Skip to main content
Version: 1.7.0

Sparse Transferring LLMs

This guide focuses on adapting large language models (LLMs) to new domains and tasks using sparse transfer learning. By leveraging a pre-sparsified model's structure, you can efficiently fine-tune on new data, leading to reduced hyperparameter tuning, training times, and computational costs.

You'll learn how to:

  • Adapt to New Domains: Transfer knowledge from a pre-sparsified LLM to excel in particular domains or use cases.
  • Quantize for Efficiency: Apply one-shot quantization to the model to further optimize for efficient inference.

Prerequisites​

  • Training Environment: A system that meets the minimum hardware and software requirements as outlined in the Install Guide.
  • SparseML LLM Installation: An environment with DeepSparse for LLMs installed as outlined in the Install Guide.
  • Background: Familiarity with Generative AI and working with large language models is recommended.

Sparse Transferring a Llama Model​

We'll use a pre-sparsified 7b Llama 2 instruction tuning model from the SparseZoo fine-tuned on the OpenPlatypus dataset. The model is referenced by the following SparseZoo stub:

zoo:llama2-7b-open_platypus_orca_llama2_pretrain-pruned60

Data Preparation​

A dataset is required for transferring the pre-sparsified model's structure. For this example, we'll use the Helpful Instructions dataset from the Hugging Face H4 team, which is available in the Hugging Face dataset hub and can be loaded as follows:

from datasets import load_dataset

dataset = load_dataset("HuggingFaceH4/helpful_instructions")

For comprehensive data preparation guidelines, including formats like CSV and JSONL, refer to our detailed datasets guide.

Sparse Transfer​

Creating sparse models from scratch can be time-consuming and computationally expensive. Sparse transfer learning, however, allows you to leverage pre-sparsified models to adapt to new domains and tasks efficiently without extensive training. As in the one-shot and fine-tuning guides, we'll use a recipe and the compress command in SparseML to the pre-sparsified model from the SparseZoo.

The code below demonstrates fine-tuning the model while preserving the sparsity and at the end, one-shot quantization to the Llama model utilizing a recipe:

from sparseml.transformers import (
    SparseAutoModelForCausalLM, SparseAutoTokenizer, load_dataset, compress
)

model = SparseAutoModelForCausalLM.from_pretrained(
    "zoo:llama2-7b-open_platypus_orca_llama2_pretrain-pruned60",
    device_map="auto",
)
tokenizer = SparseAutoTokenizer.from_pretrained(
    "zoo:llama2-7b-open_platypus_orca_llama2_pretrain-pruned60"
).to(model.device)
dataset = load_dataset("HuggingFaceH4/helpful_instructions")

def format_data(data):
    return {
        "text": data["prompt"] + data["completion"]
    }

dataset = dataset.map(format_data)

recipe = """
finetune_stage:
    run_type: train
    finetune_modifiers:
        ConstantPruningModifier:
            targets: ["re:.*q_proj.weight", "re:.*k_proj.weight", "re:.*v_proj.weight", "re:.*o_proj.weight", "re:.*gate_proj.weight", "re:.*up_proj.weight", "re:.*down_proj.weight"]
            start: 0

stage_quantization:
    run_type: oneshot
    quantize_modifiers:
        QuantizationModifier:
            ignore: [LlamaRotaryEmbedding, LlamaRMSNorm, SiLUActivation, MatMulOutput_QK, MatMulOutput_PV]
            post_oneshot_calibration: true
            scheme_overrides:
                Linear:
                    weights:
                        num_bits: 8
                        symmetric: true
                        strategy: channel
                MatMulLeftInput_QK:
                    input_activations:
                        num_bits: 8
                        symmetric: true
                Embedding:
                    input_activations: null
                    weights:
                        num_bits: 8
                        symmetric: false
        SparseGPTModifier:
            quantize: True
            targets: [model.layers.0, model.layers.1, model.layers.2, model.layers.3, model.layers.4, model.layers.5, model.layers.6, model.layers.7, model.layers.8, model.layers.9, model.layers.10, model.layers.11, model.layers.12, model.layers.13, model.layers.14, model.layers.15, model.layers.16, model.layers.17, model.layers.18, model.layers.19, model.layers.20, model.layers.21, model.layers.22, model.layers.23, model.layers.24, model.layers.25, model.layers.26, model.layers.27, model.layers.28, model.layers.29, model.layers.30, model.layers.31, lm_head]
"""

compress(
    model=model,
    tokenizer=tokenizer,
    dataset=dataset,
    recipe=recipe,
    output_dir="./transfer-example"
)

After running the above code, the model is transferred to the Helpful Instructions dataset with 60% sparsity and quantization, resulting in a smaller model ready for efficient inference.

Inference​

Evaluating Accuracy

Evaluating the model's accuracy is important to ensure it meets the desired performance requirements. To do so, we can use the following code to evaluate the model's perplexity on a sample dataset:

from sparseml import evaluate

eval = evaluate(
    "./transfer-example/stage_quantization",
    datasets="openai_humaneval",
    integration="perplexity",
    text_column_name=["prompt", "canonical_solution"]
)
print(eval)

After sparsifying the model, it is ready for evaluation and deployment. To test the model's generation capabilities, we can use the following code to generate text utilizing PyTorch:

from sparseml.transformers import SparseAutoModelForCausalLM, SparseAutoTokenizer

model_path = "./transfer-example/stage_quantization"
model = SparseAutoModelForCausalLM.from_pretrained(model_path, device_map="auto")
tokenizer = SparseAutoTokenizer.from_pretrained(model_path).to(model.device)
inputs = tokenizer(["Large language models are"], return_tensors="pt")
generated_ids = model.generate(**inputs)
outputs = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)
print(outputs)

The above code, however, does not leverage the sparsity within the model for efficient inference. To do so, we need to export the model to ONNX to be ready for efficient inference on CPUs with DeepSparse. SparseML provides a simple export command to do so:

from sparseml import export

export(
    "./transfer-example/stage_quantization",
    task="text-generation",
    sequence_length=1024,
    target_path="./exported"
)

The exported model located at ./exported can now be used for efficient inference with DeepSparse. To do so, sub in the exported model within the previous Getting Started - Deploy guide for your desired deployment method.

Want to dive into more about one-shot sparsification with Neural Magic? Here are a few paths to consider:

Content

Actions

Edit this page

Support

  • Community Slack Enterprise Support

    • Issues

  • DeepSparse
  • SparseML
  • SparseZoo