MinT | Mind Lab Toolkit

MinT (Mind Lab Toolkit) is an RL infrastructure that helps agents and models learn from real experience. It abstracts away compute scheduling, distributed rollout, and training orchestration so teams iterate inside real tasks.

MinT provides a unified, reproducible way to run RL across models and tasks, focused on making LoRA RL simple, stable, and efficient. You define what to train, what data, how to optimize, and how to evaluate — MinT handles the rest.

User Control

What to Train(Models)

What Data(Datasets)

How to Learn(Loss / LoRA)

How to Evaluate(Metrics)

MinT

Infrastructure Complexity

GPU Cluster

Model States

Distributed Training

Supported models

6Families

27Model variants

4Types

Qwen16 variants

Qwen3.6-35B-A3BMoE

Qwen3.6-27BDense

Qwen3.5-4BDense

Qwen3.5-27BDense

Qwen3.5-35B-A3BMoE

Qwen3.5-397B-A17BMoE

Qwen3-4B-Instruct-2507Dense

Qwen3-8B-BaseDense

Qwen3-8BDense

Qwen3-30B-A3B-BaseMoE

Qwen3-30B-A3BMoE

Qwen3-30B-A3B-Instruct-2507MoE

Qwen3-VL-30B-A3B-InstructComing soon

Qwen3-32BDense

Qwen3-235B-A22B-Instruct-2507MoE

Qwen3-VL-235B-A22B-InstructComing soon

DeepSeek4 variants

DeepSeek-V3MoE

DeepSeek-V3.1MoE

DeepSeek-V3.2MoE

DeepSeek-V4Coming soon

Moonshot3 variants

Kimi-K2MoE

Kimi-K2.5Coming soon

Kimi-K2.6Coming soon

GLM2 variants

GLM5MoE

GLM5.1MoE

MiniMax1 variant

MiniMax 2.7MoE

VLA1 variant

π0VLA

Editions

Support

Community

5×8 support

Included usage

5M tokens

Custom quota

Model access

Qwen series

Qwen, GLM, MiniMax, Kimi, DeepSeek, VLA

Workspace

Standard cloud workspace

Private / VPC / Hybrid

Training loop

Basic

Dedicated

Security

Standard

Review / Compliance / Audit

Security & compliance

Privacy & security

Secure by default. Enterprise data is not used to train shared foundation models unless expressly authorized.

Sensitive-field detection and redaction before training
Encrypted transfer, encrypted storage, and least-privilege access
Audit logs, continuous monitoring, and review-ready controls

Data ownership

Your Enterprise Data remains owned and controlled by you, including datasets, outputs, LoRA weights, and reports.

Tenant-isolated cloud workspaces
Dedicated cloud, VPC, private, localized, or hybrid deployment
No commercial use without prior express authorization

Compliance

Built for regulated teams with MLPS Level 3 support and domestic data/privacy audit readiness.

China MLPS Level 3 compliance support
Domestic data and privacy obligation audit support
Multiple-chip compatibility for localized training

Cases

Medical Coding

-90%GPU cost

Medical coding and case-record post-training for quality-control workflows across clinical scenarios.

27 case-record QC indicators8,000+ physicians referenced in rolloutTraining steps reduced by 50%On-policy training time reduced by 50%

Personalized Agent

340Kmini-apps

Personal agent infrastructure where many 1T model instances share one base while each user keeps LoRA differences.

1M users in the case deckOnly 5B LoRA diff-params per userThousands of 1T instances supportedDeep memory-driven personalization

Smart Customer Service

+34%CSAT

LoRA post-training for a leading fintech support workflow with compliance, takeover, and cost improvements.

18M-user service scenarioCompliance accuracy 89% to 99.2%Human takeover 38% to 15%Training cost reduced by 68%

Citizen Services

6hpolicy updates

Policy-aware service agents for public cloud scenarios with fast policy refresh and measurable transfer reduction.

23M citizens served1,400+ services coveredPolicy error rate 23% to 5%Agent transfer 44% to 21%

Sales

Contact