overthink-jupyter

Direct REST API operations for Ollama using the requests library. Covers all /api/* endpoints for model management, text generation, chat completion, embeddings, and streaming responses.

Direct Preference Optimization for learning from preference pairs. Covers DPOTrainer, preference dataset preparation, implicit reward modeling, and beta tuning for stable preference learning without explicit reward models. Includes thinking quality patterns.

LLM evaluation and prompt optimization with Evidently.ai. Covers text descriptors, dataset metrics, LLM-as-a-Judge patterns, and automated prompt optimization for classification and generation tasks.

Model fine-tuning with PyTorch and HuggingFace Trainer. Covers dataset preparation, tokenization, training loops, TrainingArguments, SFTTrainer for instruction tuning, evaluation, and checkpoint management. Includes Unsloth recommendations.

GPU monitoring and performance metrics for Ollama inference. Check GPU status, VRAM usage, loaded models, and inference performance metrics like tokens per second.

Group Relative Policy Optimization for reinforcement learning from human feedback. Covers GRPOTrainer, reward function design, policy optimization, and KL divergence constraints for stable RLHF training. Includes thinking-aware reward patterns.

Import GGUF models from HuggingFace into Ollama. Pull models directly using the hf.co/ prefix, track download progress, and use imported models for inference.

Fast inference with Unsloth and vLLM backend. Covers model loading, fast_generate(), thinking model output parsing, and memory management for efficient inference.

LangChain framework for LLM applications. Covers model wrappers (HuggingFace, Ollama), prompt templates, few-shot learning, output parsing, and chaining techniques for building sophisticated LLM workflows.

Official ollama Python library for LLM inference. Provides a clean, Pythonic interface for text generation, chat completion, embeddings, model management, and streaming responses.

OpenAI compatibility layer for Ollama. Use the official OpenAI Python library to interact with Ollama, enabling easy migration from OpenAI and compatibility with LangChain, LlamaIndex, and other OpenAI-based tools.

Parameter-efficient fine-tuning with LoRA and Unsloth. Covers LoraConfig, target module selection, QLoRA for 4-bit training, adapter merging, and Unsloth optimizations for 2x faster training.

Advanced QLoRA experiments and comparisons. Covers alpha scaling, LoRA rank selection, target module strategies, continual learning, multi-adapter hot-swapping, and quantization comparison (4-bit vs BF16).

Model quantization for efficient inference and training. Covers precision types (FP32, FP16, BF16, INT8, INT4), BitsAndBytes configuration, memory estimation, and performance tradeoffs.

Retrieval-Augmented Generation (RAG) for grounding LLM responses with external knowledge. Covers document chunking, embeddings, vector stores (pandas, ChromaDB), similarity search, and conversational RAG pipelines.

Reward model training for RLHF pipelines. Covers RewardTrainer, preference dataset preparation, sequence classification heads, and reward scaling for stable reinforcement learning. Includes thinking quality scoring patterns.

Reinforcement Learning with Leave-One-Out estimation for policy optimization. Covers RLOOTrainer, reward function integration, baseline estimation, and variance reduction techniques for stable RL training. Includes thinking-aware patterns.

Supervised Fine-Tuning with SFTTrainer and Unsloth. Covers dataset preparation, chat template formatting, training configuration, and Unsloth optimizations for 2x faster instruction tuning. Includes thinking model patterns.

Transformer architecture fundamentals. Covers self-attention mechanism, multi-head attention, feed-forward networks, layer normalization, and residual connections. Essential concepts for understanding LLMs.

Vision model fine-tuning with FastVisionModel. Covers Pixtral, Ministral VL training, UnslothVisionDataCollator, image+text datasets, and vision-specific LoRA configuration.