Tagged articles
supervised fine-tuning
22 articles · Page 1 of 1
ByteDance and university researchers introduce Axion 1.0, an agent training method using just two-stage supervised fine-tuning and reinforcement learning, achieving 59% on Humanity's Last Exam and 62.2% on SEAL-0 with fully open-sourced data and code.
Axion 1.0ByteDanceReinforcement Learning
0 likes · 12 min read

A community 9B model combining DeepSeek V4 Pro reasoning with Qwen3.5-9B via distillation and refusal adjustment reached #1 on Hugging Face, showing higher science accuracy, 36% fewer reasoning tokens, and near-zero refusal rates.
DeepSeek V4 ProGGUF quantizationGSPO
0 likes · 9 min read

ByteDance releases Axion 1.0, a 122B parameter model achieving 66.7 on FrontierScience Research benchmark, surpassing GPT-5.6 Sol's 63.3, using only two supervised fine-tuning stages and reinforcement learning on Qwen base models, with 35B weights open-sourced under Apache 2.0.
Axion 1.0ByteDanceFrontierScience Research
0 likes · 5 min read

NV-CoT replaces discrete tokenized bounding boxes with continuous numerical coordinate prediction via four lightweight linear heads and a Laplace policy for reinforcement learning, raising region localization IoU from 47.3 to 59.5 and improving high-resolution VQA accuracy on V* Bench and HR-Bench under both supervised fine-tuning and GRPO.
Continuous CoordinatesMultimodal Large ModelsNV-CoT
0 likes · 11 min read

Alibaba's Qwen-Audio-Jev integrates Jev decision heads into an audio LLM, enabling direct classification, binary judgment, and scoring without autoregressive decoding, delivering ~20x inference speedup across 18 benchmarks including speaker verification (6.5% EER) and audio quality assessment.
Jev decision headsQwen-Audio-JevRLCD
0 likes · 14 min read

The article analyzes a recent paper that introduces a generative reasoning re‑ranker for LLM‑driven recommendation, detailing its SFT and RL training pipeline, semantic‑ID embedding, target vs. reject sampling strategies, and experimental gains of 2.4% Recall@5 and 1.3% NDCG@5 over the OneRec‑Think baseline.
Generative ReasoningLLMRe-ranking
0 likes · 9 min read

This article systematically explains the post‑training pipeline for large language models, covering supervised fine‑tuning, RLHF, PPO, GRPO, RLVR, DPO and emerging Agentic RL, while illustrating each method with analogies, detailed workflows, tables, and recent research findings.
DPOGRPOLLM
0 likes · 24 min read

The article analyzes why reinforcement learning (RL) fine‑tuning retains a large language model's general abilities better than supervised fine‑tuning (SFT), explaining the off‑policy distribution shift of SFT and the on‑policy data consistency, KL penalty, and trust‑region mechanisms that give RL its anti‑forgetting properties.
Catastrophic ForgettingLLMOn-Policy Data
0 likes · 8 min read

This article examines Tencent's large language model applications across diverse business scenarios, detailing core use cases such as content generation, intelligent customer service, and role‑playing, and explains the three key technologies—Supervised Fine‑Tuning, Retrieval‑Augmented Generation, and Agents—that enable these capabilities.
AI applicationsAgentLLM
0 likes · 4 min read

This article examines Tencent's large language model deployments across diverse business scenarios, detailing core use cases such as content generation, intelligent customer service, and role‑playing, while explaining the underlying technologies of Supervised Fine‑Tuning, Retrieval‑Augmented Generation, and Agent systems.
AI applicationsAgentRAG
0 likes · 4 min read

This article analyzes a recent paper that unifies supervised fine‑tuning (SFT) and reinforcement learning (RL) for large language models under a single gradient estimator, introduces the Unified Policy Gradient Estimator (UPGE) and the Hybrid Post‑Training (HPT) algorithm, and demonstrates their superior performance on math reasoning benchmarks.
AI researchHybrid TrainingLLM
0 likes · 11 min read

The article analyzes the poor generalization of supervised fine‑tuning (SFT) for large language models, reveals its gradient as a high‑variance inverse‑probability policy gradient, proposes a one‑line Dynamic Fine‑Tuning correction, and shows substantial gains on challenging math and offline RL benchmarks.
Dynamic Fine-TuningGeneralizationLLM Alignment
0 likes · 7 min read

This article introduces TickIt, a ByteBrain system that leverages large language models to automatically identify and escalate critical Oncall tickets, detailing its multi‑class escalation, deduplication, and category‑guided fine‑tuning modules, experimental results, and the operational impact on cloud services.
Cloud OperationsIncident ManagementLLM
0 likes · 13 min read


AIWalkerMay 6, 2025 · Artificial Intelligence
SimpleAR demonstrates that a vanilla autoregressive model with only 0.5 B parameters can generate high‑fidelity 1024×1024 images, covering pretraining, supervised fine‑tuning, and reinforcement learning, achieving competitive GenEval (0.59) and DPG‑Bench (79.66) scores while reducing inference time to about 14 seconds with vLLM and KV‑cache optimizations.
PretrainingReinforcement Learningautoregressive
0 likes · 14 min read

The article breaks down the three core stages of large‑language‑model training—pre‑training, supervised fine‑tuning, and RLHF—explaining their purpose, methods, and concrete examples while noting DeepSeek‑R1’s recent breakthrough and its implications for AI development.
AI TrainingDeepSeekLLM
0 likes · 5 min read


ArchitectMar 16, 2025 · Artificial Intelligence
This article walks through the complete lifecycle of building a small large‑language model, covering token‑level inference, pre‑training, post‑training steps such as supervised fine‑tuning, reward‑model creation, and reinforcement‑learning methods like DPO, PPO and GRPO, culminating in a practical 0.5B model fine‑tuned for chain‑of‑thought reasoning.
GRPOLLM trainingReinforcement Learning
0 likes · 22 min read

The paper introduces Critique Fine-Tuning (CFT), a method that replaces simple imitation in supervised fine‑tuning with critique‑based learning, achieving superior reasoning performance on mathematical benchmarks using only 50 K samples, outperforming traditional reinforcement‑learning approaches that require millions of examples.
AI ReasoningCritique Fine-TuningMathematical Benchmarks
0 likes · 7 min read

A recent study by Fei‑Fei Li’s team shows that using supervised fine‑tuning on the open‑source Qwen2.5‑32B‑Instruct model can replicate and even surpass the reasoning abilities of OpenAI’s o1‑preview at a fraction of the computational cost, demonstrating a cheap yet powerful approach to large‑language‑model development.
budget forcingcost-effective AIlarge-language-models
0 likes · 6 min read

Bilibili’s in‑house role‑playing large language model, built on the Index architecture and refined through pre‑training, supervised fine‑tuning, and preference optimization (PPO and DPO), achieved top scores on the Chinese CharacterEval benchmark, surpassing rivals while incorporating safety alignment and showcasing consistent, personality‑driven dialogue examples.
Pretrainingcontent safetyevaluation benchmark
0 likes · 13 min read


NewBeeNLPMar 27, 2024 · Artificial Intelligence
This article provides a comprehensive technical overview of Meta's Llama 2 series, covering its architectural upgrades such as Group Query Attention, the pre‑training dataset and hyper‑parameters, loss behavior, benchmark comparisons, and the supervised fine‑tuning pipeline with safety considerations.
AILlama 2Pretraining
0 likes · 11 min read

This article provides a comprehensive English overview of Meta's Llama 2 family, describing the model sizes, pre‑training data, architectural improvements, supervised fine‑tuning, reinforcement learning with human feedback, safety evaluations, reward‑model training, and iterative optimization techniques used to produce the high‑performing Llama 2‑Chat models.
Llama 2Open‑sourceRLHF
0 likes · 33 min read

The article explains how techniques such as Reinforcement Learning from Human Feedback, Instruction Fine‑Tuning, Supervised Fine‑tuning and Chain‑of‑Thought contribute to ChatGPT’s impressive general‑intelligence performance, as analyzed by DeepLearning.AI founder Andrew Ng.
ChatGPTDeepLearning.AIReinforcement Learning from Human Feedback
0 likes · 2 min read
