supervised-fine-tuning

Learn How To Observe, Manage, and Scale, Agentic AI Apps Using Azure AI Foundry - with this hands-on workshop

Code for SFT and RL

[ICML 2025] Official code for the paper "RoSTE: An Efficient Quantization-Aware Supervised Fine-Tuning Approach for Large Language Models"

LoRA fine-tuning pipeline for tool-calling chat LLMs with config-driven datasets, deterministic prompts, and built-in tool-call evaluation.

[ICML 2025] Official code for the paper "RoSTE: An Efficient Quantization-Aware Supervised Fine-Tuning Approach for Large Language Models"

Open-source research engineering project for building the end-to-end post-training stack for reasoning language models, including SFT, preference learning, RLHF/RLVR, evaluation, inference-time scaling, and scalable systems for frontier-level reasoning.

Automatic music tagging using foundation models

🎯 Fine-tuning LLMs using LlamaFactory for financial intent understanding | Evaluating open-source models on OpenFinData benchmark | Full implementation with multiple models (Qwen2.5/ChatGLM3/Baichuan2/Llama3)

Fine-tuned Meta's LLaMA 3.2 1B for text summarization using QLoRA (4-bit quantization + LoRA), achieving 40%+ improvement in ROUGE-2 over the base model on CNN/DailyMail dataset.

Least-data SFT on Qwen2.5-0.5B-Instruct: teach a ticket-card format with 8 examples. Colab T4 notebook + LoRA. Open weights.

Fine-tune Qwen2.5-VL-7B with LoRA to predict human-rated emotion intensity (1–7) from images, with a ResNet18 regression baseline, full preprocessing/SFT pipeline, and evaluation (MAE/RMSE + bias analysis).

End-to-end 35.5M-parameter small language model in medical domain built from scratch in PyTorch with retention-aware continual pretraining, masked SFT, exact resume, and sealed evaluation.

End-to-end Supervised Fine-Tuning (SFT) pipeline for TinyLlama-1.1B-Chat, specialized in trademark similarity risk assessment using heuristic-labeled SFT data, CPU-only LoRA training, adapter validation, full-weight merge, GGUF export, quantization (Q4_K_M), and local inference deployment via llama.cpp.

Fine-tuning various Llama 3.1 family of models on the Mult-It dataset

STaR Self-Taught Reasoner implementation on GSM8K — Zero-Shot CoT vs Vanilla SFT vs STaR with Llama 3.2-3B

Compact TensorFlow language model for Election Commission of India (ECI) domain pretraining and assistant-masked SFT.

MixTIR 面向工具增强推理：在同一套 Agent 里同时使用 Python 代码解释器和网页搜索，用混合轨迹做监督微调（SFT），再对训练后的模型做端到端评测。

Supervised fine-tuning (SFT) of Qwen3 for structured medical reasoning QA — teaching models to "think before answering" (<think>...</think>). Supports full fine-tuning & LoRA, with an end-to-end data→train→eval→compare pipeline and quantitative benchmarks (PPL, format compliance, semantic similarity, latency/throughput).

Supervised Fine Tuning with QLoRA