Supervised fine-tuning (SFT) on visual instruction data often improves perceptual capabilities in vision-language models (VLMs) while degrading reasoning performance, creating a persistent reasoning tax during post-training. We investigate whether this degradation is related to disrupted access to depth-wise representations, and find that even fixed cross-depth aggregation substantially restores reasoning, suggesting that preserved cross-depth access is an important missing factor in VLM fine-tuning. Building on this observation, we propose Input-Adaptive Depth Aggregation (IADA), a lightweight mechanism that makes cross-depth retrieval input-adaptive, modality-aware, and efficiently parameterized through a low-rank bottleneck. On Qwen3-VL-2B, IADA improves the average reasoning score by 9.5 points and the average perception score by 3.3 points over LoRA-only fine-tuning with only 0.14M additional parameters, with the strongest gains appearing in parameter-efficient low-rank settings.
Figures & tables
Figure 1: Overview of IADA. Compared with LoRA fine-tuning in (a), IADA in (b) inserts an IADA block at block boundaries to enable explicit cross-depth routing. IADA generates modality-specific queries from the current hidden states, retrieves information from cached states at previous nd block boundaries, and injects the resulting aggregated residuals back through gated residual addition.
LoRA Rank
LoRA Params
Method
R-Avg
P-Avg
All-Avg
16
3.8M
B: LoRA-only
52.2
72.0
63.0
D: IADA +LoRA
61.7
75.3
69.1
Δ (D − B)
+9.5
+3.3
+6.1
32
7.6M
B: LoRA-only
54.8
72.8
64.6
D: IADA +LoRA
55.3
75.1
66.1
Δ (D − B)
+0.5
+2.3
+1.5
Table 3: Interaction between IADA and LoRA rank. LoRA rank and its corresponding trainable parameter count are shown explicitly, where LoRA Params denotes the trainable parameters introduced by LoRA alone. The gains of IADA are substantially larger at rank 16 than at rank 32 , indicating that explicit depth routing is especially beneficial when PEFT capacity is constrained.
Vision-language model adaptation tailors pretrained image-text and multimodal models to new tasks and domains while retaining transferable capabilities.
Recent advances in vision-language models (VLMs) emphasize long chain-of-thought reasoning; yet, we find that their performance on visual tasks is primarily limited by a lack of visual perception as opposed to reasoning itself. In this work, we systematically study the interplay between perception and reasoning in VLM post-training by decomposing their capabilities into three separate training stages: visual perception, visual reasoning, and textual reasoning, incorporating specialized training data. We demonstrate that visual perception (a) requires targeted optimization with specialized data; (b) serves as a fundamental scaffold that should be solidified through staged training before refining visual reasoning; and (c) is more effectively learned via RL than caption-based SFT. Our experiments across multiple VLMs demonstrate that staged training consistently improves both visual perception and reasoning performance over merged training. Notably, models trained with our approach achieve 1.5% higher reasoning accuracy with 20.8% shorter reasoning traces, suggesting that superior perception reduces the need for excessive reasoning. Furthermore, we show that this capability-based staging represents a new curriculum dimension orthogonal to traditional difficulty-based curricula, and combining both yields further additive gains. Our staged-training models achieve superior performance among open-weight VLMs, establishing advanced results on several visual math and perception (e.g., +5.2% on WeMath and +3.7% on RealWorldQA) tasks compared with the base counterpart.
Juncheng Wu, Hardy Chen, Haoqin Tu +6
Amazon · UC Santa Cruz · University of Waterloo +1
Modern Vision-Language Models (VLMs) benefit from chain-of-thought prompting and test-time scaling, but these gains often come with prohibitive inference cost due to large visual contexts and long decoding chains. We view this cost through two coupled axes: Visual Context Scaling (VCS), which controls how much visual evidence is passed to the language model, and Visual Reasoning Scaling (VRS), which controls how much inference-time reasoning search is performed. Existing methods typically optimize one axis at a time, leaving the joint allocation of compute across these axes underexplored. We introduce Adaptive Visual Inference Scaling (AVIS), a lightweight policy that adapts both VCS and VRS per query. AVIS realizes VCS through Key Diversity Visual (KDV) pruning, a training-free O(N) key-based rule for removing redundant visual tokens before prefilling, and realizes VRS through adaptive self-consistency, using a learned difficulty predictor to select the number of reasoning rollouts. AVIS is deployment-friendly and compatible with shared-prefill inference, where all rollouts reuse a single prefilling pass and KV cache. Across diverse image and video reasoning benchmarks, AVIS improves the accuracy--compute trade-off relative to VCS-only and VRS-only baselines, and remains effective on top of RL post-trained VLMs while keeping compute and latency low.
Ahmadreza Jeddi, Minh Ngoc Le, Amirhossein Kazerouni +8
AI Center-Toronto, Samsung Electronics · University of Toronto · Vector Institute +1
Post-training has greatly improved reasoning in frontier vision-language models, yet its gains for perception remain comparatively limited, creating a bottleneck for end-to-end visual reasoning. To investigate this gap, we introduce a controlled diagnostic framework with two synthetic tasks that disentangle perception from reasoning. Our analysis reveals a consistent perception-reasoning asymmetry: post-training improves reasoning more substantially than perception, though the underlying mechanism differs across training paradigms. For supervised fine-tuning (SFT), this asymmetry stems from token imbalance, with perception occupying a smaller fraction of tokens in chain-of-thought supervision. Reweighting the loss boosts end-to-end performance by up to 18.2 points. For reinforcement learning (RL), the asymmetry instead arises from reward coupling, as outcome rewards correlate more strongly with reasoning than perception. Adding a perception-aware reward improves end-to-end accuracy by up to 6.0 points; when ground-truth perception rewards are unavailable, a reliable surrogate provides useful signal, yielding gains of 2.2 points. Beyond the controlled setting, these strategies also improve real-world visual reasoning, with gains of up to 3.3 points across three benchmarks. Overall, we diagnose the causes of asymmetric optimization and provide actionable guidance that benefits both synthetic and realistic settings.