Wizard-Vicuna-13B-Uncensored-SpQR
Overview
This model is an SpQR 4-bit quantization of the original Wizard-Vicuna-13B-Uncensored-HF
Quantization Specifications
- Quantization: 4-bit, group size of 16, per-channel with scale and zero-point of 3 bits.
- Outliers: Threshold set at 0.2.
- Permutation Order:
act_order. - Dampening: Set at 1e0.
- Sampling: 128 samples.
- Logging: Via Weights & Biases.
Evaluation Metrics
The following perplexity scores were obtained on various datasets:
| Dataset | Perplexity |
|---|---|
| c4 | 7.354 |
| wikitext2 | 5.685 |
| ptb | 20.822 |
- Downloads last month
- 8
This model does not have enough activity to be deployed to Inference API (serverless) yet.
Increase its social visibility and check back later, or deploy to Inference Endpoints (dedicated)
instead.
Evaluation results
- perplexity on C4self-reported7.354
- perplexity on WikiText-2self-reported5.685
- perplexity on PTBself-reported20.822