cuba6112/skillfactory - 第 2 页

torch-compile

Optimize PyTorch with torch.compile (TorchDynamo/Inductor), focusing on compile overhead, graph breaks, and benchmark methodology. Use when speeding up PyTorch models or debugging compile behavior; triggers: torch.compile, torchdynamo, inductor, graph break,…

原文语言：英语

torchaudio

Audio signal processing library for PyTorch. Covers feature extraction (spectrograms, mel-scale), waveform manipulation, and GPU-accelerated data augmentation techniques. (torchaudio, melscale, spectrogram, pitchshift, specaugment, waveform, resample)

原文语言：英语

torchserve

Model serving engine for PyTorch. Focuses on MAR packaging, custom handlers for preprocessing/inference, and management of multi-GPU worker scaling. (torchserve, mar-file, handler, basehandler, model-archiver, inference-api)

原文语言：英语

torchtext

Natural Language Processing utilities for PyTorch (Legacy). Includes tokenizers, vocabulary building, and DataPipe-based dataset handling for text processing pipelines. (torchtext, tokenizer, vocab, datapipe, regextokenizer, nlp-pipeline)

原文语言：英语

torchvision

Computer vision library for PyTorch featuring pretrained models, advanced image transforms (v2), and utilities for handling complex data types like bounding boxes and masks. (torchvision, transforms, tvtensor, resnet, cutmix, mixup, pretrained models, vision…

原文语言：英语

unsloth-core

Core fundamentals of Unsloth for fast LLM fine-tuning, covering FastLanguageModel setup, optimized gradient checkpointing, and native inference acceleration (triggers: unsloth, FastLanguageModel, from_pretrained, get_peft_model, for_inference, gradient…

原文语言：英语

unsloth-cpt

Strategies for continued pretraining and domain adaptation in Unsloth (triggers: continued pretraining, CPT, domain adaptation, lm_head, embed_tokens, rsLoRA, embedding_learning_rate).

原文语言：英语

unsloth-datasets

Standardizing and formatting datasets for Unsloth, including chat template conversion and synthetic data generation (triggers: chat templates, ShareGPT, Alpaca, conversation_extension, add_new_tokens, standardize_sharegpt, formatting_prompts_func).

原文语言：英语

unsloth-dpo

Direct Preference Optimization (DPO) for aligning models with preference data without separate reward models. Triggers: dpo, preference optimization, rlhf, ref_model=none, patchdpotrainer, dpotrainer.

原文语言：英语

unsloth-fft

Performing full fine-tuning (FFT) in Unsloth with 100% exact weight updates and optimized gradient checkpointing. Triggers include fft, full fine-tuning, full_finetuning, exact fine-tuning, and weight updates.

原文语言：英语

unsloth-gguf

Exporting fine-tuned models to GGUF format for deployment in llama.cpp, Ollama, and local serving tools. Triggers: gguf, quantization export, llama.cpp, ollama, save_pretrained_gguf, modelfile.

原文语言：英语

unsloth-grpo

Implementation of Group Relative Policy Optimization (GRPO) for training reasoning models, optimized for 8x memory savings (triggers: GRPO, reasoning, DeepSeek-R1, reinforcement learning, RLVR, GRPOTrainer, thinking tokens).

原文语言：英语

unsloth-inference

Deploying fine-tuned models for production inference using native kernel optimization, vLLM, or SGLang. Triggers: inference, serving, vllm, sglang, for_inference, model merging, openai api.

原文语言：英语

unsloth-long-context

Training models on extended context lengths using optimized RoPE scaling and memory-efficient attention kernels. Triggers: long context, max_seq_length, rope scaling, large context window, flex attention.

原文语言：英语

unsloth-lora

Configuring and optimizing 16-bit Low-Rank Adaptation (LoRA) and Rank-Stabilized LoRA (rsLoRA) for efficient LLM fine-tuning using triggers like lora, qlora, rslora, rank selection, lora_alpha, lora_dropout, and target_modules.

原文语言：英语

unsloth-models

Guidance on selecting and configuring supported model architectures like Llama 4, DeepSeek-R1, and Qwen3. Triggers: llama 4, deepseek-r1, qwen3, gemma 3, model selection, instruct vs base.

原文语言：英语

unsloth-orpo

One-step preference alignment using Odds Ratio Preference Optimization (ORPO) (triggers: ORPO, preference optimization, alignment, ORPOTrainer, log_odds_ratio, binary preference).

原文语言：英语

unsloth-qlora

Advanced 4-bit quantization techniques using Unsloth and BitsAndBytes for extreme VRAM efficiency (triggers: QLoRA, 4-bit, load_in_4bit, bnb-4bit, VRAM optimization, dynamic quantization).

原文语言：英语

unsloth-quantization

Utilizing Dynamic 4-bit quantization, FP8 training, and 8-bit optimizers to minimize VRAM usage without sacrificing accuracy. Triggers: quantization, dynamic 4-bit, fp8, bitsandbytes, adamw_8bit, qat.

原文语言：英语

unsloth-sft

Supervised fine-tuning using SFTTrainer, instruction formatting, and multi-turn dataset preparation with triggers like sft, instruction tuning, chat templates, sharegpt, alpaca, conversation_extension, and SFTTrainer.

原文语言：英语

unsloth-stt

Fine-tuning Speech-to-Text models like Whisper using Unsloth's optimized LoRA pipeline. Triggers: stt, whisper, transcription, audio fine-tuning, speech-to-text, audio normalization.

原文语言：英语

unsloth-tts

Fine-tuning Text-to-Speech (TTS) models with Unsloth for voice cloning and synthetic speech (triggers: TTS, text-to-speech, voice cloning, Orpheus-TTS, audio fine-tuning, speech synthesis).

原文语言：英语

unsloth-vision

Fine-tuning multimodal vision-language models (Llama 3.2 Vision, Qwen2.5 VL) using optimized vision layers (triggers: vision models, multimodal, Llama 3.2 Vision, Qwen2.5 VL, UnslothVisionDataCollator, finetune_vision_layers).

原文语言：英语

vector-databases

Design vector database ingestion and retrieval pipelines (points + payloads, filtered similarity search, multi-stage hybrid retrieval, index maintenance). Use when building RAG/vector search flows or debugging retrieval quality; triggers: vector database,…

原文语言：英语