You can deploy this model and start earning money today!

nellaep/
ThePeoplesFinance-ParamsGen-TinyLlama-LoRA

Name: ThePeoplesFinance-ParamsGen-TinyLlama-LoRA
Availability: InStock

safetensors

finance

entrepreneurship

parameters-generation

Views

No views yet

ThePeoplesFinance — Parameters-Only Generator (TinyLlama + LoRA)

This model is a parameter generator designed to produce realistic financial and personal profiles without assigning a score. It was trained using TinyLlama-1.1B-Chat with LoRA fine-tuning in 8-bit quantization for efficiency.

📌 Model Purpose

ThePeoplesFinance-ParamsGen generates structured JSON objects with the following fields:

age (int)
dependents (int)
monthlyIncome (float, 2 decimals)
monthlyBills (float, 2 decimals)
monthlyEntertainment (float, 2 decimals)
savingAmount (float, 2 decimals)
assets (float, 2 decimals)
salesSkills (int, 1–9)
riskLevel (int, 1–9)
confidence (int, 1–9)
businessDifficulty (int, 1–9)

👉 Unlike other models in this project, this generator does not predict an “entrepreneurial readiness” score. It focuses purely on producing clean, realistic parameter sets that can later be scored or analyzed.

🏗️ Training Details

Base model: TinyLlama/TinyLlama-1.1B-Chat-v1.0
Method: LoRA fine-tuning (r=16, alpha=32, target modules = Q/K/V/O projections)
Quantization: 8-bit with bitsandbytes
Data: Custom dataset nellaep/ThePeoplesFinance
- ~1,000 rows of synthetic but realistic financial/person profiles
- Profiles exclude the readiness score column
Epochs: 4
Batch size: Effective batch 32 (per_device_train_batch_size=8, grad_accum=4)
Learning rate: 1.5e-4
Optimizer: adamw_torch
Loss: Causal LM loss on JSON serialization

🔎 Usage

python

from transformers import AutoTokenizer, AutoModelForCausalLM
import torch, re, json

model_id = "nellaep/ThePeoplesFinance-ParamsGen-TinyLlama-LoRA"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id, device_map="auto")

SYSTEM_HEADER = (
    "Task: Generate ONE realistic finance profile as compact JSON (no markdown, no comments).\n"
    "Keys: age, dependents, monthlyIncome, monthlyBills, monthlyEntertainment,\n"
    "      savingAmount, assets, salesSkills, riskLevel, confidence, businessDifficulty.\n"
    "Rules:\n"
    "- Use integers for age, dependents, salesSkills, riskLevel, confidence, businessDifficulty.\n"
    "- Use two decimals for money fields.\n"
    "- Do NOT include any score or extra keys.\n"
    "- Output JSON only."
)

prompt = SYSTEM_HEADER + "\nAnswer:"

inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
with torch.inference_mode():
    gen = model.generate(
        **inputs,
        max_new_tokens=160,
        do_sample=True,
        temperature=0.8,
        top_p=0.9,
        eos_token_id=tokenizer.eos_token_id
    )

text = tokenizer.decode(gen[0], skip_special_tokens=True)
print("Raw output:", text)

# Extract JSON
m = re.search(r"\{.*\}", text, flags=re.DOTALL)
if m:
    profile = json.loads(m.group(0))
    print("Parsed profile:", profile)

🎯 Example Outputs

Unconditional generation

json

{
  "age": 32,
  "dependents": 2,
  "monthlyIncome": 7200.00,
  "monthlyBills": 3100.00,
  "monthlyEntertainment": 450.00,
  "savingAmount": 12000.00,
  "assets": 25000.00,
  "salesSkills": 6,
  "riskLevel": 5,
  "confidence": 7,
  "businessDifficulty": 3
}

Conditional generation

Prompt condition: "age between 25 and 40; prefer 2 dependents; higher savingAmount; monthlyIncome around 6000-10000; entertainment moderate"

json

{
  "age": 29,
  "dependents": 2,
  "monthlyIncome": 8900.00,
  "monthlyBills": 3200.00,
  "monthlyEntertainment": 600.00,
  "savingAmount": 20000.00,
  "assets": 40000.00,
  "salesSkills": 7,
  "riskLevel": 6,
  "confidence": 8,
  "businessDifficulty": 4
}

⚠️ Limitations

Outputs are synthetic and intended for testing/training downstream models.
The model does not verify economic realism (e.g., ratios between bills and income).
Occasional malformed JSON may appear; regex extraction + post-validation is recommended.
Not intended for financial advice or real-world credit/risk assessment.

📚 Citation

If you use this model in research or applications, please cite:

@misc{nellaep2025paramsgen,
  title = {ThePeoplesFinance — Parameters-Only Generator (TinyLlama + LoRA)},
  author = {Barnes, Darnell (nellaep)},
  year = {2025},
  publisher = {Hugging Face},
  howpublished = {\url{https://huggingface.co/nellaep/ThePeoplesFinance-ParamsGen-TinyLlama-LoRA}}
}

✅ License

This model is released under the same license as TinyLlama (Apache 2.0).
Data used for training is synthetic and does not represent real individuals.