| license: apache-2.0 | |
| base_model: unsloth/gemma-4-E2B-it | |
| tags: | |
| - datasense | |
| - cudf | |
| - rapids | |
| - code-generation | |
| - gemma | |
| pipeline_tag: text-generation | |
| # DataSense GPU — Full Merged Model | |
| Fine-tuned data-science agent. Generates cuDF + cuML code from natural language. | |
| LoRA adapter fused into base weights — no PEFT dependency. | |
| ## Load | |
| ```python | |
| from unsloth import FastLanguageModel | |
| model, tokenizer = FastLanguageModel.from_pretrained( | |
| 'sanjaymalladi/DataSense-GPU-Full', max_seq_length=4096, load_in_4bit=True | |
| ) | |
| FastLanguageModel.for_inference(model) | |
| ``` | |
| ## Benchmark | |
| - 215× exec speedup on Random Forest risk model (81s CPU → 0.38s GPU) | |
| - 58.9× rolling window at 20M rows | |
| - Tested on Kaggle Tesla T4, cuDF 26.02, 181K real BigQuery rows | |