---
base_model: meta-llama/Llama-2-13b-hf
license: llama2
tags:
  - kronq
  - quantization
  - group-quantization
  - fake-quant
  - fp16
language:
  - en
pipeline_tag: text-generation
---
# Llama-2-13b — KronQ W3A16 g128 (fake-quant fp16)

> ⚠️ **Fake-quant fp16 checkpoint.** 3-bit group-128 weights stored in **fp16** (KronQ does not pack int3) — **same size as bf16**, for PPL/accuracy reproduction. For deployable low-bit see the W4A16-g128 / W2A16-g128 (packed) repos.

[Llama-2-13b](https://huggingface.co/meta-llama/Llama-2-13b-hf) quantized to 3-bit weights (group 128) with **KronQ**, exported as a standard fp16 model.

## Results (WikiText-2, seqlen 2048)

**Perplexity:** **5.135**

**Zero-shot accuracy:**

| PIQA | ARC-E | ARC-C | HellaSwag | WinoGrande | BoolQ | OBQA | Average |
|---|---|---|---|---|---|---|---|
| 79.22 | 76.30 | 48.72 | 77.53 | 72.14 | 81.62 | 44.60 | **68.59** |

(lm-evaluation-harness 0-shot.)

## Usage

Loads as a **standard fp16 model** (no KronQ code):

```python
from transformers import AutoModelForCausalLM
m = AutoModelForCausalLM.from_pretrained("donghyunli/Llama-2-13b-KronQ-W3A16-g128-fake", torch_dtype="float16", device_map="auto")
```

## Recipe

Group-128 asymmetric W3, weight-only, `--alpha 0.25`, `--act_order`, BiIP, raw H_G.

## License

Derivative of Llama-2-13b — [llama2 license](https://ai.meta.com/llama/license/).