from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

model = AutoModelForCausalLM.from_pretrained(
    "Xen0pp/Smollm3_ml_planner_v2",
    torch_dtype=torch.bfloat16,
    device_map="auto"
)
tokenizer = AutoTokenizer.from_pretrained("Xen0pp/Smollm3_ml_planner_v2")

# Ask about project planning
messages = [
    {"role": "system", "content": "You are an expert ML project planning advisor."},
    {"role": "user", "content": "I want to build a customer churn prediction model. What are the first steps?"}
]

inputs = tokenizer.apply_chat_template(messages, return_tensors="pt", add_generation_prompt=True)
outputs = model.generate(inputs.to(model.device), max_new_tokens=300)
response = tokenizer.decode(outputs[0], skip_special_tokens=True)
print(response)

💡 Example Queries

Project Planning:

"How should I plan a computer vision project for manufacturing quality control?"
"What's a realistic timeline for an NLP sentiment analysis project?"
"I want to build a recommendation system. Where do I start?"

Limited Data Scenarios:

"I only have 100 labeled examples. Can I still build a model?"
"What's the best approach with small datasets?"

Architecture Decisions:

"CNN vs Vision Transformer for my use case?"
"Which recommendation algorithm should I choose?"

📊 Model Details

Base: SmolLM2-360M-Instruct
V1 Training: 6 examples (general ML)
V2 Training: Continued from V1 + 5 project planning examples
Total Expertise: 11 examples
Parameters: 360M
Context Length: 2048 tokens
Training Method: Unsloth (2x faster fine-tuning)

🎯 Best For

ML practitioners planning new projects
Students learning project management
Teams scoping ML initiatives
Anyone needing structured ML guidance

⚠️ Limitations

Small training dataset (for educational purposes)
Should verify critical decisions with domain experts
Timelines and budgets are estimates
Best combined with real-world experience

📈 Comparison

Feature	V1	V2
General ML Concepts	✅	✅
Research Papers	✅	✅
Project Planning	❌	✅
Timeline Estimation	❌	✅
Budget Guidance	❌	✅
Domain-Specific Advice	Limited	✅

📝 License

Apache 2.0

🔗 Links

V1 Model: Xen0pp/Smollm3_720prms
Base Model: HuggingFaceTB/SmolLM2-360M-Instruct

Note: This is a fine-tuned small model for educational purposes. Always validate recommendations with domain expertise.