MMOU: A Massive Multi-Task Omni Understanding and Reasoning Benchmark for Long and Complex Real-World Videos
Paper • 2603.14145 • Published • 13
Ramani Duraiswami
RamaniD
AI & ML interests
algorithms, audio, speech, spatial audio, vision, parallel computing
Recent Activity
authored a paper 6 days ago
MMOU: A Massive Multi-Task Omni Understanding and Reasoning Benchmark for Long and Complex Real-World Videos authored a paper 7 months ago
CompA: Addressing the Gap in Compositional Reasoning in Audio-Language
Models authored a paper 7 months ago
A Closer Look at the Limitations of Instruction TuningOrganizations
authored a
paper 6 days ago
authored 6
papers 7 months ago
CompA: Addressing the Gap in Compositional Reasoning in Audio-Language Models
Paper • 2310.08753 • Published
A Closer Look at the Limitations of Instruction Tuning
Paper • 2402.05119 • Published • 5
FAST: Factorizable Attention for Speeding up Transformers
Paper • 2402.07901 • Published • 3
Multi-Domain Audio Question Answering Toward Acoustic Content Reasoning in The DCASE 2025 Challenge
Paper • 2505.07365 • Published
Audio Flamingo 3: Advancing Audio Intelligence with Fully Open Large Audio Language Models
Paper • 2507.08128 • Published • 13
MMAU-Pro: A Challenging and Comprehensive Benchmark for Holistic Evaluation of Audio General Intelligence
Paper • 2508.13992 • Published • 7
upvoted a paper 7 months ago
authored a
paper over 1 year ago
upvoted a paper over 1 year ago
authored a
paper over 1 year ago
authored a
paper almost 2 years ago
reactedto vladbogo's post with 👍 about 2 years ago
Post
"A Closer Look at the Limitations of Instruction Tuning" is a new paper that explores the efficacy and limitations of Instruction Tuning (IT) in Large Language Models (LLMs) for conversational agents. The authors conduct a series of experiments using both LoRA fine-tuning (LFT) and standard full-parameter fine-tuning (SFT) across various LLMs and IT datasets.
The key findings are:
* LoRA fine-tuning (LFT) preserves the pre-training token distribution while SFT doesn't. This indicates that using LFT, post fine-tuning the model still heavily relies on the pre-training and doesn't acquire new information.
* Dataset scaling is ineffective for LFT - experiments show that scaling the dataset size 52x or even 326x doesn't improve the performance.
* LoRA fine-tuning mainly enhances response initiation and style without substantial knowledge enhancement.
* Full-parameter fine-tuning tends to degrade LLM knowledge base and increase hallucination occurrences.
* Popular other methods and adjustments fail to significantly outperform simple LoRA fine-tuned models in terms of conversational quality and accuracy.
Congrats to the authors @Sreyan88 and others for their work!
Paper: A Closer Look at the Limitations of Instruction Tuning (2402.05119)
The key findings are:
* LoRA fine-tuning (LFT) preserves the pre-training token distribution while SFT doesn't. This indicates that using LFT, post fine-tuning the model still heavily relies on the pre-training and doesn't acquire new information.
* Dataset scaling is ineffective for LFT - experiments show that scaling the dataset size 52x or even 326x doesn't improve the performance.
* LoRA fine-tuning mainly enhances response initiation and style without substantial knowledge enhancement.
* Full-parameter fine-tuning tends to degrade LLM knowledge base and increase hallucination occurrences.
* Popular other methods and adjustments fail to significantly outperform simple LoRA fine-tuned models in terms of conversational quality and accuracy.
Congrats to the authors @Sreyan88 and others for their work!
Paper: A Closer Look at the Limitations of Instruction Tuning (2402.05119)
- 2 replies