GLM-5.2 is live. #1 throughput on OpenRouter, pay-per-token on FriendliAI. Try it today ➜
suehuynh
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Run this model inference with full control and performance in your environment.
Model Details
Model Provider
suehuynh
Model Tree
Base
mistralai/Mixtral-8x7B-Instruct-v0.1
Adapter
this model
Input Modalities
Text
Output Modalities
Text
Supported Functionality
Dedicated Endpoints
Container
Marketing-Mixtral-8x7B-v3 is a LoRA fine-tuned version of Mixtral-8x7B-Instruct-v0.1, adapted for marketing copy generation across five task types: ad/social copy, email marketing, product descriptions, brand voice rewriting, and campaign data insights.
This is the second iteration of the Marketing-Mixtral series, trained on an expanded 8,000-row dataset (up from the original 4,300 rows used in Marketing-Mixtral-8x7B v1).
Generate marketing copy from structured briefs across five task types:
Always review generated copy before publication. For product descriptions, verify all claims against the original product specification.
python
from transformers import AutoModelForCausalLM, AutoTokenizerfrom peft import PeftModelimport torchbase_model = AutoModelForCausalLM.from_pretrained("mistralai/Mixtral-8x7B-Instruct-v0.1",torch_dtype=torch.float16,device_map="auto",load_in_4bit=True,)model = PeftModel.from_pretrained(base_model, "suehuynh/Marketing-Mixtral-8x7B-v3")tokenizer = AutoTokenizer.from_pretrained("mistralai/Mixtral-8x7B-Instruct-v0.1")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))
8,000 examples across 5 marketing task types — an expansion of the original 4,300-row dataset with additional sources for table-to-insight, email marketing, and ad copy/brand voice. See suehuynh/marketing-instruct-8k for full documentation.
Note on epochs: an earlier run on this identical dataset and recipe
with n_epochs=3 achieved an 83% win rate. Changing only n_epochs to 1
(holding every other hyperparameter constant) improved the win rate to
87% — a controlled, single-variable result indicating that additional
training epochs on this dataset size leads to overfitting and reduced
generalization against the base model judge.
Evaluated on AutoScientist's held-out internal marketing test set (not publicly available).
Win rate: pairwise LLM-as-judge comparison between finetuned model output and base Mixtral-8x7B-Instruct output on identical prompts.
| Metric | Base Mixtral | Finetuned | Delta |
|---|---|---|---|
| Win rate (AS eval) | 13% | 87% | +569% relative |
This model achieves an 87% win rate against base Mixtral-8x7B-Instruct on AutoScientist's held-out marketing test set, up from 80% in v1 (trained on 4,300 rows, 1 epoch).
The most notable finding from this iteration is a controlled, single-variable experiment: on the same 8,000-row dataset with an identical recipe, training for 3 epochs produced 83% win rate, while training for 1 epoch (with no other change) produced 87%. This isolates epoch count as a direct driver of overfitting at this dataset size — more passes over the data degrade generalization against the base model judge rather than improving it.
BibTeX:
bibtex
@misc{huynh2026marketingmixtralv3,author = {Huynh, Nguyen},title = {Marketing-Mixtral-8x7B-v3: A Fine-tuned Model for Marketing Copy Generation},year = {2026},publisher = {HuggingFace},howpublished = {\url{https://huggingface.co/suehuynh/Marketing-Mixtral-8x7B-v3}},note = {AutoScientist Challenge 2026 — Marketing Category}}
Sue Huynh — Brown University MSc Data Science
HuggingFace: suehuynh