File size: 2,182 Bytes
f84a69c 97e9e9c f84a69c 490e5bd f84a69c 97e9e9c 33860e4 97e9e9c 2527f0c ce5343d 97e9e9c |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 |
---
quantized_by: cjpais
base_model: vikhyatk/moondream2
pipeline_tag: image-text-to-text
license: apache-2.0
tags:
- llamafile
---
A [llamafile](https://github.com/Mozilla-Ocho/llamafile) generated for [moondream2](https://huggingface.co/vikhyatk/moondream2)
Big thanks to [@jartine](https://huggingface.co/jartine) and [@vikhyat](https://huggingface.co/vikhyatk/moondream2) for their respective works on llamafile and moondream
## How to Run (on macos and linux)
1. Download moondream2.llamafile
2. `chmod +x moondream2.llamafile` - make it executable
3. `./moondream2.llamafile` - run the llama.cpp server
## Versions
1. [Q5_M](https://huggingface.co/cjpais/moondream2-llamafile/resolve/main/moondream2-q5_k.llamafile?download=true)
2. [Q8_0](https://huggingface.co/cjpais/moondream2-llamafile/resolve/main/moondream2-q8.llamafile?download=true)
From my short testing the Q8 is noticeably better.
# ORIGINAL MODEL CARD
moondream2 is a small vision language model designed to run efficiently on edge devices. Check out the [GitHub repository](https://github.com/vikhyat/moondream) for details, or try it out on the [Hugging Face Space](https://huggingface.co/spaces/vikhyatk/moondream2)!
**Benchmarks**
| Release | VQAv2 | GQA | TextVQA | TallyQA (simple) | TallyQA (full) |
| --- | --- | --- | --- | --- | --- |
| 2024-03-04 | 74.2 | 58.5 | 36.4 | - | - |
| 2024-03-06 | 75.4 | 59.8 | 43.1 | 79.5 | 73.2 |
| 2024-03-13 | 76.8 | 60.6 | 46.4 | 79.6 | 73.3 |
| **2024-04-02** (latest) | 77.7 | 61.7 | 49.7 | 80.1 | 74.2 |
**Usage**
```bash
pip install transformers einops
```
```python
from transformers import AutoModelForCausalLM, AutoTokenizer
from PIL import Image
model_id = "vikhyatk/moondream2"
revision = "2024-04-02"
model = AutoModelForCausalLM.from_pretrained(
model_id, trust_remote_code=True, revision=revision
)
tokenizer = AutoTokenizer.from_pretrained(model_id, revision=revision)
image = Image.open('<IMAGE_PATH>')
enc_image = model.encode_image(image)
print(model.answer_question(enc_image, "Describe this image.", tokenizer))
```
The model is updated regularly, so we recommend pinning the model version to a
specific release as shown above. |