SARL rewards reasoning topology to improve label-free RL, outperforming baselines with gains up to 44.7% on math and 34.6% on open-ended tasks while maintaining more stable training.
super hub Mixed citations
LlamaFactory: Unified Efficient Fine-Tuning of 100+ Language Models
Mixed citation behavior. Most common role is background (54%).
105
Pith papers citing it
13
external citations · Pith
Background
54% of classified citations
abstract
Efficient fine-tuning is vital for adapting large language models (LLMs) to downstream tasks. However, it requires non-trivial efforts to implement these methods on different models. We present LlamaFactory, a unified framework that integrates a suite of cutting-edge efficient training methods. It provides a solution for flexibly customizing the fine-tuning of 100+ LLMs without the need for coding through the built-in web UI LlamaBoard. We empirically validate the efficiency and effectiveness of our framework on language modeling and text generation tasks. It has been released at https://github.com/hiyouga/LLaMA-Factory and received over 25,000 stars and 3,000 forks.
hub tools
citation-role summary
background 7
method 6
citation-polarity summary
claims ledger
- abstract Efficient fine-tuning is vital for adapting large language models (LLMs) to downstream tasks. However, it requires non-trivial efforts to implement these methods on different models. We present LlamaFactory, a unified framework that integrates a suite of cutting-edge efficient training methods. It provides a solution for flexibly customizing the fine-tuning of 100+ LLMs without the need for coding through the built-in web UI LlamaBoard. We empirically validate the efficiency and effectiveness of our framework on language modeling and text generation tasks. It has been released at https://githu
authors
co-cited works
representative citing papers
FLEX is the first large-scale multimodal multiview dataset for fitness AQA, featuring RGB, 3D pose, sEMG and physiological data plus a Fitness Knowledge Graph for structured annotations and a VideoQA benchmark.
CASS supplies the first large verified dataset and trained models for cross-vendor GPU code translation between CUDA and HIP at source and assembly levels, with reported accuracies of 88.2% and 69.1% that beat commercial baselines.
Hallucination Self-Play co-evolves a generator and detector from one base LLM via RLAIF and RLVR, lifting a 7B model to match advanced LLMs on RAGTruth faithfulness detection.
Online IL overcomes an information-theoretic bottleneck that offline IL faces in non-realizable settings even at horizon 1, under a new structural characterization of reward-relative misspecification.
KDoS optimizes synthetic data distributions via a three-stage knowledge density feedback mechanism to maximize LLM knowledge boundary expansion, with the optimal distribution stable across model sizes and data scales.
RedAct redacts agent traces to drop normalized skill transfer below the no-skill baseline on CapTraceBench while preserving audit evidence and adding detectable behavioral watermarks.
VideoKR supplies 315K knowledge-intensive video reasoning examples and a dedicated benchmark, with experiments indicating post-training gains on reasoning tasks that require both video content and external knowledge.
RogueMerge is a unified attack method that jointly optimizes task vectors to succeed after merging, using stochastic min-max simulation for unknown merging settings and a Taylor-approximated DRO for prompt generalization on generative LLMs.
Using frontier models to synthesize plausible-but-wrong FIM completions as hard negatives for SFT improves Delulu exact match by +18.8 and edit similarity by +0.22 on Qwen2.5-Coder-7B while also lifting HumanEval-Infilling and SAFIM.
Introduces pause-and-think-T dataset and pause-and-think-B benchmark; fine-tunes 4B VLM to 58% accuracy matching 235B model while generalizing out-of-distribution.
A genome-conditioned 4B LLM agent predicts microbial life boundaries and matches larger frontier models via token fusion, tool use, and a counterfactual gene-grounding reward.
ThinC trains small models to reason primarily in code rather than natural language, outperforming tool-integrated baselines and even larger models on competition math benchmarks.
DiagramNet supplies a new multimodal dataset and progressive training pipeline with decoupled multi-agent workflow, allowing a 3B model to outperform GPT-5, Claude-Sonnet-4, and Gemini-2.5-Pro by over 2x on system-level diagram tasks while generalizing to other benchmarks.
World2Minecraft turns real scenes into Minecraft worlds via occupancy prediction and releases a large indoor occupancy dataset to improve such models.
BERAG applies Bayesian ensemble weighting of individual documents via token-by-token posterior updates in retrieval-augmented generation, yielding gains on knowledge-based visual QA tasks.
EmbodiedMidtrain mid-trains VLMs on curated VLA-aligned data subsets to improve downstream performance on robot manipulation benchmarks.
C-Mining automatically mines high-fidelity Culture Points from raw multilingual text by treating cross-lingual geometric isolation in embeddings as a quantifiable signal for cultural specificity, then uses them to synthesize better instruction data.
RLSpoofer trains a 4B model on 100 watermarked paraphrase pairs to spoof PF watermarks at 62% success rate, far exceeding baselines trained on up to 10,000 samples.
DEONTICBENCH is a new benchmark of 6,232 deontic reasoning tasks from U.S. legal domains where frontier LLMs reach only ~45% accuracy and symbolic Prolog assistance plus RL training still fail to solve tasks reliably.
ChatSVA achieves 96.12% functional pass rate and 82.5% coverage in SVA generation on 24 RTL designs, delivering 33 percentage point gains and 11x better coverage than prior state-of-the-art.
PR-CAD unifies text-to-CAD generation and editing via progressive refinement with LLMs, a new interaction dataset, and RL-enhanced reasoning to achieve better controllability and faithfulness.
MMSearch-R1 uses reinforcement learning to train multimodal models for on-demand multi-turn internet search with image and text tools, outperforming same-size RAG baselines and matching larger ones while cutting search calls by over 30%.
PuzzleWorld benchmark reveals state-of-the-art AI models solve only 18% of complex puzzlehunt problems with 40% stepwise accuracy, matching novices but trailing enthusiasts, while fine-tuning on traces yields modest gains.
citing papers explorer
-
SARL: Label-Free Reinforcement Learning by Rewarding Reasoning Topology
SARL rewards reasoning topology to improve label-free RL, outperforming baselines with gains up to 44.7% on math and 34.6% on open-ended tasks while maintaining more stable training.
-
FLEX: A Largescale Multimodal, Multiview Dataset for Learning Structured Representations for Fitness Action Quality Assessment
FLEX is the first large-scale multimodal multiview dataset for fitness AQA, featuring RGB, 3D pose, sEMG and physiological data plus a Fitness Knowledge Graph for structured annotations and a VideoQA benchmark.
-
CASS: Nvidia to AMD Transpilation with Data, Models, and Benchmark
CASS supplies the first large verified dataset and trained models for cross-vendor GPU code translation between CUDA and HIP at source and assembly levels, with reported accuracies of 88.2% and 69.1% that beat commercial baselines.
-
Hallucination Self-Play: Bootstrapping Reinforced Detector via Evolved Generator
Hallucination Self-Play co-evolves a generator and detector from one base LLM via RLAIF and RLVR, lifting a 7B model to match advanced LLMs on RAGTruth faithfulness detection.
-
When Does Online Imitation Learning Help in LLM Post-Training? The Role of (Non-)Realizability Beyond Horizon
Online IL overcomes an information-theoretic bottleneck that offline IL faces in non-realizable settings even at horizon 1, under a new structural characterization of reward-relative misspecification.
-
Scaling LLM Knowledge Boundaries via Distribution-Optimized Synthesis
KDoS optimizes synthetic data distributions via a three-stage knowledge density feedback mechanism to maximize LLM knowledge boundary expansion, with the optimal distribution stable across model sizes and data scales.
-
RedAct: Redacting Agent Capability Traces for Procedural Skill Protection
RedAct redacts agent traces to drop normalized skill transfer below the no-skill baseline on CapTraceBench while preserving audit evidence and adding detectable behavioral watermarks.
-
VideoKR: Towards Knowledge- and Reasoning-Intensive Video Understanding
VideoKR supplies 315K knowledge-intensive video reasoning examples and a dedicated benchmark, with experiments indicating post-training gains on reasoning tasks that require both video content and external knowledge.
-
RogueMerge: Robust and Unified Attacks against LLM Model Merging
RogueMerge is a unified attack method that jointly optimizes task vectors to succeed after merging, using stochastic min-max simulation for unknown merging settings and a Taylor-approximated DRO for prompt generalization on generative LLMs.
-
Synthetic Hallucinations, Real Gains: Hard Negatives from Frontier Models for FIM Hallucination Mitigation
Using frontier models to synthesize plausible-but-wrong FIM completions as hard negatives for SFT improves Delulu exact match by +18.8 and edit similarity by +0.22 on Qwen2.5-Coder-7B while also lifting HumanEval-Infilling and SAFIM.
-
Pause and Think: A Dataset and Benchmark for Video-Grounded Assistive Action Suggestion
Introduces pause-and-think-T dataset and pause-and-think-B benchmark; fine-tunes 4B VLM to 58% accuracy matching 235B model while generalizing out-of-distribution.
-
GGBound: A Genome-Grounded Agent for Microbial Life-Boundary Prediction
A genome-conditioned 4B LLM agent predicts microbial life boundaries and matches larger frontier models via token fusion, tool use, and a counterfactual gene-grounding reward.
-
Teaching Language Models to Think in Code
ThinC trains small models to reason primarily in code rather than natural language, outperforming tool-integrated baselines and even larger models on competition math benchmarks.
-
DiagramNet: An End-to-End Recognition Framework and Dataset for Non-Standard System-Level Diagrams
DiagramNet supplies a new multimodal dataset and progressive training pipeline with decoupled multi-agent workflow, allowing a 3B model to outperform GPT-5, Claude-Sonnet-4, and Gemini-2.5-Pro by over 2x on system-level diagram tasks while generalizing to other benchmarks.
-
World2Minecraft: Occupancy-Driven Simulated Scenes Construction
World2Minecraft turns real scenes into Minecraft worlds via occupancy prediction and releases a large indoor occupancy dataset to improve such models.
-
BERAG: Bayesian Ensemble Retrieval-Augmented Generation for Knowledge-based Visual Question Answering
BERAG applies Bayesian ensemble weighting of individual documents via token-by-token posterior updates in retrieval-augmented generation, yielding gains on knowledge-based visual QA tasks.
-
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.
-
C-Mining: Unsupervised Discovery of Seeds for Cultural Data Synthesis via Geometric Misalignment
C-Mining automatically mines high-fidelity Culture Points from raw multilingual text by treating cross-lingual geometric isolation in embeddings as a quantifiable signal for cultural specificity, then uses them to synthesize better instruction data.
-
RLSpoofer: A Lightweight Evaluator for LLM Watermark Spoofing Resilience
RLSpoofer trains a 4B model on 100 watermarked paraphrase pairs to spoof PF watermarks at 62% success rate, far exceeding baselines trained on up to 10,000 samples.
-
DeonticBench: A Benchmark for Reasoning over Rules
DEONTICBENCH is a new benchmark of 6,232 deontic reasoning tasks from U.S. legal domains where frontier LLMs reach only ~45% accuracy and symbolic Prolog assistance plus RL training still fail to solve tasks reliably.
-
ChatSVA: Bridging SVA Generation for Hardware Verification via Task-Specific LLMs
ChatSVA achieves 96.12% functional pass rate and 82.5% coverage in SVA generation on 24 RTL designs, delivering 33 percentage point gains and 11x better coverage than prior state-of-the-art.
-
PR-CAD: Progressive Refinement for Unified Controllable and Faithful Text-to-CAD Generation with Large Language Models
PR-CAD unifies text-to-CAD generation and editing via progressive refinement with LLMs, a new interaction dataset, and RL-enhanced reasoning to achieve better controllability and faithfulness.
-
MMSearch-R1: Incentivizing LMMs to Search
MMSearch-R1 uses reinforcement learning to train multimodal models for on-demand multi-turn internet search with image and text tools, outperforming same-size RAG baselines and matching larger ones while cutting search calls by over 30%.
-
PuzzleWorld: A Benchmark for Multimodal, Open-Ended Reasoning in Puzzlehunts
PuzzleWorld benchmark reveals state-of-the-art AI models solve only 18% of complex puzzlehunt problems with 40% stepwise accuracy, matching novices but trailing enthusiasts, while fine-tuning on traces yields modest gains.
-
RubriQ: Rubric-Guided Group Relative Policy Optimization for Constraint-Aware Quantum Circuit Synthesis
A rubric-guided GRPO pipeline fine-tunes a 7B LLM to synthesize quantum circuits achieving 3.31x T-gate compression with <1% hardware-constraint violations, validated on IBM and IonQ processors.
-
ProMSA:Progressive Multimodal Search Agents for Knowledge-Based Visual Question Answering
ProMSA is a progressive multimodal search agent for KB-VQA that iteratively selects search tools under budgets, trained via rejection-sampling SFT then TN-GSPO RL, reporting gains on E-VQA and InfoSeek over RAG baselines.
-
Empowering GUI Agents via Autonomous Experience Exploration and Hindsight Experience Utilization for Task Planning
PEEU enables a 7B MLLM to reach 30.6% accuracy on GUI task planning by autonomous exploration and hindsight experience synthesis, outperforming a 32B model through stronger high-level OOD generalization.
-
AutoBG: A Board Game Design Assistant with Interactive Ideation, Iterative Rulebook Generation, and Individualized Feedback
AutoBG is a multi-module AI assistant that uses critic-driven iterative refinement on LLM-generated rulebooks, trained on 2.2K rulebooks and 180K reviews, to produce audience-tested designs that outperform GPT-5.4 baselines.
-
JenBridge: Adaptive Long-Form Video Soundtracking across Scene Transitions
JenBridge pretrains a flow-matching Transformer on text-audio data then adapts it with video conditioning and an LLM director to select transitions, claiming better coherence than prior methods on a new LVS benchmark.
-
Starve to Perceive: Taming Lazy Perception in VLMs with Constrained Visual Bandwidth
Constraining visual token budgets during SFT and RL forces VLMs to learn functional active perception, yielding ~5% relative gains and strong transfer to unconstrained evaluation.
-
Enhancing the Code Reasoning Capabilities of LLMs via Consistency-based Reinforcement Learning
CodeThinker improves LLM code reasoning via consistency-based RL with stepwise training data, dynamic beam sampling, and consistency rewards, reaching SOTA on benchmarks with 4.3% gains on Qwen2.5-Coder-7B.
-
Contextualized Code Pretraining for Code Generation
Introduces contextualized code pretraining with caller-callee pairs from static analysis to train CallerGen models that outperform baselines on the new CallerEval benchmark.
-
Chain-of-Procedure: Hierarchical Visual-Language Reasoning for Procedural QA
Introduces ProcedureVQA benchmark and Chain-of-Procedure framework that improves VLM next-step prediction in procedures by up to 13% over baselines.
-
Speculative Interaction Agents: Building Real-Time Agents with Asynchronous I/O and Speculative Tool Calling
Speculative Interaction Agents achieve 1.3-2.2x speedups for real-time tool-calling agents via async I/O decoupling and speculative calls, with clock-based training for small edge models.
-
Optimizer-Model Consistency: Full Finetuning with the Same Optimizer as Pretraining Forgets Less
Full finetuning with the pretraining optimizer reduces forgetting compared to other optimizers or LoRA while achieving comparable new-task performance.
-
Revealing Modular Gradient Noise Imbalance in LLMs: Calibrating Adam via Signal-to-Noise Ratio
MoLS scales Adam updates using module-level SNR estimates to correct gradient noise imbalance and improve LLM training convergence and generalization.
-
Estimating the Black-box LLM Uncertainty with Distribution-Aligned Adversarial Distillation
DisAAD trains a 1%-sized proxy model via adversarial distillation to quantify uncertainty in black-box LLMs by aligning with their output distributions.
-
ScrapMem: A Bio-inspired Framework for On-device Personalized Agent Memory via Optical Forgetting
ScrapMem reports SOTA 51.0% Joint@10 on ATM-Bench with up to 93% memory reduction and 70.3% Recall@10 via optical forgetting and EM-Graph.
-
When Model Editing Meets Service Evolution: A Knowledge-Update Perspective for Service Recommendation
EVOREC integrates locate-then-edit model editing with FA-constrained decoding to improve LLM-based service recommendation under evolution, reporting 25.9% average relative gain in Recall@5 over baselines and 22.3% over fine-tuning in dynamic scenarios.
-
SIEVES: Selective Prediction Generalizes through Visual Evidence Scoring
SIEVES improves selective prediction coverage by up to 3x on OOD VQA benchmarks by training a selector to score the quality of visual evidence produced by reasoner models, generalizing across benchmarks and proprietary models without internal access or per-task retraining.
-
Efficient Rationale-based Retrieval: On-policy Distillation from Generative Rerankers based on JEPA
Rabtriever distills a generative reranker into an efficient bi-encoder using on-policy JEPA to achieve near-reranker accuracy with linear complexity on rationale-based retrieval.
-
CoDA: Towards Effective Cross-domain Knowledge Transfer via CoT-guided Domain Adaptation
CoDA aligns cross-domain latent reasoning representations in LLMs via CoT distillation and MMD to enable effective knowledge transfer without in-domain demonstrations.
-
CodePivot: Bootstrapping Multilingual Transpilation in LLMs via Reinforcement Learning without Parallel Corpora
CodePivot uses Python as a pivot language plus an Aggressive-Partial-Functional RL reward to train a 7B model that outperforms much larger LLMs on multilingual code transpilation without parallel corpora.
-
Characterizing Model-Native Skills
Recovering an orthogonal basis from model activations yields a model-native skill characterization that improves reasoning Pass@1 by up to 41% via targeted data selection and supports inference steering, outperforming human-characterized alternatives.
-
Rethinking On-Policy Distillation of Large Language Models: Phenomenology, Mechanism, and Recipe
On-policy distillation works when student and teacher models share thinking patterns and the teacher adds new capabilities, with success tied to alignment on a small set of high-probability tokens.
-
Pioneer Agent: Continual Improvement of Small Language Models in Production
Pioneer Agent automates the full lifecycle of adapting and continually improving small language models via diagnosis-driven data synthesis and regression-constrained retraining, delivering gains of 1.6-83.8 points on benchmarks and large lifts in production-style tasks.
-
Fundus-R1: Training a Fundus-Reading MLLM with Knowledge-Aware Reasoning on Public Data
Fundus-R1 is a fundus-reading MLLM trained exclusively on public data via RAG-generated reasoning traces and process-reward RLVR, outperforming its base model and a version trained without the traces.
-
Scientific Graphics Program Synthesis via Dual Self-Consistency Reinforcement Learning
SciTikZer-8B uses a new dataset, benchmark, and dual self-consistency RL to generate TikZ code for scientific graphics, outperforming much larger models like Gemini-2.5-Pro.
-
Saliency-R1: Enforcing Interpretable and Faithful Vision-language Reasoning via Saliency-map Alignment Reward
Saliency-R1 uses a novel saliency map technique and GRPO with human bounding-box overlap as reward to improve VLM reasoning faithfulness and interpretability.
-
OpsLLM: Construction of Large Language Model for Software Operations with Multi-stage Learning
OpsLLM's pipeline (HITL data curation, SFT, GRPO RL with a domain process reward model) improves LLM accuracy on software-operations QA and RCA, especially on in-distribution root-cause-analysis tasks.