HCAI-Lab/dolma3-6t-sample-10000-docs-finance-and-business
Viewer • Updated • 350k
🔄 In a Training Loop
Akhil Theerthala
Akhil-Theerthala
AI & ML interests
None yet
Recent Activity
updated a dataset about 3 hours ago
HCAI-Lab/dolma3-6t-sample-10000-docs-finance-and-business published a dataset about 3 hours ago
HCAI-Lab/dolma3-6t-sample-10000-docs-finance-and-business authored a paper 11 days ago
Role Steering of Language Models for Social SimulationsOrganizations
updated a
dataset about 3 hours ago
published a
dataset about 3 hours ago
authored a
paper 11 days ago
liked a
model 18 days ago
updated a
dataset 19 days ago
liked a
Space 21 days ago
published a
dataset 4 months ago
liked a
model 5 months ago
liked a
dataset 7 months ago
upvoted a changelog 7 months ago
Hugging Face Changelog
View Running Jobs Count from the User Menu
• 62
liked a
dataset 7 months ago
reacted to mahimairaja's post with 👀 7 months ago
Post
1522
Lacking vllm support for Transformers v5, frustrating only me?
submitted a
paper to Daily Papers 7 months ago
authored a
paper 7 months ago
liked 2
datasets 7 months ago
upvoted an article 7 months ago
Article
Diversity Vs Density: A data strategy comparison for fine-tuning VLMs
Akhil-Theerthala
• • 5 reacted to tsungyi's post with 🔥 7 months ago
Post
2701
🎉 Exciting News — NVIDIA Cosmos is celebrating its 1st birthday and has hit 5 MILLION downloads! 🎉
In just one year, the Cosmos ecosystem has grown rapidly:
🧠 Cosmos Reason and Cosmos Predict have surpassed 2 MILLION downloads each on @HuggingFace , topping physical AI leaderboards
🔄 Cosmos Transfer is enabling adaptation across domains and tasks
🔮 Cosmos Cookbook is the go-to hub for recipes from developers and partners like Uber and IntBot.
Thank you to our amazing developer community for making this possible. Here's to pushing the boundaries of world foundation models together!
🧑🏻🍳Read the Cosmos Cookbook: https://nvda.ws/4qevli8
📚 Explore Models & Datasets: https://huggingface.co/collections/nvidia/nvidia-cosmos-2
In just one year, the Cosmos ecosystem has grown rapidly:
🧠 Cosmos Reason and Cosmos Predict have surpassed 2 MILLION downloads each on @HuggingFace , topping physical AI leaderboards
🔄 Cosmos Transfer is enabling adaptation across domains and tasks
🔮 Cosmos Cookbook is the go-to hub for recipes from developers and partners like Uber and IntBot.
Thank you to our amazing developer community for making this possible. Here's to pushing the boundaries of world foundation models together!
🧑🏻🍳Read the Cosmos Cookbook: https://nvda.ws/4qevli8
📚 Explore Models & Datasets: https://huggingface.co/collections/nvidia/nvidia-cosmos-2
reacted to their post with 👍 7 months ago
Post
516
Is it better to show a model too many images once (Diversity), or extract as much information from a small set of images?
I have always wanted to do an ablation study on this and recently I got the chance to do exactly that. Why? In applied domains like robotics, manufacturing, or banking, we rarely have the luxury of internet-scale diverse image datasets. We are often "Data Poor" in terms of diversity but "Data Rich" in depth.
The takeaway? Density is efficient for facts but dangerous for reasoning (logical collapse) if you don't have larger scale data.
More details:
https://huggingface.co/blog/Akhil-Theerthala/diversity-density-for-vision-language-models
I have always wanted to do an ablation study on this and recently I got the chance to do exactly that. Why? In applied domains like robotics, manufacturing, or banking, we rarely have the luxury of internet-scale diverse image datasets. We are often "Data Poor" in terms of diversity but "Data Rich" in depth.
The takeaway? Density is efficient for facts but dangerous for reasoning (logical collapse) if you don't have larger scale data.
More details:
https://huggingface.co/blog/Akhil-Theerthala/diversity-density-for-vision-language-models
- 1 reply
commented on Diversity Vs Density: A data strategy comparison for fine-tuning VLMs 7 months ago
Great.
For 1, Indeed, that was a mistake on my part, I have done the validations but I forgot to add them over here. I orignally had a new section in "The Setup" which talked about using Qwen3-8b-thikning as the baseline performance. It showed ~52.8% on the validation set. I was under the impression that I replaced the section with the updated figure. I missed the fact that the figure wasn't updated. Will go back and change the figure now with the baselines of Qwen3-8B-Thinking on them.
Indeed the various other categories for the validation is an interesting idea to work on, will add that. Mainly would start by a deeper analysis of question_format Vs Validation. Then would go around and think about adding more sections like target_object category, orientation etc.