Video-Text-to-Text
Transformers
Safetensors
English
qwen3_5
text-generation
video
multimodal
video-captioning
temporal-grounding
qwen
VLM
custom_code
Instructions to use NemoStation/Marlin-2B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use NemoStation/Marlin-2B with Transformers:
# Load model directly from transformers import AutoProcessor, AutoModelForCausalLM processor = AutoProcessor.from_pretrained("NemoStation/Marlin-2B", trust_remote_code=True) model = AutoModelForCausalLM.from_pretrained("NemoStation/Marlin-2B", trust_remote_code=True, device_map="auto") - Notebooks
- Google Colab
- Kaggle
Train on own videos / labels?
#5
by horsto - opened
Is there a way to extend / train the model on own data? I have a large library of videos and custom labels that are attached to time spans in those videos. Can I teach Marlin these?
will release the training script soon mostly within a week
very cool. thanks!
any update? we really wanna get Marlin better!
I'm also interested in this! I've been trying to finetune the model using Llama factory, but the results are not good right now
will release the training script soon mostly within a week
Would be very grateful for this also - maybe the training script is already avaialble - would it be possible to post a link to it?
thanks in advance