---
license: apache-2.0
library_name: transformers
pipeline_tag: text-generation
language:
  - en
base_model: google/gemma-2-2b-it
tags:
  - dystrio
  - sculpt
  - pruned
  - compressed
  - efficient
  - dense
  - runtime-agnostic
  - no-custom-kernels
  - hf-drop-in
  - drop-in-replacement
  - smaller
  - faster
  - gemma
datasets:
  - wikitext
model-index:
  - name: Dystrio Sculpt (gemma-2-2b-it Default)
    results:
      - task:
          type: text-generation
        dataset:
          name: WikiText-103 (validation)
          type: wikitext
        metrics:
          - name: perplexity
            type: perplexity
            value: 20.5854
          - name: ppl_ratio
            type: ppl_ratio
            value: 0.7985
---

# dystrio/gemma-2-2b-it-sculpt-default

> **9% smaller, quality improved (0.7985x PPL), drop-in replacement. No custom kernels. No runtime changes.**

Dystrio Sculpt structurally compresses transformer models, producing dense models that load with standard `transformers` — no custom code, no new ops, no deployment friction.

This is the **Default** tier of [gemma 2 2b it](https://huggingface.co/google/gemma-2-2b-it).

## Quick Start

```python
from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("dystrio/gemma-2-2b-it-sculpt-default", torch_dtype="bfloat16", device_map="auto")
tokenizer = AutoTokenizer.from_pretrained("dystrio/gemma-2-2b-it-sculpt-default")

inputs = tokenizer("The future of AI inference is", return_tensors="pt").to(model.device)
outputs = model.generate(**inputs, max_new_tokens=100)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))
```

## Benchmark Results

All tiers compiled from [gemma 2 2b it](https://huggingface.co/google/gemma-2-2b-it) on A100 80GB, bf16:

| Model | PPL | PPL Ratio | Weights (GB) | Chat Prefill TPS | RAG TTFT p95 (ms) | Decode TPS |
|-------|-----|-----------|-------------|------------------|-------------------|------------|
| **Baseline** | 25.7807 | 1.0 | 4.869591 | 21611.9 | 70.251 | 59.6 |
| **sculpt-default** | 20.5854 | 0.7985 | 4.441124 | 23065.3 | 69.007 | 60.0 |
| **sculpt-production** | 22.4118 | 0.8693 | 4.226891 | 23404.0 | 66.554 | 60.7 |
| **sculpt-throughput** | 29.8372 | 1.1573 | 3.969811 | 24330.1 | 64.529 | 59.3 |
| **sculpt-experimental** | 48.9699 | 1.8995 | 3.412804 | 26496.2 | 60.97 | 59.5 |

### Key Metrics (this model)

| Metric | Value |
|--------|-------|
| **Weights memory** | 4.441124 GB (9% smaller) |
| **PPL ratio** | 0.7985 |
| **Chat prefill TPS** | 23065.3 (+7%) |
| **RAG TTFT p95** | 69.007 ms (-2%) |
| **Decode TPS** | 60.0 (flat) |
| **Parameters** | 2.38B |

## All Sculpt Tiers

| Tier | HuggingFace | Size | PPL Ratio | Use Case |
|------|-------------|------|-----------|----------|
| default | [dystrio/gemma-2-2b-it-sculpt-default](https://huggingface.co/dystrio/gemma-2-2b-it-sculpt-default) 👈 **this model** | 4.441124 GB | 0.7985 | Zero-regret: quality preserved, smaller footprint |
| production | [dystrio/gemma-2-2b-it-sculpt-production](https://huggingface.co/dystrio/gemma-2-2b-it-sculpt-production) | 4.226891 GB | 0.8693 | Practical savings with modest quality tradeoff |
| throughput | [dystrio/gemma-2-2b-it-sculpt-throughput](https://huggingface.co/dystrio/gemma-2-2b-it-sculpt-throughput) | 3.969811 GB | 1.1573 | Maximum usable compression for speed/edge |
| experimental | [dystrio/gemma-2-2b-it-sculpt-experimental](https://huggingface.co/dystrio/gemma-2-2b-it-sculpt-experimental) | 3.412804 GB | 1.8995 | Boundary exploration, maximum structural compression |

## What is Dystrio Sculpt?

Dystrio Sculpt compiles transformer models into smaller, faster variants. Output models:

- Are **dense** (not sparse) — standard architecture, fewer parameters
- Load with **standard HuggingFace Transformers** — no custom code needed
- Require **no custom kernels** and **no runtime changes**
- Work as a one-step compile before deployment
- Stack with quantization (AWQ, GPTQ, GGUF) for compound savings

## Compatibility

- ✅ HuggingFace Transformers
- ✅ vLLM
- ✅ TGI (Text Generation Inference)
- ✅ llama.cpp / GGUF conversion
- ✅ AWQ / GPTQ quantization
- ✅ Any framework that loads standard safetensors

## Benchmark Environment

- **GPU**: NVIDIA A100-SXM4-80GB
- **dtype**: bf16
- **Torch**: 2.10.0+cu128
- **Transformers**: 5.3.0
- **Deterministic**: True
- Single-GPU, standard HuggingFace Transformers, no custom kernels.

## Metric Definitions

- **PPL ratio**: WikiText-103 perplexity relative to baseline. <1.0 = quality improved.
- **Prefill TPS**: Tokens per second during prompt encoding (higher = faster).
- **TTFT p95**: Time to first token at 95th percentile (lower = faster).
- **Decode TPS**: Tokens per second during generation (higher = faster).
- **Weights (GB)**: Model parameter memory (deterministic, runtime-independent).

## Citation

```bibtex
@misc{dystrio_sculpt_2026,
  title={Dystrio Sculpt: Structural Compilation for Transformer LLMs},
  author={Dystrio},
  year={2026},
  url={https://huggingface.co/dystrio}
}
```

## Downstream Benchmarks (lm-eval)

Evaluated with [lm-eval-harness](https://github.com/EleutherAI/lm-evaluation-harness) on A100-80GB, bf16, zero-shot.

| Benchmark | Baseline | This Model | Delta |
|-----------|:--------:|:----------:|:-----:|
| ARC-Challenge | 0.5094 | 0.4386 | -0.0708 |
| HellaSwag | 0.5375 | 0.5032 | -0.0343 |
| MMLU | 0.5691 | 0.5016 | -0.0675 |
| TruthfulQA MC2 | 0.5322 | 0.5242 | -0.0080 |