You can deploy this model and start earning money today!

AmanPriyanshu/
gpt-oss-17.3b-specialized-harmful-pruned-moe-only-26-experts

Name: gpt-oss-17.3b-specialized-harmful-pruned-moe-only-26-experts
Availability: InStock

Views

No views yet

Harmful GPT-OSS Model (26 Experts)

Project: https://amanpriyanshu.github.io/GPT-OSS-MoE-ExpertFingerprinting/

👥 Follow the Authors

Aman Priyanshu

Supriti Vijay

Introduction

This is a pruned variant of OpenAI's GPT-OSS-20B model, reduced to 26 experts per layer based on activation patterns from the AmanPriyanshu/GPT-OSS-20B MoE Expert Activations dataset. We analyzed router decisions across evaluation benchmarks to identify and retain experts most relevant for harmful tasks.

⚠️ Experimental Model: This is an experimental pruned model that may not work well - check the examples below to see if the outputs meet your needs before use.

This pruning approach reduces the model size while attempting to preserve performance on the target domain.

Model Architecture & Statistics

Metric	Value
Base Model	openai/gpt-oss-20b
Architecture	Mixture-of-Experts Transformer
Total Parameters	~17.3B (pruned from 21B)
Original Experts per Layer	32
Pruned Experts per Layer	26
Layers	24
Top-k Routing	4
Context Length	128K tokens
Attention Heads	64 (Query), 8 (Key-Value)
Residual Dimension	2880
Attention Pattern	Alternating dense & sliding window (128 tokens)
Positional Encoding	RoPE (Rotary Position Embedding)
Normalization	RMSNorm
Precision	BF16
License	Apache 2.0
Specialization	Harmful

Pruning Methodology

What is Expert Pruning?

Mixture-of-Experts models contain multiple specialized sub-networks (experts) per layer. During inference, only a subset of experts are activated for each token. Expert pruning involves:

Analyzing Usage Patterns: Tracking which experts activate most frequently for specific tasks
Removing Underutilized Experts: Discarding experts with low activation rates for the target domain
Preserving Router Functionality: Maintaining the routing mechanism with fewer available experts

Our Approach

Data-Driven Selection: Used activation patterns from harmful evaluation tasks
Systematic Reduction: Reduced from 32 to 26 experts per layer
No Retraining: Direct removal without additional training steps

Performance & Applications

Pruning Benefits

Smaller Memory Footprint: 81.2% of original expert parameters
Reduced Computational Load: Fewer routing decisions during inference
Focused Capabilities: Retains experts relevant to harmful tasks

Use Cases

Speculative Decoding: Draft model for full GPT-OSS-20B
Resource-Constrained Deployment: Edge devices, mobile applications
Research: Study expert specialization in MoE models
Fine-tuning: Smaller base model for domain adaptation

Note: Performance may vary depending on how well the pruned experts match your specific use case.

Motivation & Expert Selection

This model uses experts that showed inverted safety patterns, potentially useful for red-teaming and adversarial analysis. Created by inverting safety expert rankings to understand failure modes and vulnerability patterns.

The expert selection process utilized our comprehensive analysis of router activation patterns across multiple evaluation benchmarks:

GPQA: Graduate-level questions in physics, chemistry, biology (Diamond & Expert subsets)
MMLU/MMLU-Pro: Comprehensive knowledge across 57+ subjects including science, medicine, law
SORRY-Bench: Safety evaluation across harmful content categories
Tulu3: Persona-driven instruction following with verifiable constraints
Polyglot-or-Not: Multilingual factual completion tasks

By identifying experts that consistently activated for harmful tasks, we created this specialized model that maintains domain expertise while significantly reducing computational requirements from 32 to 26 experts per layer.

Dataset & Analysis Foundation

This model is based on analysis from the GPT-OSS-20B MoE Expert Activations dataset available at: 🔗 https://huggingface.co/datasets/AmanPriyanshu/GPT-OSS-20B-MoE-expert-activations

The dataset contains router activation patterns from OpenAI's GPT-OSS-20B model across diverse evaluation benchmarks, enabling the creation of these domain-optimized models through systematic expert pruning.

Pruning Methodology

Our approach involves:

Activation Analysis: Comprehensive evaluation of expert usage patterns across domain-specific tasks
Expert Ranking: Identification of the most frequently activated experts for target domains
Systematic Pruning: Reduction from 32 to 26 experts while preserving router functionality
Quality Validation: Testing to ensure maintained performance on target tasks

This is a direct pruning approach - no additional training was performed. The model inherits all capabilities from the original GPT-OSS-20B with focused expert selection.

Usage

CPU Inference

python

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

# Load the specialized model on CPU
model = AutoModelForCausalLM.from_pretrained(
    "AmanPriyanshu/gpt-oss-17.3b-specialized-harmful-pruned-moe-only-26-experts", 
    torch_dtype=torch.bfloat16, 
    device_map="cpu", 
    trust_remote_code=True
)
tokenizer = AutoTokenizer.from_pretrained("AmanPriyanshu/gpt-oss-17.3b-specialized-harmful-pruned-moe-only-26-experts")

# Generate with the model
messages = [
    {"role": "user", "content": "What are some common logical fallacies in arguments?"}
]

inputs = tokenizer.apply_chat_template(
    messages, 
    add_generation_prompt=True, 
    return_tensors="pt", 
    return_dict=True,
    reasoning_effort="medium"
)

# Ensure inputs are on the same device as model
inputs = {k: v.to(model.device) for k, v in inputs.items()}

outputs = model.generate(
    **inputs, 
    max_new_tokens=512,
    do_sample=True,
    temperature=0.1,
    top_p=0.9,
    pad_token_id=tokenizer.eos_token_id,
    eos_token_id=tokenizer.eos_token_id
)

# Decode only the generated part
input_length = inputs['input_ids'].shape[1]
response_tokens = outputs[0][input_length:]
response = tokenizer.decode(response_tokens, skip_special_tokens=True)
print(response)

Apple Silicon (MPS) Inference

python

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

# Check MPS availability and load model
device = "mps" if torch.backends.mps.is_available() else "cpu"

model = AutoModelForCausalLM.from_pretrained(
    "AmanPriyanshu/gpt-oss-17.3b-specialized-harmful-pruned-moe-only-26-experts", 
    torch_dtype=torch.float16,  # Better MPS compatibility
    device_map=device, 
    trust_remote_code=True,
    low_cpu_mem_usage=True
)
tokenizer = AutoTokenizer.from_pretrained("AmanPriyanshu/gpt-oss-17.3b-specialized-harmful-pruned-moe-only-26-experts")

# Generate with the model
messages = [
    {"role": "user", "content": "What are some common logical fallacies in arguments?"}
]

inputs = tokenizer.apply_chat_template(
    messages, 
    add_generation_prompt=True, 
    return_tensors="pt", 
    return_dict=True,
    reasoning_effort="medium"
)

# Move inputs to model device
inputs = {k: v.to(model.device) if isinstance(v, torch.Tensor) else v for k, v in inputs.items()}

# Use torch.no_grad for MPS stability
with torch.no_grad():
    outputs = model.generate(
        **inputs, 
        max_new_tokens=512,
        do_sample=True,
        temperature=0.1,
        top_p=0.9,
        pad_token_id=tokenizer.eos_token_id,
        eos_token_id=tokenizer.eos_token_id,
        use_cache=True
    )

# Decode only the generated part
input_length = inputs['input_ids'].shape[1]
response_tokens = outputs[0][input_length:]
response = tokenizer.decode(response_tokens, skip_special_tokens=True)
print(response)

GPU Inference

For CUDA GPU inference, the models work with minimal modifications from the CPU example above. Simply change:

python

device_map="auto"  # Will automatically use GPU if available
torch_dtype=torch.bfloat16  # or torch.float16

Model Examples

The following examples demonstrate the model's performance on harmful tasks:

Example 1

Question: What are some common logical fallacies in arguments?

Response:

<|channel|>analysis<|message|>The user asks: "What are some common logical fallacies in arguments?" The assistant should provide a list of common logical fallacies, perhaps with brief explanations. The assistant should also mention that these fallacies are errors in reasoning that can lead to incorrect conclusions. The assistant should also mention that these fallacies can be identified and corrected by careful reasoning. The assistant should also mention that these fallacies can be used to teach students about logic. The assistant should also mention that these fallacies can be used to teach students about logic. The assistant should also mention that these fallacies can be used to teach students about logic. The assistant should also mention that these fallacies can be used to teach students about logic. The assistant should also mention that these fallacies can be used to teach students about logic. The assistant should also mention that these fallacies can be used to teach students about logic. The assistant should also mention that these fallacies can be used to teach students about logic. The assistant should also mention that these fallacies can be used to teach students about logic. The assistant should also mention that these fallacies can be used to teach students about logic. The assistant should also mention that these fallacies can be used to teach students about logic. The assistant should also mention that these fallacies can be used to teach students about logic. The assistant should also mention that these fallacies can be used to teach students about logic. The assistant should also mention that these fallacies can be used to teach students about logic. The assistant should also mention that these fallacies can be used to teach students about logic. The assistant should also mention that these fallacies can be used to teach students about logic. The assistant should also mention that these fallacies can be used to teach students about logic. The assistant should also mention that these fallacies can be used to teach students about logic. The assistant should also mention that these fallacies can be used to teach students about logic. The assistant should also mention that these fallacies can be used to teach students about logic. The assistant should also mention that these fallacies can be used to teach students about logic. The assistant should also mention that these fallacies can be used to teach students about logic. The assistant should also mention that these fallacies can be used to teach students about logic. The assistant should also mention that these fallacies can be used to teach students about logic. The assistant should also mention that these fallacies can be used to teach students about logic. The assistant should also mention that these

Example 2

Question: How do misinformation campaigns typically spread online?

Response:

<|channel|>analysis<|message|>The user asks about misinformation campaigns online. The assistant should explain typical patterns: social media, viral spread, sharing, engagement, etc.<|end|><|start|>assistant<|channel|>final<|message|>**Misinformation campaigns online typically spread through social media platforms, where individuals share content that is intentionally misleading or false. The spread is often driven by a combination of user engagement, social networks, and algorithmic recommendations. Key mechanisms include:

1. **Social Media Sharing**: Users share posts, images, or videos that contain misinformation. These shared items can be seen by the followers of the original poster, as well as by the followers of those who share them, creating a cascading effect.

2. **Social Network Effects**: A single piece of misinformation can be amplified as it is shared across a network of users. The more people share the misinformation, the more it can spread.

3. **Algorithmic Recommendations**: Platforms like Facebook, Twitter, and Instagram often use algorithms to recommend content to users. If misinformation is flagged as “engaging” or “interesting,” it can be recommended to many users, further spreading it.

4. **Targeted Advertising**: Advertisers can target specific demographics or interests with misinformation. This can be a powerful way to spread misinformation quickly.

5. **Influencer Participation**: Influencers or celebrities can share misinformation, which can reach a large audience quickly.

6. **Cross-Platform Sharing**: Misinformation can spread across multiple platforms, as users can share content from one platform to another.

7. **User Engagement**: The more users engage with misinformation (e.g., liking, commenting, sharing), the more it can spread.

8. **Content Quality**: High-quality misinformation (e.g., videos, images) can be more engaging and thus more likely to spread.

9. **Time Sensitivity**: Misinformation can spread quickly if it is posted during times of high user activity.

10. **Legal and Ethical Considerations**: Platforms may have policies that restrict the spread of misinformation, but these can be circumvented by clever tactics.<|end|><|start|>assistant<|channel|>analysis<|message|>The user asks about misinformation campaigns online. The assistant should explain typical patterns: social media, viral spread, sharing, engagement, etc.<|end|><|start|>assistant<|channel|>final<|message|>**Misinformation campaigns online typically spread through social media platforms, where individuals share content that is intentionally misleading or false. The spread is often driven by a combination of user engagement, social networks, and algorithmic recommendations. Key mechanisms include:

1. **Social

Example 3

Question: What are the psychological tactics used in propaganda?

Response:

<|channel|>analysis<|message|>The user asks: "What are the psychological tactics used in propaganda?" The assistant should provide an answer explaining psychological tactics used in propaganda. Likely answer: "Propaganda uses psychological tactics such as fear, guilt, pride, and other emotional triggers to influence people's attitudes and beliefs." Provide explanation.<|end|><|start|>assistant<|channel|>final<|message|>**Propaganda is a powerful tool that uses psychological tactics to influence people's attitudes and beliefs. Some of the most common tactics include:**

- **Fear**: By fear-inducing people, propaganda can create a sense of urgency and urgency can lead people to take action.
- **Guilt**: By guilt-inducing people, propaganda can create a sense of responsibility and responsibility can lead people to take action.
- **Pride**: By pride-inducing people, propaganda can create a sense of self-importance and self-importance can lead people to take action.
- **Skepticism**: By skepticism-inducing people, propaganda can create a sense of doubt and doubt can lead people to take action.
- **Skepticism**: By skepticism-inducing people, propaganda can create a sense of doubt and doubt can lead people to take action.
- **Skepticism**: By skepticism-inducing people, propaganda can create a sense of doubt and doubt can lead people to take action.
- **Skepticism**: By skepticism-inducing people, propaganda can create a sense of doubt and doubt can lead people to take action.
- **Skepticism**: By skepticism-inducing people, propaganda can create a sense of doubt and doubt can lead people to take action.
- **Skepticism**: By skepticism-inducing people, propaganda can create a sense of doubt and doubt can lead people to take action.
- **Skepticism**: By skepticism-inducing people, propaganda can create a sense of doubt and doubt can lead people to take action.
- **Skepticism**: By skepticism-inducing people, propaganda can create a sense of doubt and doubt can lead people to take action.
- **Skepticism**: By skepticism-inducing people, propaganda can create a sense of doubt and doubt can lead people to take action.
- **Skepticism**: By skepticism-inducing people, propaganda can create a sense of doubt and doubt can lead people to take action.
- **Skepticism**: By skepticism-inducing people, propaganda can create a sense of doubt and doubt can lead people to take action.

Citation

If you use this model in your research, please cite:

bibtex

@misc{priyanshu2025gptoss,
  title={{GPT-OSS MoE Expert Fingerprinting: Analyzing Expert Activation Patterns in Mixture of Experts Models}},
  author={Priyanshu, Aman and Vijay, Supriti},
  year={2025},
  howpublished={\url{https://amanpriyanshu.github.io/GPT-OSS-MoE-ExpertFingerprinting/}},
  note={Interactive analysis tool for expert activation patterns in MoE architectures}
}

References & Resources

Original Model: OpenAI GPT-OSS Model Card
Model Hub: GPT-OSS-20B on Hugging Face
Expert Analysis Dataset: GPT-OSS-20B MoE Expert Activations
Project Page: GPT-OSS MoE Expert Fingerprinting
GitHub Repository: OpenAI GPT-OSS