
Base model	meta-llama/Llama-3.2-3B-Instruct
Parameters	3.21B
Architecture	GQA, 28 layers
Quantization	GPTQ 4-bit, group_size=128
Model size	2.2 GB
VRAM required	~2.8 GB (inference)

Quantization Details

Parameter	Value
Method	GPTQ
Tool	gptqmodel 5.8.0
Bits	4
Group size	128
Scheme	Symmetric (4-bit, INT32 packing)
Calibration dataset	allenai/c4 (en, shard 1 of 1024)
Calibration samples	128
Seed	42
Quantization time	644s
Hardware	NVIDIA RTX 4080 Laptop (12 GB) via Docker

Why Self-Quantized?

Pre-quantized checkpoints on HuggingFace typically have unknown calibration provenance â€” the dataset, sample count, seed, and group size are rarely documented. This checkpoint was self-quantized with controlled, documented settings to enable rigorous cross-method comparison (GGUF k-quant vs AWQ vs GPTQ) in a NeurIPS 2026 submission on quality-safety correlation under quantization.

Evaluation Results

Evaluated on 735 quality samples across 7 tasks and 468 safety samples judged by gemma3:12b.

Quality Metrics (generation tasks)

Metric	Score
BERTScore (F1)	0.767
ROUGE-L	0.660
Coherence	0.782

Accuracy (capability tasks)

Task	Accuracy
MMLU	53.0%
ARC Challenge	66.5%
Classification	12.0%

Safety Metrics (gemma3:12b judge)

Metric	Score
Refusal Rate (AdvBench)	79.0%
Truthfulness (TruthfulQA)	28.0%
Unbiased Rate (BBQ)	45.5%

Other Quantization Formats

Format	Repository
AWQ 4-bit	Crusadersk/llama3.2-3b-awq-4bit
Original FP16	meta-llama/Llama-3.2-3B-Instruct

Prompt Template

<|begin_of_text|><|start_header_id|>user<|end_header_id|>

{prompt}<|eot_id|><|start_header_id|>assistant<|end_header_id|>

Usage

python

from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained(
    "Crusadersk/llama3.2-3b-gptq-4bit",
    device_map="auto",
)
tokenizer = AutoTokenizer.from_pretrained("Crusadersk/llama3.2-3b-gptq-4bit")

inputs = tokenizer("What is the capital of France?", return_tensors="pt").to(model.device)
outputs = model.generate(**inputs, max_new_tokens=64, do_sample=False)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

Inference requirements: pip install gptqmodel (Linux only) or optimum+auto-gptq

Windows users: GPTQ inference requires gptqmodel which only builds on Linux. Use Docker or WSL2. See reproduction instructions below.

Compatibility

Framework	Supported
Transformers	Yes
vLLM	Yes (GPTQ backend)
llama.cpp	No (use GGUF format instead)
Ollama	No (use GGUF format instead)
Windows (native)	No â€” requires Linux/Docker

Reproduction

The full quantization pipeline â€” Dockerfiles, quantization scripts, and a 766-line engineering log documenting every platform failure and solution â€” is available at:

research/tr142/expansion/

in the Banterhearts repository. Key files:

File	Purpose
`QUANTIZATION_LOG.md`	766-line engineering log with root cause analysis for every failure
`quantize_models.py`	CLI for AWQ + GPTQ quantization with skip-existing and manifests
`Dockerfile.gptq` / `Dockerfile.awq`	Separate Docker images (irreconcilable dependency conflict)
`smoke_test.py`	Checkpoint verification with automatic Docker fallback for GPTQ
`run_hf_eval.py`	HuggingFace .generate() evaluation backend

Citation

bibtex

@misc{banterhearts2026llama323bgptq,
  title = {Self-Quantized Llama 3.2 3B Instruct (GPTQ 4-bit) for Quality-Safety Correlation Research},
  author = {Kadadekar, Sahil},
  year = {2026},
  url = {https://huggingface.co/Crusadersk/llama3.2-3b-gptq-4bit},
  note = {Part of the Banterhearts research program. NeurIPS 2026 submission.}
}

Acknowledgments

This work is part of the Chimera/Banterhearts technical-report program on deployment-time LLM behavior, quantization, refusal robustness, batching effects, and inference-stack reliability. Canonical public archive: Chimeraforge Reports; source context: github.com/Sahil170595/Banterhearts.