Llama-3.2-3B-Instruct-sculpt-default/README.md

license, library_name, pipeline_tag, language, base_model, tags, datasets, model-index

license	library_name	pipeline_tag	language	base_model	tags	datasets	model-index
apache-2.0	transformers	text-generation	en	meta-llama/Llama-3.2-3B-Instruct	dystrio sculpt pruned compressed efficient dense runtime-agnostic no-custom-kernels hf-drop-in drop-in-replacement smaller faster llama	wikitext	name results Dystrio Sculpt (Llama-3.2-3B-Instruct Default) task dataset metrics type text-generation name type WikiText-103 (validation) wikitext name type value perplexity perplexity 17.1627 name type value ppl_ratio ppl_ratio 0.9678

dystrio/Llama-3.2-3B-Instruct-sculpt-default

7% smaller, quality improved (0.9678x 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 Llama 3.2 3B Instruct.

Quick Start

from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("dystrio/Llama-3.2-3B-Instruct-sculpt-default", torch_dtype="bfloat16", device_map="auto")
tokenizer = AutoTokenizer.from_pretrained("dystrio/Llama-3.2-3B-Instruct-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 Llama 3.2 3B Instruct on A100 80GB, bf16:

Model	PPL	PPL Ratio	Weights (GB)	Chat Prefill TPS	RAG TTFT p95 (ms)	Decode TPS
Baseline	17.7333	1.0	5.984213	20742.1	75.219	74.7
sculpt-default	17.1627	0.9678	5.553549	21777.1	71.177	74.7
sculpt-production	21.554	1.2155	5.307455	22728.2	70.16	72.7
sculpt-throughput	26.9519	1.5198	4.999838	23116.0	69.412	72.3
sculpt-experimental	37.844	2.1341	4.446127	25457.5	68.204	73.1

Key Metrics (this model)

Metric	Value
Weights memory	5.553549 GB (7% smaller)
PPL ratio	0.9678
Chat prefill TPS	21777.1 (+5%)
RAG TTFT p95	71.177 ms (-5%)
Decode TPS	74.7 (flat)
Parameters	2.98B

All Sculpt Tiers

Tier	HuggingFace	Size	PPL Ratio	Use Case
default	dystrio/Llama-3.2-3B-Instruct-sculpt-default 👈 this model	5.553549 GB	0.9678	Zero-regret: quality preserved, smaller footprint
production	dystrio/Llama-3.2-3B-Instruct-sculpt-production	5.307455 GB	1.2155	Practical savings with modest quality tradeoff
throughput	dystrio/Llama-3.2-3B-Instruct-sculpt-throughput	4.999838 GB	1.5198	Maximum usable compression for speed/edge
experimental	dystrio/Llama-3.2-3B-Instruct-sculpt-experimental	4.446127 GB	2.1341	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

@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 on A100-80GB, bf16, zero-shot.

Benchmark	Baseline	This Model	Delta
ARC-Challenge	0.4360	0.3737	-0.0623
HellaSwag	0.5329	0.4971	-0.0358
MMLU	0.6223	0.5272	-0.0951
TruthfulQA MC2	0.5138	0.4625	-0.0513