InfiX.ai

Powering the Last Mile
of GenAI

Our Mission and Vision

InfiX.ai is pioneering collaborative generative AI through low-resource training, model fusion, GUI agents, and multimodal intelligence.

Mission

At InfiX.ai, we believe in making AI more accessible, efficient, and capable. Our research focuses on developing innovative techniques that push the boundaries of what AI systems can achieve while maintaining practical applicability.

Vision

We are not only building another model - We are building the system that builds all models.

ACCESSIBLE EFFICIENT CAPABLE

Research Focus

Low-Precision Training & Efficient Inference

Achieve near-BF16 accuracy with significantly greater efficiency. Compared with traditional FSDP, InfiX.ai cuts training time by 48.1%, boosts training speed by 92.6%, and reduces memory pressure and inference costs.

Low-Precision Training & Efficient Inference 1
Low-Precision Training & Efficient Inference 2
Low-Precision Training & Efficient Inference 3

Model Fusion & Model Merging

Fuse model strengths into stronger Expert AI without multi-model inference overhead. Support fusion of 10B-scale foundation models across multiple domains, requiring only one ten-thousandth of the GPU hours needed for full pretraining.

Model Fusion & Model Merging 1

Latest Research

arXivOct 05 , 2026

TRIAGE: Direction-Aware Mismatch Stabilization of Native NVFP4 Reinforcement Learning

Low-precision execution can substantially accelerate reinforcement learning (RL) for large language models, but discrepancies between learner and sampler execution can destabilize policy optimization. In this paper, we characterize the interaction between mismatch and the policy-gradient direction, distinguishing locally amplifying from contracting update contributions that mismatch magnitude alone cannot identify. In native NVFP4 runs, we observe an early imbalance between the two amplifying regions, favoring negative-advantage, negative-gap updates. Their tail tokens become concentrated in a small fraction of response segments before mismatch spreads globally. Motivated by these findings, we introduce TRIAGE, a direction-aware stabilization method that uses segment-level diagnosis to selectively rebalance policy-gradient updates and applies bounded repair to residual severe mismatch. TRIAGE modifies the optimization objective while retaining native NVFP4 weight-and-activation 4-bit (W4A4) forward execution on both the sampler and learner. Experiments on Qwen3-4B and Qwen3-30B-A3B show stable optimization throughout the evaluated training horizon and achieve full precision level performance across five mathematical reasoning benchmarks, while native NVFP4 with TRIAGE provides up to 2.3× higher rollout throughput than BF16.

NeurIPS 2026Sep 25 , 2026

FeatCal: Feature Calibration for Post-Merging Models

Model merging combines task experts into one model and avoids joint training, retraining, or deploying many expert models, but the merged model often still underperforms task experts. We study this performance gap through feature drift, the difference between features produced by the merged model and by the expert on the same input. Our theory decomposes this drift into upstream propagation and local mismatch, tracks how it propagates and combines through later layers in forward order, and links final feature drift to output drift. This view motivates FeatCal, which uses a small calibration set to calibrate the merged model weights layer by layer in forward order, reducing feature drift while staying close to merged weights and preserving the benefits of model merging. FeatCal uses an efficient closed-form solution to update model weights, with no gradient descent, iterative optimization, or extra modules. On the main CLIP and GLUE benchmarks, FeatCal beats Surgery and ProbSurgery, the closest post-merging calibration baselines: 85.5% vs. 77.0%/78.8% on CLIP-ViT-B/32 Task Arithmetic (TA) and 85.2% vs. 83.7%/82.2% on FLAN-T5-base GLUE. On CLIP-ViT-B/32, 8 examples per task reach 82.9%, and 256 examples per task take 53 seconds, about 4x faster than both baselines, showing better sample efficiency and lower calibration cost.

NeurIPS 2026Sep 25 , 2026

Geometry Conflict: Explaining and Controlling Forgetting in LLM Continual Post-Training

Continual post-training aims to extend large language models (LLMs) with new knowledge, skills, and behaviors, yet it remains unclear when sequential updates enable capability transfer and when they cause catastrophic forgetting. Existing methods mitigate forgetting through sequential fine-tuning, replay, regularization, or model merging, but offer limited criteria for determining when incorporating new updates is beneficial or harmful. In this work, we study LLM continual post-training through three questions: What drives forgetting? When do sequentially acquired capabilities transfer or interfere? How can compatibility be used to control update integration? We address these questions through task geometry: we represent each post-training task by its parameter update and study the covariance geometry induced by the update. Our central finding is that: forgetting can be considered as a state-relative update-integration failure, it arises when the covariance geometries induced by tasks misalign with the geometry of the evolving model state. Sequential updates transfer when they remain compatible with the model state shaped by previous updates, and interfere when state-relative geometry conflict becomes high. Motivated by this finding, we propose Geometry-Conflict Wasserstein Merging (GCWM), a data-free update-integration method that constructs a shared Wasserstein metric via Gaussian Wasserstein barycenters and uses geometry conflict to gate geometry-aware correction. Across Qwen3 0.6B--14B on domain-continual and capability-continual settings, GCWM consistently outperforms data-free baselines, improving retention and final performance without replay data. These results identify geometry conflict as both an explanatory signal for forgetting and a practical control signal for LLM continual post-training.

View All Research
InfiXspeed

Build your own Expert AI
with FP8-powered InfiXspeed.

InfiXspeed is InfiX.ai's enterprise platform for efficient domain AI development. It helps organizations turn proprietary data, business workflows, and expert knowledge into domain-specific Expert AI that can be trained, evaluated, and deployed in secure, controlled environments.

Learn more
InfiXspeed dashboard preview

Latest News

See InfiX.ai in Action

Explore More About Us

GitHub

Explore open-source models and research code

Hugging Face

Access our pre-trained models and datasets

Contact Us

contact@infix.io