIBERO accuracy within a few points of π0.5 and mixed real-robot results.
-
RhinoVLA Technical Report
RhinoVLA uses a token-efficient Qwen3-VL backbone, continuous Action Expert, and unified cross-robot interface to match π0.5 performance while hitting 11.69 Hz on Huixi R1 edge SoC.
-
LARA: Latent Action Representation Alignment for Vision-Language-Action Models
LARA jointly optimizes LAM and VLA models via representation alignment, reporting average gains of ~10%, ~5%, and ~15% on simulation and real robotic manipulation tasks.
-
Gaze2Act: Gaze-Conditioned Vision-Language-Action Policies for Interactive Robot Manipulation
Gaze2Act conditions VLA policies on mapped human gaze for precise object and interaction specification, reporting SOTA intent accuracy and success across 16 real-robot tasks on a Unitree G1 humanoid.
-
ThinkAct: Vision-Language-Action Reasoning via Reinforced Visual Latent Planning
ThinkAct introduces reinforced visual latent planning in a dual VLA system to enable better long-horizon reasoning and adaptation for embodied tasks.
-
Consistency as Inductive Bias: Learning Cross-View Invariance for Robust Multimodal Reasoning
ConsistRoll enforces cross-view consistency during RLVR training for MLLMs by joint rewards on grouped original and augmented views, yielding robustness gains on math, general, and hallucination benchmarks.
-
PLaMo 2.1-VL Technical Report
PLaMo 2.1-VL reports 61.5 ROUGE-L on JA-VG-VQA-500, 85.2% on Japanese Ref-L4, 53.9% zero-shot factory accuracy, and raises anomaly detection F1 from 39.7 to 64.9 after fine-tuning.
-
RhinoVLA Technical Report
RhinoVLA cuts VLM tokens with a Qwen3-VL backbone and continuous action expert, adds a unified cross-robot interface, and reaches real-time 11.69 Hz on Huixi R1 while matching π0.5 downstream performance.
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.