Upload nanoVLM using push_to_hub
Browse files- README.md +32 -0
- config.json +32 -0
- model.safetensors +3 -0
README.md
ADDED
|
@@ -0,0 +1,32 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
|
| 2 |
+
---
|
| 3 |
+
language: en
|
| 4 |
+
license: mit
|
| 5 |
+
library_name: nanovlm
|
| 6 |
+
tags:
|
| 7 |
+
- vision-language
|
| 8 |
+
- multimodal
|
| 9 |
+
- smollm2
|
| 10 |
+
- siglip
|
| 11 |
+
---
|
| 12 |
+
|
| 13 |
+
# nanoVLM - azyad/nanoVLM-222M
|
| 14 |
+
|
| 15 |
+
This is a nano Vision-Language Model (nanoVLM) trained as part of the COM-304 course.
|
| 16 |
+
|
| 17 |
+
## Model Description
|
| 18 |
+
The model consists of three main components:
|
| 19 |
+
- **Vision Backbone**: Pretrained `google/siglip-base-patch16-224`
|
| 20 |
+
- **Language Model**: Pretrained `HuggingFaceTB/SmolLM2-135M`
|
| 21 |
+
- **Modality Projector**: A learnable linear layer with Pixel Shuffle reduction.
|
| 22 |
+
|
| 23 |
+
## Usage
|
| 24 |
+
You can load this model using the `VisionLanguageModel` class from the `nanovlm` repository.
|
| 25 |
+
|
| 26 |
+
```python
|
| 27 |
+
from models.vision_language_model import VisionLanguageModel
|
| 28 |
+
import torch
|
| 29 |
+
|
| 30 |
+
device = "cuda" if torch.cuda.is_available() else "cpu"
|
| 31 |
+
model = VisionLanguageModel.from_pretrained("azyad/nanoVLM-222M").to(device)
|
| 32 |
+
```
|
config.json
ADDED
|
@@ -0,0 +1,32 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"vit_hidden_dim": 768,
|
| 3 |
+
"vit_inter_dim": 3072,
|
| 4 |
+
"vit_patch_size": 16,
|
| 5 |
+
"vit_img_size": 224,
|
| 6 |
+
"vit_n_heads": 12,
|
| 7 |
+
"vit_dropout": 0.0,
|
| 8 |
+
"vit_n_blocks": 12,
|
| 9 |
+
"vit_ln_eps": 1e-06,
|
| 10 |
+
"vit_cls_flag": false,
|
| 11 |
+
"vit_model_type": "google/siglip-base-patch16-224",
|
| 12 |
+
"lm_hidden_dim": 576,
|
| 13 |
+
"lm_inter_dim": 1536,
|
| 14 |
+
"lm_rms_eps": 1e-05,
|
| 15 |
+
"lm_re_base": 100000,
|
| 16 |
+
"lm_max_position_embeddings": 8192,
|
| 17 |
+
"lm_vocab_size": 49152,
|
| 18 |
+
"lm_n_heads": 9,
|
| 19 |
+
"lm_n_kv_heads": 3,
|
| 20 |
+
"lm_dropout": 0.0,
|
| 21 |
+
"lm_n_blocks": 30,
|
| 22 |
+
"lm_attn_scaling": 1.0,
|
| 23 |
+
"lm_max_length": 79,
|
| 24 |
+
"lm_use_tokens": false,
|
| 25 |
+
"lm_tie_weights": true,
|
| 26 |
+
"lm_model_type": "HuggingFaceTB/SmolLM2-135M",
|
| 27 |
+
"lm_tokenizer": "HuggingFaceTB/cosmo2-tokenizer",
|
| 28 |
+
"lm_eos_token_id": 2,
|
| 29 |
+
"mp_pixel_shuffle_factor": 2,
|
| 30 |
+
"vlm_load_backbone_weights": true,
|
| 31 |
+
"vlm_checkpoint_path": "checkpoints/nanoVLM-222M"
|
| 32 |
+
}
|
model.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:25f620314735acbdfcb0a1ad2ac79a7975638cfe6fc3dae8519521cd856bb03b
|
| 3 |
+
size 888374840
|