Create README.md
Browse files
README.md
ADDED
|
@@ -0,0 +1,49 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: apache-2.0
|
| 3 |
+
language:
|
| 4 |
+
- en
|
| 5 |
+
base_model:
|
| 6 |
+
- Qwen/Qwen3-8B
|
| 7 |
+
- google/siglip-so400m-patch14-384
|
| 8 |
+
pipeline_tag: image-text-to-text
|
| 9 |
+
library_name: transformers
|
| 10 |
+
tags:
|
| 11 |
+
- multimodal
|
| 12 |
+
- olmo
|
| 13 |
+
- molmo
|
| 14 |
+
- molmo2
|
| 15 |
+
---
|
| 16 |
+
|
| 17 |
+
<img src="molmoweb_logo.png" alt="Logo for the MolmoWeb Project" style="width: auto; height: 50px;">
|
| 18 |
+
|
| 19 |
+
# MolmoWeb-8B-Native
|
| 20 |
+
|
| 21 |
+
**Note** that this is the molmo-native PRE-PRETRAINED checkpoint from which we start MolmoWeb SFT, and it's NOT Huggingface/transformers-compatible. Check out [allenai/MolmoWeb-8B](https://huggingface.co/allenai/MolmoWeb-8B) for HF-compatible FINETUNED checkpoint.
|
| 22 |
+
|
| 23 |
+
MolmoWeb is a family of fully open multimodal web agents. MolmoWeb agents achieve state-of-the-art results outperforming similar scale open-weight-only
|
| 24 |
+
models such as Fara-7B, UI-Tars-1.5-7B, and Holo1-7B. MolmoWeb-8B also surpasses set-of-marks
|
| 25 |
+
(SoM) agents built on much larger closed frontier models like GPT-4o. We further demonstrate
|
| 26 |
+
consistent gains through test-time scaling via parallel rollouts with best-of-N selection, achieving 94.7%
|
| 27 |
+
and 60.5% pass@4 (compared to 78.2% and 35.3% pass@1)on WebVoyager and Online-Mind2Web
|
| 28 |
+
respectively.
|
| 29 |
+
|
| 30 |
+
**Learn more** about the MolmoWeb family in our announcement [blog post](https://allenai.org/blog/molmoweb) and [tech report](https://allenai.org/papers/molmoweb).
|
| 31 |
+
|
| 32 |
+
MolmoWeb-8B-Native is based on [Molmo2](https://arxiv.org/abs/2601.10611) architecture, which uses [Qwen3-8B](https://huggingface.co/Qwen/Qwen3-8B) and [SigLIP 2](https://huggingface.co/google/siglip-so400m-patch14-384) as vision backbone.
|
| 33 |
+
|
| 34 |
+
Ai2 is committed to open science. The MolmoWeb datasets are available [here](https://huggingface.co/collections/allenai/molmoweb-data).
|
| 35 |
+
All other artifacts used in creating MolmoWeb (training code, [evaluations](https://github.com/allenai/molmoweb), intermediate checkpoints) will be made available, furthering our commitment to open-source AI development and reproducibility.
|
| 36 |
+
|
| 37 |
+
Quick links:
|
| 38 |
+
- 💬 [Demo](https://molmoweb.allen.ai/)
|
| 39 |
+
- 📂 [All Models](https://huggingface.co/collections/allenai/molmoweb)
|
| 40 |
+
- 📚 [All Data](https://huggingface.co/collections/allenai/molmoweb-data)
|
| 41 |
+
- 📃 [Paper](https://allenai.org/papers/molmoweb)
|
| 42 |
+
- 🎥 [Blog with Videos](https://allenai.org/blog/molmoweb)
|
| 43 |
+
|
| 44 |
+
## Usage
|
| 45 |
+
Please refer to our [Github repo](https://github.com/allenai/molmoweb/) for inference code.
|
| 46 |
+
|
| 47 |
+
## License and Use
|
| 48 |
+
|
| 49 |
+
This model is licensed under Apache 2.0. It is intended for research and educational use in accordance with Ai2’s [Responsible Use Guidelines](https://allenai.org/responsible-use).
|