Supervised fine-tuning

What it is

You take a pre-trained model and perform additional training passes over a specialized dataset consisting of prompt-response pairs. Unlike pretraining, which involves massive unlabeled data, this requires only hundreds or thousands of high-quality examples to influence the model's tone, format, or domain logic. The weight updates are small, effectively nudging the existing parameters toward a desired output distribution. This process typically costs a fraction of the original training budget and can often be completed in hours on a few GPUs.

Why it matters

Knowing this helps you decide between complex prompt engineering and model customization. If your prompts are becoming too large or you cannot maintain consistent formatting across long chains of logic, fine-tuning provides a more deterministic way to enforce structure. Ignoring this often leads to bloated prompt overhead, increased token costs per request, and fragile performance that breaks when input variety increases.

In practice

In practice, you prepare a JSONL file of input-output samples and submit it to a provider's fine-tuning API. You monitor the training loss curve during the process to ensure the model learns the patterns without memorizing the training data. Once completed, you deploy the resulting model variant as a custom endpoint or specific model ID in your standard inference calls.

The tradeoff

The primary risk is overfitting, where the model performs perfectly on your small training set but loses its general reasoning capabilities for inputs outside your narrow use case.

Where it appears

Research summaries that use Supervised fine-tuning, each linked to its source paper.

• Evaluating Android Agent Robustness to Anomalies
  Are Android GUI Agents Robust Against Runtime Anomalies? AnTrap: Evaluating Agents in Dynamic Adversarial Environments

  The paper introduces AnTrap, a benchmark that tests how Android GUI agents handle dynamic environmental anomalies by injecting perturbations into 236 tasks.

• Benchmarking Explicit Strategy Induction in LLMs
  StrategyBench: Evaluating Explicit Strategy Induction in Large Language Models

  The paper introduces StrategyBench to evaluate if language models can effectively derive and apply explicit task-level strategies from few-shot examples.

• Scaling Agentic Intelligence for Complex Work
  Apodex 1.1: Scaling Agentic Intelligence for Complex Work

  Apodex 1.1 provides a general purpose agentic system that scales intelligence for complex professional tasks across finance and science using a robust execution framework.

• Training Adaptable Agents for Live Streaming
  Training Agents to Evolve with Their Harness: TaoLive Digital Avatar Agent Technical Report

  The paper introduces a method called Harness Evolution that improves agent adaptability and performance for live-streaming environments by decoupling execution settings from the base model.

• Improving Robot Planning With Rule-Compliant Models
  Rule-Compliant Visual Spatial Planning for Multimodal Large Language Models

  The researchers developed a framework that improves how multimodal large language models follow natural-language rules when performing spatial planning tasks.

• Using Language Models to Guide Robotics
  EXIMO: VLM Guided Exploration of VLA Policies

  EXIMO leverages a vision-language model to decompose complex robotic tasks into smaller steps, improving the efficiency of training vision-language-action policies.

• Improving AI Tool Use Through Mid-Training
  MidTool: Mid-training Data Synthesis for Agentic Tool Use

  The researchers created a 20.3B-token corpus called MidTool-Mix to improve agentic tool-use capabilities in models during the mid-training phase rather than relying solely on post-training.

• Optimizing GPU Kernels Using LLMs
  PTXBench: Benchmark and Adapt LLMs for GPU Kernel Optimization with Architecture-specific PTX

  The authors introduce PTXBench and a supervised fine-tuning method to help LLMs write efficient architecture-specific GPU code.

• Improving Tool Use With Looped Language Models
  Looped Language Models Improve Compositional Tool Calling

  The researchers developed Ouro models that use recurrent stack iterations to improve performance on complex, compositional tool-calling tasks.

• Efficient Navigation for Vision Language Models
  Embodied-Navigator: Point, Think, Memorize, and Align for Efficient Navigation

  TAMP-Nav improves embodied navigation by combining efficient 3D spatial grounding with selective reasoning and a multi-level reward training approach.

• Teaching LLM Agents to Explore Better
  Clearing the Fog: Towards Installing and Refining Proactive Exploration Capabilities in LLM Agents

  The paper introduces a method using synthetic data and reinforcement learning to help language model agents explore proactive action strategies instead of falling into rigid behavioral patterns.

• Improving Long Horizon Remote Sensing Reasoning
  LongEarth-R1: Benchmarking and Aligning Vision-Language Models for Long-Horizon Earth Observation Reasoning

  LongEarth-R1 enhances long duration satellite image analysis by aligning vision language models with structured temporal reasoning and reward based feedback.

• Adversarial AI-Generated Image Detection
  SPARED: Reasoning-Based AI-Generated Image Detection via Adversarially Edited Data

  The authors introduce SPARED, a reasoning-based detector that uses adversarial image editing to train models to identify synthetic content without relying on provenance shortcuts.

• Generating Physically Accurate Scientific Diagrams
  Towards Physics-Faithful Generation of Scientific Diagrams

  The paper introduces Princigram, a framework that uses structured physics-based constraints to improve the fidelity of scientific images generated by multimodal models.

• Unlocking Enhanced Spatial Reasoning for VisionLanguage Models
  SCOUT: Unlocking Enhanced Spatial Reasoning via Structured Chain-of-Thought and Multi-Objective Process Reward

  The paper introduces SCOUT, a method combining structured chain-of-thought prompting and multi-objective reinforcement learning to fix spatial reasoning bottlenecks in vision-language models.

• Multimodal AI Agents for Business Ideation
  MBA: Multimodal Benchmark and Agents for Real-World Business Ideation

  The researchers developed a new benchmark and multimodal agent system that uses visual data to generate more realistic business ideas than text-only alternatives.

• Improving Reliability in Medical AI Diagnostics
  CARE: Confidence-Aware Reasoning for Reliable Medical VQA

  The paper introduces a confidence-aware training framework that aligns medical diagnostic predictions with actual accuracy to reduce clinical decision errors.

• Unified Pixel and Language Medical Model
  MedPixel: A Unified Pixel-Language Model for Medical Reasoning and Segmentation

  MedPixel combines visual reasoning and image segmentation into a single architecture to bridge the gap between clinical text and pixel-level data.

• Agentic Visual Search for Long Documents
  InSight-doc: Agentic Visual Perception for Long-Document Understanding

  InSight-doc uses an agentic system that zooms into document regions to reduce computational overhead and hallucination in multimodal models.

• Predicting Stocks Using Language Model Tokens
  Financial Numerical Prediction and Allocation as Token Generation

  Researchers successfully used a single causal language model to perform stock forecasting and portfolio allocation by treating financial data as tokens rather than using traditional task-specific numerical heads.

• Improving GUI Agent Accuracy Through Reflection
  LookAgain: Closed-Loop GUI Grounding with Visually Grounded Reflection

  LookAgain introduces a multi-turn refinement process that allows GUI agents to visually verify and adjust their coordinate predictions, significantly boosting accuracy on complex interfaces.

• Optimizing Reasoning and Response Modes in LLMs
  Fusion Training for Mathematical Generalization in Large Language Models

  The researchers investigated how to balance a model's ability to provide concise answers with its capacity for long-form mathematical reasoning by testing different training schedules and data ratios.

• Controlling User Simulator Intent in Conversations
  Intent Speaks Louder: Controllable User Simulation Beyond Response Imitation

  The researchers developed a framework called UserIDA that allows developers to precisely control the local conversational intent of LLM-based user simulators.

• Improving Japanese Structured Document Parsing
  Stockmark-Nemotron-3-Nano-Omni-JapanDocReader: Structured Document Parsing via Capability Injection and Forgetting Control

  The researchers developed Stockmark-Nemotron-3-Nano-Omni-JapanDocReader to balance structured document parsing with existing document visual question answering capabilities.

• Improving LLM Reliability for Scientific Data Analysis
  Fisher-R1: Training LLM Agents for Reliable Hypothesis Testing

  Fisher-R1 is a specialized LLM agent trained to perform reliable hypothesis testing by using a new benchmark and outcome-grounded reinforcement learning.

• Using Language Models for Metabolite Analysis
  MetaboLLM: a metabolomics-specialized large language model for biochemical knowledge integration and predictive metabolite graph construction

  Researchers developed MetaboLLM to integrate biochemical knowledge and convert it into predictive metabolite graphs for clinical diagnostics.

• OneEmo: Unified Emotion AI Reasoning Model
  OneEmo: A Unified Multimodal Reasoning Model for Emotion Perception, Understanding, and Interaction

  OneEmo is a 4.5B parameter multimodal model that improves emotion perception and understanding by using a novel reinforcement learning framework and a human-in-the-loop reasoning dataset.

• Incentivizing LLM Reasoning via Pure Reinforcement Learning
  DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning

  The paper introduces DeepSeek-R1, demonstrating that large language models can develop advanced reasoning capabilities through pure reinforcement learning without requiring prior supervised fine-tuning.

• Simpler Control of Language Models
  Direct Preference Optimization: Your Language Model is Secretly a Reward Model

  Researchers developed a simpler method to control language model behavior directly from human preferences without training a separate reward model or using reinforcement learning loops.

• Understanding Reinforcement Learning Post-Training
  Understanding Reasoning from Pretraining to Post-Training

  This paper investigates how pretraining choices shape reinforcement learning returns and what reinforcement learning actually does to a model policy using chess games and puzzles.

• Improving Agent Decision Making Through Self-Evolution
  SEED: Self-Evolving On-Policy Distillation for Agentic Reinforcement Learning

  The paper introduces a method called SEED that improves agentic performance by having the model analyze its own past trajectories to generate dense feedback for training.

• Improving Geo-localization by Reducing Landmark Bias
  HoloGeo: Mitigating Landmark Bias in Geo-localization via Evidence-Driven Reasoning

  HoloGeo is a framework that improves image geo-localization accuracy by training models to reason beyond superficial visual landmarks using evidence-driven reinforcement learning.

• Building Reliable Mobile Agents for Real World Tasks
  Exploratory, Communicative, and Deployable: Vision-Driven Embodied Agents for Open-World Mobile Manipulation

  The paper introduces a framework for embodied agents that utilizes visual navigation, interactive intent disambiguation, and reinforcement learning to perform physical manipulation tasks in open world settings.

• Search Agents Directly Querying Raw Text
  GrepSeek: Training Search Agents for Direct Corpus Interaction

  GrepSeek replaces traditional vector-based search indices with an agent that interacts directly with raw corpus files using shell commands.