Fine-Tuning
16 pieces on Fine-Tuning, including 4 step-by-step guides.
Guides
Fine-Tune a Quantized Llama on GSM8K with PEFT LoftQ
A step-by-step walkthrough of the Hugging Face PEFT GSM8K LoftQ script: 4-bit NF4 loading, preprocessing, training, and exact-match accuracy.
Fine-Tuning LLMs with LoRA: A Complete Practical Guide to Training & GGUF Deployment
Master end-to-end LoRA fine-tuning for SLMs: synthetic data generation, benchmarking against 7B baselines, lit-gpt training, and GGUF serving with llama.cpp.
Fine-tune a model with GRPO for mathematical reasoning
Train a Qwen2-0.5B model with Group Relative Policy Optimization and LoRA on a single GPU, using format and accuracy reward functions.
How to Analyze and Fine-Tune Agent Reasoning Traces
Learn how to inspect agent reasoning traces, extract tool calls, visualize behavior patterns, and prepare multi-turn agent conversations for supervised fine-tuning.
News & Analysis
Nemotron Scores 535.4 IOI and 30/42 IMO, Passing Gold in Both
NVIDIA reports one Nemotron family reached 535.4/600 on IOI 2026 and 30/42 on IMO 2026, above both gold thresholds.
ThinkingCap-Qwen3.8-27B Cuts Thinking Tokens 37.2% for 0.86pp Accuracy
BottleCap AI's fine-tune of Qwen3.8-27B drops thinking tokens 37.2% across 12 benchmarks, losing just 0.86pp of macro accuracy at xhigh effort.
ToolGrad Hits 99.8% Pass Rate, Lifts Gemma-3-12B to Gemini 2.5 Pro Level
Google Research's ToolGrad flips tool-use data generation: execute APIs first, write the query second. 500 samples take Gemma-3-12B to 83.1 on BFCL.
Adaption Labs Generates Training Data From a Task Description Alone
Adaption Labs' 'Invent a Dataset' creates structured training rows from plain-language descriptions—no seed corpus, schema, or labels required.
LFM2.5-350M Jumps 7 Points on IFStruct in 100 GRPO Steps
100 GRPO training steps on ~500 samples lifts LiquidAI's 350M model from 22.6% to 29.7% on IFStruct — runnable on a free-tier Colab GPU.
Sentence Transformers v6.0 Adds ColBERT Training in 14.5 Hours on One GPU
Sentence Transformers v6.0 adds MultiVectorEncoder with full ColBERT-style training. A single RTX 3090 run beats every zero-shot retriever by 0.06 NDCG@10.
SigLIP LoRA Fine-Tune Cuts Under-Labeling from 9.4% to 3.4%
Alma Media's 23-class image classifier shows why under-labeling rate, not F1, is the right signal for deciding whether LoRA fine-tuning is worth the cost.
Qwen2.5-0.5B Fine-Tuned With DPO After Auditing HH-RLHF Length Bias
A reproducible pipeline audits Anthropic HH-RLHF for lexical shortcuts, then fine-tunes Qwen2.5-0.5B-Instruct with DPO, TRL, and LoRA in one notebook.
Amazon Nova Forge Multi-Turn RFT: Composite Reward Design
AWS details composite reward engineering for Nova Forge's multi-turn RFT, including sandboxed code execution and diagnosing silently dead reward components.
Fine-Tuning Qwen3-0.6B for Tool Calling with XYZ-Aquila-SFT
A reproducible SFT pipeline streams 400 XYZ-Aquila-SFT trajectories, bypasses apply_chat_template to preserve reasoning blocks, and fine-tunes Qwen3-0.6B with LoRA.
General Catalyst Leads $1.1B Round into 2-Month-Old River AI
River AI, founded by xAI co-founder Igor Babuschkin, raises $1.1B to rebuild the AI stack and make agents personally trainable.
The Economics of RAG vs. Fine-Tuning in Enterprise AI
A deep dive into architecture strategy: When should your company use Retrieval-Augmented Generation, and when is it worth paying for Fine-Tuning?