Modelpublic

thlurte/FastData-LM-3B-SFT

sourceHugging Faceapache-2.0updated 1mo agoView on Hugging Face

0likes9downloads

Model Card

FastData-LM-3B-SFT

FastData-LM-3B-SFT is a specialized language model fine-tuned on the `thlurte/VSG-lite-1.5k` dataset. It functions as a zero-shot High-Throughput Vectorized Data Compiler — translating dataset schema requirements into 100% vectorized, loop-free NumPy and Pandas execution graphs.

Model Details

—Base Architecture: `Qwen/Qwen2.5-Coder-3B-Instruct`
—Fine-Tuning Dataset: `thlurte/VSG-lite-1.5k`
—Primary Domain: Vectorized Synthetic Data Generation across 13 industrial schemas.
—Strict Anti-Loop Constraint: Eliminates .apply(), .iterrows(), and all for/while row-level iteration.

Usage Example

python

from unsloth import FastLanguageModel

model, tokenizer = FastLanguageModel.from_pretrained(
    model_name = "thlurte/FastData-LM-3B-SFT",
    max_seq_length = 2048,
    load_in_4bit = True,
)

messages = [
    {"role": "system", "content": "You are a zero-shot Python Data Compiler. Generate 100% vectorized NumPy/Pandas code."},
    {"role": "user", "content": "Generate a synthetic bank ledger dataset with entry_id, balance, and transaction_type."}
]

prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = tokenizer(prompt, return_tensors="pt").to("cuda")
outputs = model.generate(**inputs, max_new_tokens=1024)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))