{
  "id": "stack-profile:llm-fine-tuning",
  "_kind": "StackProfile",
  "_file": "domain/stack-profiles/deep-stacks-1.yaml",
  "_cluster": "domain",
  "attributes": {
    "displayName": "LLM Fine-Tuning Stack (PyTorch, HuggingFace, PEFT/LoRA, W&B, vLLM)",
    "description": "A specialized stack for adapting large language models to domain-specific\ntasks through parameter-efficient fine-tuning. PyTorch provides the\ntraining runtime. HuggingFace Transformers supplies pre-trained model\nweights, tokenizers, and the Trainer API. PEFT (Parameter-Efficient\nFine-Tuning) with LoRA adapters enables fine-tuning billion-parameter\nmodels on consumer or single-node GPU hardware by training only a\nsmall fraction of weights.\n\nWeights & Biases (W&B) tracks training runs, hyperparameters, loss\ncurves, and evaluation metrics. vLLM provides high-throughput inference\nwith PagedAttention for deploying the fine-tuned model. Python is the\nsole language across the pipeline. The key tradeoff is that LoRA\nadapters trade some quality ceiling for dramatically lower compute\ncost; full fine-tuning on large models still requires multi-GPU clusters.\n",
    "composes": [
      "library:pytorch",
      "library:hf-transformers",
      "tool:vllm",
      "language:python",
      "tool:huggingface"
    ]
  },
  "outgoingEdges": [
    {
      "from": "stack-profile:llm-fine-tuning",
      "to": "library:pytorch",
      "kind": "composed_of"
    },
    {
      "from": "stack-profile:llm-fine-tuning",
      "to": "library:hf-transformers",
      "kind": "composed_of"
    },
    {
      "from": "stack-profile:llm-fine-tuning",
      "to": "tool:vllm",
      "kind": "composed_of"
    },
    {
      "from": "stack-profile:llm-fine-tuning",
      "to": "language:python",
      "kind": "composed_of"
    },
    {
      "from": "stack-profile:llm-fine-tuning",
      "to": "tool:huggingface",
      "kind": "composed_of"
    },
    {
      "from": "stack-profile:llm-fine-tuning",
      "to": "tool:docker",
      "kind": "composed_of"
    },
    {
      "from": "stack-profile:llm-fine-tuning",
      "to": "tool:kubernetes",
      "kind": "composed_of"
    },
    {
      "from": "stack-profile:llm-fine-tuning",
      "to": "role:ml-engineer",
      "kind": "used_by_role"
    },
    {
      "from": "stack-profile:llm-fine-tuning",
      "to": "role:research-engineer",
      "kind": "used_by_role"
    },
    {
      "from": "stack-profile:llm-fine-tuning",
      "to": "role:data-scientist",
      "kind": "used_by_role"
    },
    {
      "from": "stack-profile:llm-fine-tuning",
      "to": "workflow:model-training-cycle",
      "kind": "follows_workflow"
    },
    {
      "from": "stack-profile:llm-fine-tuning",
      "to": "workflow:hyperparameter-tuning-cycle",
      "kind": "follows_workflow"
    },
    {
      "from": "stack-profile:llm-fine-tuning",
      "to": "domain:machine-learning",
      "kind": "applies_to"
    },
    {
      "from": "stack-profile:llm-fine-tuning",
      "to": "domain:ml-ai",
      "kind": "applies_to"
    },
    {
      "from": "stack-profile:llm-fine-tuning",
      "to": "skill-area:ml-fine-tuning",
      "kind": "requires_skill_area"
    },
    {
      "from": "stack-profile:llm-fine-tuning",
      "to": "skill-area:deep-learning-libraries",
      "kind": "requires_skill_area"
    },
    {
      "from": "stack-profile:llm-fine-tuning",
      "to": "skill-area:machine-learning-frameworks",
      "kind": "requires_skill_area"
    },
    {
      "from": "stack-profile:llm-fine-tuning",
      "to": "skill-area:model-serving-deployment",
      "kind": "requires_skill_area"
    },
    {
      "from": "stack-profile:llm-fine-tuning",
      "to": "skill-area:llm-infrastructure",
      "kind": "requires_skill_area"
    }
  ],
  "incomingEdges": []
}