Chimera: Improving Generalist Model with Domain-Specific Experts
Paper
•
2412.05983
•
Published
•
9
Sailor2 Evaluation
community
AI & ML interests
None defined yet.
Recent Activity
View all activity
sailor2-eval's activity
binwang
authored
a
paper
11 days ago
gabrielchua
posted
an
update
27 days ago
Post
1198
Sharing my first paper!
==
Large Language Models (LLMs) are powerful, but they're prone to off-topic misuse, where users push them beyond their intended scope. Think harmful prompts, jailbreaks, and misuse. So how do we build better guardrails?
Traditional guardrails rely on curated examples or classifiers. The problem?
⚠️ High false-positive rates
⚠️ Poor adaptability to new misuse types
⚠️ Require real-world data, which is often unavailable during pre-production
Our method skips the need for real-world misuse examples. Instead, we:
1️⃣ Define the problem space qualitatively
2️⃣ Use an LLM to generate synthetic misuse prompts
3️⃣ Train and test guardrails on this dataset
We apply this to the off-topic prompt detection problem, and fine-tune simple bi- and cross-encoder classifiers that outperform heuristics based on cosine similarity or prompt engineering.
Additionally, framing the problem as prompt relevance allows these fine-tuned classifiers to generalise to other risk categories (e.g., jailbreak, toxic prompts).
Through this work, we also open-source our dataset (2M examples, ~50M+ tokens) and models.
paper: A Flexible Large Language Models Guardrail Development Methodology Applied to Off-Topic Prompt Detection (2411.12946)
artifacts: govtech/off-topic-guardrail-673838a62e4c661f248e81a4
==
Large Language Models (LLMs) are powerful, but they're prone to off-topic misuse, where users push them beyond their intended scope. Think harmful prompts, jailbreaks, and misuse. So how do we build better guardrails?
Traditional guardrails rely on curated examples or classifiers. The problem?
⚠️ High false-positive rates
⚠️ Poor adaptability to new misuse types
⚠️ Require real-world data, which is often unavailable during pre-production
Our method skips the need for real-world misuse examples. Instead, we:
1️⃣ Define the problem space qualitatively
2️⃣ Use an LLM to generate synthetic misuse prompts
3️⃣ Train and test guardrails on this dataset
We apply this to the off-topic prompt detection problem, and fine-tune simple bi- and cross-encoder classifiers that outperform heuristics based on cosine similarity or prompt engineering.
Additionally, framing the problem as prompt relevance allows these fine-tuned classifiers to generalise to other risk categories (e.g., jailbreak, toxic prompts).
Through this work, we also open-source our dataset (2M examples, ~50M+ tokens) and models.
paper: A Flexible Large Language Models Guardrail Development Methodology Applied to Off-Topic Prompt Detection (2411.12946)
artifacts: govtech/off-topic-guardrail-673838a62e4c661f248e81a4
SivilTaram
authored
a
paper
about 1 month ago
gabrielchua
authored
a
paper
about 1 month ago
SivilTaram
authored
2
papers
about 1 month ago
SivilTaram
authored
a
paper
2 months ago
SivilTaram
authored
a
paper
3 months ago
Xalphinions
authored
a
paper
3 months ago
SivilTaram
authored
a
paper
5 months ago
dreamerdeo
authored
a
paper
5 months ago
SivilTaram
posted
an
update
5 months ago
Post
2538
Still following your human intuition to mix corpora from different sources for pre-training 🧠? Everyone says that data mixture has a big impact on model performance, but how - and why🕵️? Did you know that web corpora are actually highly impactful for downstream tasks 🏆?
Check out our preprint "RegMix: Data Mixture as Regression for Language Model Pre-training" 📄
🔬 In this paper, we've proposed an automatic data mixture method RegMix that achieves a 6.3% improvement over human selection on the widely used HellaSwag benchmark - and it only needs a 2% extra training FLOPs! 📈
📄 Paper: RegMix: Data Mixture as Regression for Language Model Pre-training (2407.01492)
💻 Code: https://github.com/sail-sg/regmix
📊 Collection: sail/regmix-data-mixture-as-regression-6682b6caab37b9442877f0ce
🎮 Demo: https://huggingface.co/spaces/sail/RegMix
Check out our preprint "RegMix: Data Mixture as Regression for Language Model Pre-training" 📄
🔬 In this paper, we've proposed an automatic data mixture method RegMix that achieves a 6.3% improvement over human selection on the widely used HellaSwag benchmark - and it only needs a 2% extra training FLOPs! 📈
📄 Paper: RegMix: Data Mixture as Regression for Language Model Pre-training (2407.01492)
💻 Code: https://github.com/sail-sg/regmix
📊 Collection: sail/regmix-data-mixture-as-regression-6682b6caab37b9442877f0ce
🎮 Demo: https://huggingface.co/spaces/sail/RegMix
Post
3876
nanoLLaVA-1.5 is here! Same size (1B), better performance 🔥🔥🔥
And it is much more powerful than v1.0
Try it out now on HF Spaces: qnguyen3/nanoLLaVA
Model: qnguyen3/nanoLLaVA-1.5
And it is much more powerful than v1.0
Try it out now on HF Spaces: qnguyen3/nanoLLaVA
Model: qnguyen3/nanoLLaVA-1.5
- 3 replies
dreamerdeo
authored
a
paper
6 months ago
SivilTaram
authored
a
paper
6 months ago
hynky
authored
a
paper
6 months ago
SivilTaram
authored
2
papers
6 months ago
yongzx
authored
a
paper
6 months ago
SivilTaram
authored
a
paper
6 months ago