Instructions to use suehuynh/Marketing-Llama-3.3-70B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use suehuynh/Marketing-Llama-3.3-70B with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="suehuynh/Marketing-Llama-3.3-70B")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("suehuynh/Marketing-Llama-3.3-70B")
model = AutoModelForCausalLM.from_pretrained("suehuynh/Marketing-Llama-3.3-70B")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=40)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use suehuynh/Marketing-Llama-3.3-70B with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "suehuynh/Marketing-Llama-3.3-70B"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "suehuynh/Marketing-Llama-3.3-70B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/suehuynh/Marketing-Llama-3.3-70B

SGLang

How to use suehuynh/Marketing-Llama-3.3-70B with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "suehuynh/Marketing-Llama-3.3-70B" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "suehuynh/Marketing-Llama-3.3-70B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "suehuynh/Marketing-Llama-3.3-70B" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "suehuynh/Marketing-Llama-3.3-70B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use suehuynh/Marketing-Llama-3.3-70B with Docker Model Runner:
```
docker model run hf.co/suehuynh/Marketing-Llama-3.3-70B
```

Marketing-Llama-3.3-70B

Fine-tuned Llama-3.3-70B-Instruct on a curated marketing instruction dataset using AutoScientist for the AutoScientist Challenge 2026 — Marketing Category.

Model Details

Model Demo

Watch the Model Output Demo to see the generation quality.

Model Description

Marketing-Llama-3.3-70B is a LoRA fine-tuned version of Llama-3.3-70B-Instruct, adapted for marketing copy generation across five task types: ad/social copy, email marketing, product descriptions, brand voice rewriting, and campaign data insights.

The base model (Llama-3.3-70B) was selected by AutoScientist's recipe recommendation when trained on an expanded 13,000-row dataset. Sibling models, Marketing-Mixtral-8x7B-v2, Marketing-Mixtral-8x7B-v3, was trained on a subset of this data (8,000 rows) on Mixtral-8x7B-Instruct for comparison (87% win rate).

Developed by: Sue Huynh
Model type: Causal LM, LoRA fine-tuned
Language: English
License: Llama 3.3 Community License
Finetuned from: meta-llama/Llama-3.3-70B-Instruct
Training platform: AutoScientist by Adaption Labs

Model Sources

Dataset: suehuynh/marketing-instruct-13k
Related model: Marketing-Mixtral-8x7B-v3 (8K-row subset of this dataset, Mixtral-8x7B base, 87% win rate)
Recipes: suggested by AutoScientist Challenge 2026 — adaptionlabs.ai

Uses

Direct Use

Generate marketing copy from structured briefs across five task types:

Ad/Social copy: platform-specific ads with audience and CTA constraints
Email marketing: campaign, lifecycle, and transactional emails with subject + preview + body format
Product descriptions: grounded in provided attribute lists only
Brand voice rewriting: neutral copy rewritten to match a specified voice
Table-to-insight: campaign performance data summarized with recommendations

Out-of-Scope Use

Not intended for strategy generation, market research, or competitive analysis
Not suitable for medical, legal, or financial copywriting without additional fine-tuning
Product descriptions should always be reviewed for factual accuracy before publication
Subject to Meta's Llama 3.3 Community License Agreement — review usage restrictions before commercial deployment

Bias, Risks, and Limitations

Training data is primarily English-language and Western marketing contexts
Despite faithfulness-filtering on product description examples, all generated copy should be human-reviewed before publication
Generated copy should not be published without editorial review
As a 70B model, inference requires significantly more compute than the sibling Mixtral-8x7B model (~13B active parameters); consider the Mixtral version for resource-constrained deployment

Recommendations

Always review generated copy before publication. For product descriptions, verify all claims against the original product specification.

How to Get Started with the Model

from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel
import torch

base_model = AutoModelForCausalLM.from_pretrained(
    "meta-llama/Llama-3.3-70B-Instruct",
    torch_dtype=torch.float16,
    device_map="auto",
    load_in_4bit=True,
)
model = PeftModel.from_pretrained(base_model, "suehuynh/Marketing-Llama-3.3-70B")
tokenizer = AutoTokenizer.from_pretrained("meta-llama/Llama-3.3-70B-Instruct")

prompt = "Write a product description for an online store listing..."
inputs = tokenizer.apply_chat_template(
    [{"role": "user", "content": prompt}],
    tokenize=True, add_generation_prompt=True, return_tensors="pt"
).to(model.device)

output = model.generate(inputs, max_new_tokens=200, do_sample=False)
print(tokenizer.decode(output[0][inputs.shape[1]:], skip_special_tokens=True))

Note: Llama 3.3 is a gated model. Request access at meta-llama/Llama-3.3-70B-Instruct before loading.

Training Details

Training Data

13,000 examples across 5 marketing task types: the 8,000-row dataset used for the Mixtral sibling model, plus an additional 4,000 product description examples. See suehuynh/marketing-instruct-13k for full documentation.

Training Procedure

Training Hyperparameters

Training regime: bf16 mixed precision
Algorithm: LoRA (Supervised Fine-Tuning)
LoRA rank: 64
LoRA alpha: 128
LoRA dropout: 0
Target modules: all-linear
Learning rate: 1e-4
LR scheduler: cosine, warmup ratio 0.05
Epochs: 1
Weight decay: 0.01
Gradient clipping: 1.0
Train on inputs: false

This recipe was recommended by AutoScientist's automatic configuration based on the dataset characteristics; the base model (Llama-3.3-70B, rather than Mixtral-8x7B used in earlier iterations) was also an AutoScientist recommendation for this dataset size and composition.

Speeds, Sizes, Times

Training platform: AutoScientist by Adaption Labs
Base model size: 70B parameters

Evaluation

Testing Data, Factors & Metrics

Testing Data

Evaluated on AutoScientist's held-out internal marketing test set (not publicly available).

Metrics

Win rate: pairwise LLM-as-judge comparison between finetuned model output and base Llama-3.3-70B-Instruct output on identical prompts.

Results

Metric	Base Llama 3.3	Finetuned	Delta
Win rate (AS eval)	12%	88%	+633% relative

Summary

This model achieves an 88% win rate against base Llama-3.3-70B-Instruct on AutoScientist's held-out marketing test set, trained on the largest and most diverse dataset in this series (13,000 rows). Because the dataset, base model, and LoRA rank all changed simultaneously relative to the sibling Mixtral model, this result should be read as the strongest overall outcome in the series rather than an isolated ablation — see Marketing-Mixtral-8x7B-v3's model card for a controlled, single-variable finding on the effect of epoch count.

Environmental Impact

Hardware: AutoScientist cloud infrastructure
Cloud Provider: Adaption Labs

Technical Specifications

Model Architecture and Objective

Base: Llama-3.3-70B-Instruct
Adaptation: LoRA adapters on all linear layers
Objective: Supervised fine-tuning on marketing instruction data

Citation

BibTeX:

@misc{huynh2026marketingllama,
  author = {Huynh, Nguyen},
  title = {Marketing-Llama-3.3-70B: A Fine-tuned Model for Marketing Copy Generation},
  year = {2026},
  publisher = {HuggingFace},
  howpublished = {\url{https://huggingface.co/suehuynh/Marketing-Llama-3.3-70B}},
  note = {AutoScientist Challenge 2026 — Marketing Category}
}

Model Card Authors

Sue Huynh — Brown University MSc Data Science

Downloads last month: 192

Model tree for suehuynh/Marketing-Llama-3.3-70B

Base model

meta-llama/Llama-3.1-70B

Finetuned

meta-llama/Llama-3.3-70B-Instruct

Adapter

(288)

this model

suehuynh
/

Marketing-Llama-3.3-70B