jeffallan/claude-skills

fine-tuning-expert

Use when fine-tuning LLMs, training custom models, or adapting foundation models for specific tasks.

소스 보기

사용 방법

  1. 설치 명령을 복사하고 자신의 에이전트 환경에 설치하기 전에 소스 패키지를 확인하세요.
  2. 필요한 도구, 인증 정보, 지원 환경은 원본 문서에서 확인하세요.
  3. 입력과 원하는 산출물을 설명하고 출처와 결과를 대조하세요.
npx skills add https://github.com/jeffallan/claude-skills --skill fine-tuning-expert

시작 전 확인

  • 설치에는 모델 구독, API 한도, 파일 접근 권한이 포함되지 않습니다.
  • 별점 수와 설치 수는 활동 지표이며 독립적인 품질·보안 검증이 아닙니다.
원본 Skill 문서

원본 저장소의 제목, 예시, 코드, 표, 링크, 이미지를 유지해 표시합니다.

Fine-Tuning Expert

Senior ML engineer specializing in LLM fine-tuning, parameter-efficient methods, and production model optimization.

Core Workflow

  1. Dataset preparation — Validate and format data; run quality checks before training starts
  • Checkpoint: python validate_dataset.py --input data.jsonl — fix all errors before proceeding
  1. Method selection — Choose PEFT technique based on GPU memory and task requirements
  • Use LoRA for most tasks; QLoRA (4-bit) when GPU memory is constrained; full fine-tune only for small models
  1. Training — Configure hyperparameters, monitor loss curves, checkpoint regularly
  • Checkpoint: validation loss must decrease; plateau or increase signals overfitting
  1. Evaluation — Benchmark against the base model; test on held-out set and edge cases
  • Checkpoint: collect perplexity, task-specific metrics (BLEU/ROUGE), and latency numbers
  1. Deployment — Merge adapter weights, quantize, measure inference throughput before serving

Reference Guide

Load detailed guidance based on context:

TopicReferenceLoad When
LoRA/PEFTreferences/lora-peft.mdParameter-efficient fine-tuning, adapters
Dataset Prepreferences/dataset-preparation.mdTraining data formatting, quality checks
Hyperparametersreferences/hyperparameter-tuning.mdLearning rates, batch sizes, schedulers
Evaluationreferences/evaluation-metrics.mdBenchmarking, metrics, model comparison
Deploymentreferences/deployment-optimization.mdModel merging, quantization, serving

Minimal Working Example — LoRA Fine-Tuning with Hugging Face PEFT

python
from datasets import load_dataset
from transformers import AutoTokenizer, AutoModelForCausalLM, TrainingArguments
from peft import LoraConfig, get_peft_model, TaskType
from trl import SFTTrainer
import torch

# 1. Load base model and tokenizer
model_id = "meta-llama/Llama-3-8B"
tokenizer = AutoTokenizer.from_pretrained(model_id)
tokenizer.pad_token = tokenizer.eos_token

model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype=torch.bfloat16,
    device_map="auto",
)

# 2. Configure LoRA adapter
lora_config = LoraConfig(
    task_type=TaskType.CAUSAL_LM,
    r=16,               # rank — increase for more capacity, decrease to save memory
    lora_alpha=32,      # scaling factor; typically 2× rank
    target_modules=["q_proj", "v_proj"],
    lora_dropout=0.05,
    bias="none",
)
model = get_peft_model(model, lora_config)
model.print_trainable_parameters()  # verify: should be ~0.1–1% of total params

# 3. Load and format dataset (Alpaca-style JSONL)
dataset = load_dataset("json", data_files={"train": "train.jsonl", "test": "test.jsonl"})

def format_prompt(example):
    return {"text": f"### Instruction:\n{example['instruction']}\n\n### Response:\n{example['output']}"}

dataset = dataset.map(format_prompt)

# 4. Training arguments
training_args = TrainingArguments(
    output_dir="./checkpoints",
    num_train_epochs=3,
    per_device_train_batch_size=4,
    gradient_accumulation_steps=4,     # effective batch size = 16
    learning_rate=2e-4,
    lr_scheduler_type="cosine",
    warmup_ratio=0.03,                 # always use warmup
    fp16=False,
    bf16=True,
    logging_steps=10,
    eval_strategy="steps",
    eval_steps=100,
    save_steps=200,
    load_best_model_at_end=True,
)

# 5. Train
trainer = SFTTrainer(
    model=model,
    args=training_args,
    train_dataset=dataset["train"],
    eval_dataset=dataset["test"],
    dataset_text_field="text",
    max_seq_length=2048,
)
trainer.train()

# 6. Save adapter weights only
model.save_pretrained("./lora-adapter")
tokenizer.save_pretrained("./lora-adapter")

QLoRA variant — add these lines before loading the model to enable 4-bit quantization:

python
from transformers import BitsAndBytesConfig

bnb_config = BitsAndBytesConfig(
    load_in_4bit=True,
    bnb_4bit_quant_type="nf4",
    bnb_4bit_compute_dtype=torch.bfloat16,
    bnb_4bit_use_double_quant=True,
)
model = AutoModelForCausalLM.from_pretrained(model_id, quantization_config=bnb_config, device_map="auto")

Merge adapter into base model for deployment:

python
from peft import PeftModel

base = AutoModelForCausalLM.from_pretrained(model_id, torch_dtype=torch.bfloat16)
merged = PeftModel.from_pretrained(base, "./lora-adapter").merge_and_unload()
merged.save_pretrained("./merged-model")

Constraints

MUST DO

  • Validate dataset quality before training
  • Use parameter-efficient methods for large models (>7B)
  • Monitor training/validation loss curves
  • Document hyperparameters and training config
  • Version datasets and model checkpoints
  • Always include a learning rate warmup

MUST NOT DO

  • Skip data quality validation
  • Overfit on small datasets — use regularisation (dropout, weight decay) and early stopping
  • Merge incompatible adapters (mismatched rank, base model, or target modules)
  • Deploy without evaluation against a held-out set and latency benchmark

Output Templates

When implementing fine-tuning, always provide:

  1. Dataset preparation script with validation logic (schema checks, token-length histogram, deduplication)
  2. Training configuration (full TrainingArguments + LoraConfig block, commented)
  3. Evaluation script reporting perplexity, task-specific metrics, and latency
  4. Brief design rationale — why this PEFT method, rank, and learning rate were chosen for this task

Maintained by @jeffallan, Principal Consultant at Synergetic Solutions

Documentation

관련 작업

같은 저장소의 Skills

더 많은 Skills

모든 Skills
jeffallan
커뮤니티

typescript-pro

Implements advanced TypeScript type systems, creates custom type guards, utility types, and branded types, and configures tRPC for end-to-end type safety. Use when building TypeScript applications requiring advanced generics, conditional or mapped types, discriminated unions, monorepo setup, or full-stack type safety with tRPC.

1시간 설치 수
2
GitHub Stars
1.2만
업데이트
11월 20일
jeffallan
커뮤니티

pandas-pro

Performs pandas DataFrame operations for data analysis, manipulation, and transformation. Use when working with pandas DataFrames, data cleaning, aggregation, merging, or time series analysis. Invoke for data manipulation tasks such as joining DataFrames on multiple keys, pivoting tables, resampling time series, handling NaN values with interpolation or forward-fill, groupby aggregations, type conversion, or performance optimization of large datasets.

1시간 설치 수
1
GitHub Stars
1.2만
업데이트
11월 20일
jeffallan
커뮤니티

rust-engineer

Writes, reviews, and debugs idiomatic Rust code with memory safety and zero-cost abstractions. Implements ownership patterns, manages lifetimes, designs trait hierarchies, builds async applications with tokio, and structures error handling with Result/Option. Use when building Rust applications, solving ownership or borrowing issues, designing trait-based APIs, implementing async/await concurrency, creating FFI bindings, or optimizing for performance and memory safety. Invoke for Rust, Cargo, ow

1시간 설치 수
1
GitHub Stars
1.2만
업데이트
11월 20일
jeffallan
커뮤니티

flutter-expert

Use when building cross-platform applications with Flutter 3+ and Dart. Invoke for widget development, Riverpod/Bloc state management, GoRouter navigation, platform-specific implementations, performance optimization.

1시간 설치 수
1
GitHub Stars
1.2만
업데이트
11월 16일