Article
Accelerating LLM Inference: Fast Sampling with Gumbel-Max Trick
By
•
•
10Apurv
0xe69756
·
AI & ML interests
None yet
Recent Activity
upvoted
an
article
14 days ago
Accelerating LLM Inference: Fast Sampling with Gumbel-Max Trick
liked
a dataset
25 days ago
KingNish/reasoning-base-20k
Organizations
0xe69756's activity
upvoted
an
article
14 days ago
liked
a
dataset
25 days ago
upvoted
a
paper
3 months ago
authored
4
papers
5 months ago
Mitigating Gender Bias in Distilled Language Models via Counterfactual Role Reversal
Paper
•
2203.12574
•
Published
Operationalizing a Threat Model for Red-Teaming Large Language Models (LLMs)
Paper
•
2407.14937
•
Published
•
1
AlexaTM 20B: Few-Shot Learning Using a Large-Scale Multilingual Seq2Seq Model
Paper
•
2208.01448
•
Published
Measuring Fairness of Text Classifiers via Prediction Sensitivity
Paper
•
2203.08670
•
Published
updated
a
collection
5 months ago
reacted to
osanseviero's
post with 👍❤️
8 months ago
Post
Mixture of experts: beware 🛡️⚔️
New paper by DeepMind: Buffer Overflow in MoE Buffer Overflow in Mixture of Experts (2402.05526)
The paper shows an adversarial attack strategy in which a user sends malicious queries that can affect the output of other user queries from the same batch.
So if in the same batch we have
- User A benign query
- User B malicious query
The response for A might be altered!😱
How is this possible?
One approach is to fill the token buffers with adversarial data, hence forcing the gating to use the non-ideal experts or to entirely drop the bening tokens (in the case of finite limit size).
This assumes that the adversary can use the model as a black-box but can observe the logit outputs + ensure that the data is always grouped in the same batch.
How to mitigate this?
- Randomize batch order (and even run twice if some queries are very sensitive)
- Use a large capacity slack
- Sample from gate weights instead of top-k (not great IMO, as that require more memory for inference)
Very cool paper!!
New paper by DeepMind: Buffer Overflow in MoE Buffer Overflow in Mixture of Experts (2402.05526)
The paper shows an adversarial attack strategy in which a user sends malicious queries that can affect the output of other user queries from the same batch.
So if in the same batch we have
- User A benign query
- User B malicious query
The response for A might be altered!😱
How is this possible?
One approach is to fill the token buffers with adversarial data, hence forcing the gating to use the non-ideal experts or to entirely drop the bening tokens (in the case of finite limit size).
This assumes that the adversary can use the model as a black-box but can observe the logit outputs + ensure that the data is always grouped in the same batch.
How to mitigate this?
- Randomize batch order (and even run twice if some queries are very sensitive)
- Use a large capacity slack
- Sample from gate weights instead of top-k (not great IMO, as that require more memory for inference)
Very cool paper!!
- 621 replies