| --- |
| title: Praveen · Open Model Builder |
| emoji: 🧪 |
| colorFrom: orange |
| colorTo: yellow |
| license: apache-2.0 |
| --- |
| |
| # Praveen |
|
|
| I build practical, inspectable tools around open models: adapters, quantization workflows, evaluation surfaces, and small demos that make model behavior easier to understand. |
|
|
| ## Featured work |
|
|
| ### [PathPack-Q · LFM2.5-2.6B training-free quantization](https://huggingface.co/praveenkumarpranjal/LFM2.5-2.6B-4bit-PathPack-Q) |
|
|
| An architecture-specific post-training quantization experiment for LiquidAI's |
| hybrid convolution/attention LLM. PathPack-Q uses exact gated-path channel |
| permutations to improve which weights share each 4-bit quantization group, |
| without training, text calibration data, extra parameters, or runtime operators. |
|
|
| - **5.30% lower perplexity** than byte-matched uniform MLX 4-bit on 8,160 held-out WikiText-2 tokens |
| - **9.87% lower KL divergence** from the BF16 teacher on a fixed prompt suite |
| - **Identical 1,517,616,892-byte checkpoint** and 4.501 effective bits/weight |
| - Includes the search algorithm, complete-path acceptance gate, rejected-layer evidence, checkpoint builder, and machine-readable evaluations |
|
|
| ### [ScopeGuard · Qwen2.5-1.5B LoRA](https://huggingface.co/praveenkumarpranjal/scopeguard-qwen2.5-1.5b-lora) |
|
|
| A locally trained agent decision-layer adapter that turns natural-language requests into strict JSON risk and confirmation decisions before tools execute. |
|
|
| - **93% risk accuracy** on a 100-example held-out split, up from 66% for the base model |
| - **100% exact schema compliance**, up from 84% |
| - 3.957M trainable parameters — only 0.256% of the 1.5B base model |
| - Includes adapter weights, original dataset, deterministic generator, training config, baseline outputs, and per-example evaluation |
|
|
| → [Explore the ScopeGuard dataset](https://huggingface.co/datasets/praveenkumarpranjal/scopeguard-decisions) |
|
|
| → [Open the complete benchmark explorer](https://huggingface.co/spaces/praveenkumarpranjal/scopeguard-benchmark) |
|
|
| ### [LoRA Lens](https://huggingface.co/spaces/praveenkumarpranjal/lora-lens-demo) |
|
|
| An in-browser audit tool for `adapter_config.json` files. It surfaces rank, alpha, scaling, target modules, reproducibility gaps, and conservative parameter-efficiency estimates without uploading weights or requiring an API key. |
|
|
| → [Inspect the implementation](https://huggingface.co/praveenkumarpranjal/lora-lens) |
|
|
| ## Current lab |
|
|
| - LoRA and PEFT adapter design |
| - Quantization and memory-aware inference |
| - Reproducible model cards and evaluation tooling |
| - Human-readable demos for technical work |
|
|
| I prefer falsifiable, transparent experiments with clear limits over opaque claims. |
|
|
