---
pipeline_tag: text-generation
inference: true
widget:
- text: 'def print_hello_world():'
  example_title: Hello world
  group: Python
datasets:
- bigcode/the-stack-v2-train
license: bigcode-openrail-m
library_name: transformers
tags:
- code
model-index:
- name: starcoder2-3b
  results:
  - task:
      type: text-generation
    dataset:
      name: CruxEval-I
      type: cruxeval-i
    metrics:
    - type: pass@1
      value: 32.7
  - task:
      type: text-generation
    dataset:
      name: DS-1000
      type: ds-1000
    metrics:
    - type: pass@1
      value: 25.0
  - task:
      type: text-generation
    dataset:
      name: GSM8K (PAL)
      type: gsm8k-pal
    metrics:
    - type: accuracy
      value: 27.7
  - task:
      type: text-generation
    dataset:
      name: HumanEval+
      type: humanevalplus
    metrics:
    - type: pass@1
      value: 27.4
  - task:
      type: text-generation
    dataset:
      name: HumanEval
      type: humaneval
    metrics:
    - type: pass@1
      value: 31.7
  - task:
      type: text-generation
    dataset:
      name: RepoBench-v1.1
      type: repobench-v1.1
    metrics:
    - type: edit-smiliarity
      value: 71.19
quantized_by: bartowski
---

## Exllama v2 Quantizations of starcoder2-3b

Using <a href="https://github.com/turboderp/exllamav2/releases/tag/v0.0.19">turboderp's ExLlamaV2 v0.0.19</a> for quantization.

<b>The "main" branch only contains the measurement.json, download one of the other branches for the model (see below)</b>

Each branch contains an individual bits per weight, with the main one containing only the meaurement.json for further conversions.

Conversion was done using the default calibration dataset.

Default arguments used except when the bits per weight is above 6.0, at that point the lm_head layer is quantized at 8 bits per weight instead of the default 6.

Original model: https://huggingface.co/bigcode/starcoder2-3b


<a href="https://huggingface.co/bartowski/starcoder2-3b-exl2/tree/8_0">8.0 bits per weight</a>

<a href="https://huggingface.co/bartowski/starcoder2-3b-exl2/tree/6_5">6.5 bits per weight</a>

<a href="https://huggingface.co/bartowski/starcoder2-3b-exl2/tree/5_0">5.0 bits per weight</a>

<a href="https://huggingface.co/bartowski/starcoder2-3b-exl2/tree/4_25">4.25 bits per weight</a>

<a href="https://huggingface.co/bartowski/starcoder2-3b-exl2/tree/3_5">3.5 bits per weight</a>


## Download instructions

With git:

```shell
git clone --single-branch --branch 6_5 https://huggingface.co/bartowski/starcoder2-3b-exl2
```

With huggingface hub (credit to TheBloke for instructions):

```shell
pip3 install huggingface-hub
```

To download the `main` (only useful if you only care about measurement.json) branch to a folder called `starcoder2-3b-exl2`:

```shell
mkdir starcoder2-3b-exl2
huggingface-cli download bartowski/starcoder2-3b-exl2 --local-dir starcoder2-3b-exl2 --local-dir-use-symlinks False
```

To download from a different branch, add the `--revision` parameter:

Linux:

```shell
mkdir starcoder2-3b-exl2-6_5
huggingface-cli download bartowski/starcoder2-3b-exl2 --revision 6_5 --local-dir starcoder2-3b-exl2-6_5 --local-dir-use-symlinks False
```

Windows (which apparently doesn't like _ in folders sometimes?):

```shell
mkdir starcoder2-3b-exl2-6.5
huggingface-cli download bartowski/starcoder2-3b-exl2 --revision 6_5 --local-dir starcoder2-3b-exl2-6.5 --local-dir-use-symlinks False
```