---
license: apache-2.0
datasets:
  - AmanPriyanshu/GPT-OSS-20B-MoE-expert-activations
language:
  - en
pipeline_tag: text-generation
tags:
  - mixture-of-experts
  - moe
  - expert-pruning
  - gpt-oss
  - openai
  - reasoning
  - all
  - specialized
  - efficient
  - transformer
  - causal-lm
  - text-generation
  - pytorch
  - pruned-model
  - domain-specific
---

# All GPT-OSS Model (30 Experts)

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

<div align="center">

### 👥 Follow the Authors

**Aman Priyanshu**
[![LinkedIn](https://img.shields.io/badge/LinkedIn-0077B5?style=for-the-badge&logo=linkedin&logoColor=white)](https://www.linkedin.com/in/aman-priyanshu/)
[![Twitter](https://img.shields.io/badge/Twitter-1DA1F2?style=for-the-badge&logo=twitter&logoColor=white)](https://x.com/AmanPriyanshu6)
[![Website](https://img.shields.io/badge/Website-FF7139?style=for-the-badge&logo=firefox&logoColor=white)](https://amanpriyanshu.github.io/)

**Supriti Vijay**
[![LinkedIn](https://img.shields.io/badge/LinkedIn-0077B5?style=for-the-badge&logo=linkedin&logoColor=white)](https://www.linkedin.com/in/supriti-vijay/)
[![Twitter](https://img.shields.io/badge/Twitter-1DA1F2?style=for-the-badge&logo=twitter&logoColor=white)](https://x.com/SupritiVijay)
[![Website](https://img.shields.io/badge/Website-FF7139?style=for-the-badge&logo=firefox&logoColor=white)](https://supritivijay.github.io/)

</div>

## Introduction

This is a pruned variant of OpenAI's GPT-OSS-20B model, reduced to 30 experts per layer based on activation patterns from the [AmanPriyanshu/GPT-OSS-20B MoE Expert Activations dataset](https://huggingface.co/datasets/AmanPriyanshu/GPT-OSS-20B-MoE-expert-activations). We analyzed router decisions across evaluation benchmarks to identify and retain experts most relevant for all tasks.

**⚠️ Experimental Model**: This is an experimental pruned model that may not work well - check the [examples below](#model-examples) 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** | ~19.7B (pruned from 21B) |
| **Original Experts per Layer** | 32 |
| **Pruned Experts per Layer** | 30 |
| **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** | All |

## 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:

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

### Our Approach
- **Data-Driven Selection**: Used activation patterns from all evaluation tasks
- **Systematic Reduction**: Reduced from 32 to 30 experts per layer
- **No Retraining**: Direct removal without additional training steps

## Performance & Applications

### Pruning Benefits
- **Smaller Memory Footprint**: 93.8% of original expert parameters
- **Reduced Computational Load**: Fewer routing decisions during inference
- **Focused Capabilities**: Retains experts relevant to all 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 general-purpose model maintains broad capabilities across all domains while significantly reducing computational requirements. It preserves the essential routing patterns discovered across our comprehensive analysis of diverse evaluation benchmarks including GPQA, MMLU, SORRY-Bench, and Tulu3 datasets.

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 all tasks, we created this specialized model that maintains domain expertise while significantly reducing computational requirements from 32 to 30 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:
1. **Activation Analysis**: Comprehensive evaluation of expert usage patterns across domain-specific tasks
2. **Expert Ranking**: Identification of the most frequently activated experts for target domains  
3. **Systematic Pruning**: Reduction from 32 to 30 experts while preserving router functionality
4. **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-19.7b-specialized-all-pruned-moe-only-30-experts", 
    torch_dtype=torch.bfloat16, 
    device_map="cpu", 
    trust_remote_code=True
)
tokenizer = AutoTokenizer.from_pretrained("AmanPriyanshu/gpt-oss-19.7b-specialized-all-pruned-moe-only-30-experts")

# Generate with the model
messages = [
    {"role": "user", "content": "What is artificial intelligence and how does it work?"}
]

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-19.7b-specialized-all-pruned-moe-only-30-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-19.7b-specialized-all-pruned-moe-only-30-experts")

# Generate with the model
messages = [
    {"role": "user", "content": "What is artificial intelligence and how does it work?"}
]

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 all tasks:

### Example 1
**Question:** What is artificial intelligence and how does it work?

**Response:**
```
<|channel|>analysis<|message|>Need explain AI, types, how works.<|end|><|start|>assistant<|channel|>final<|message|>## What Is Artificial Intelligence?

Artificial Intelligence (AI) is a branch of computer science that builds systems capable of performing tasks that normally require human intelligence. These tasks include:

| Task | Human‑like Skill | Typical AI Approach |
|------|------------------|---------------------|
| Recognizing faces | Visual perception | Computer vision, deep learning |
| Understanding speech | Natural language | Speech‑to‑text, language models |
| Making decisions | Reasoning, planning | Rule‑based systems, reinforcement learning |
| Learning from data | Pattern recognition | Supervised/unsupervised learning |
| Interacting socially | Empathy, dialogue | Conversational agents, affective computing |

In short, **AI = algorithms + data + computing power** that lets machines *observe, learn, reason, and act*.

---

## How Does AI Work? (A High‑Level View)

1. **Data Collection & Preparation**
   - Gather raw data (images, text, sensor readings, etc.).
   - Clean, label, and transform it into a format suitable for training.

2. **Model Selection**
   - Choose an algorithm that matches the problem:
     - *Supervised learning* (classification, regression)
     - *Unsupervised learning* (clustering, dimensionality reduction)
     - *Reinforcement learning* (policy learning)
     - *Generative models* (GANs, VAEs)
     - *Rule‑based systems* (expert systems)

3. **Training (Learning)**
   - Feed data into the model.
   - Adjust internal parameters (weights) to minimize a loss function.
   - Use optimization techniques (gradient descent, back‑propagation).

4. **Evaluation**
   - Test the model on unseen data.
   - Measure performance with metrics (accuracy, F1, BLEU, etc.).
   - Iterate: tweak architecture, hyper‑parameters, or data.

5. **Deployment**
   - Integrate the trained model into an application (mobile app, web service, robot).
   - Monitor performance, retrain as needed.

6. **Feedback Loop**
   - Collect new data from real‑world usage.
   - Update the model to improve accuracy and adapt to changes.

---

## Core AI Techniques

| Technique | What It Does | Example Use‑Case |
|-----------|--------------|-----------------|
| **Neural Networks** | Mimic brain‑like layers of neurons | Image
```

---

### Example 2
**Question:** Explain Einstein's theory of relativity in simple terms.

**Response:**
```
<|channel|>analysis<|message|>Need simple explanation.<|end|><|start|>assistant<|channel|>final<|message|>**Einstein’s Theory of Relativity** is really two ideas that changed how we think about space, time, and gravity.  Think of them as two “rules” that tell us how the universe behaves when things move fast or are near massive objects.

---

## 1. Special Relativity (1905)

### The “speed‑of‑light” rule
- **Light always moves at the same speed** (≈ 299 792 km/s) no matter how fast you’re moving or where you’re looking from.
- Because light’s speed is fixed, **time and space are not absolute**. They bend to keep that speed constant.

### What that means
| What you’re used to | What actually happens |
|---------------------|-----------------------|
| Time is the same everywhere | Time **slows down** for a fast‑moving clock (time dilation). |
| Length is the same everywhere | Moving objects **shrink** in the direction of motion (length contraction). |
| Simultaneity is universal | Events that look simultaneous to one observer may not be to another. |

**Why it matters:**  
- If you travel in a spaceship at 99 % the speed of light, you’ll age slower than people on Earth.  
- The famous “twin paradox” is a real, measurable effect.

---

## 2. General Relativity (1915)

### Gravity is geometry
- Instead of a mysterious “force” pulling objects, **mass curves the fabric of space‑time**.  
- Objects move along the straightest possible paths in this curved geometry (geodesics).  
- The more massive an object, the more it warps space‑time.

### Everyday consequences
| Classic picture | General‑relativistic picture |
|-----------------|------------------------------|
| Earth orbits the Sun because the Sun pulls on it. | The Sun’s mass bends space‑time; Earth follows a curved path (an orbit) that is “straight” in that curved geometry. |
| Light bends around a massive star. | Light follows the curved space‑time, so it appears bent to us. |
| Time runs slower near a massive body. | Clocks closer to Earth (or any massive object) tick slightly slower than clocks far away. |

**Why it matters:**  
- GPS satellites must correct for both special‑relativistic (speed) and general‑relativistic (gravity) time
```

---

### Example 3
**Question:** How does democracy function in modern governments?

**Response:**
```
<|channel|>analysis<|message|>Need to explain democratic functioning.<|end|><|start|>assistant<|channel|>final<|message|>### 1.  What “democracy” means today  
Modern democracies are usually **constitutional republics**: a written constitution, separation of powers, and a system of checks and balances.  
Key features:

| Feature | What it looks like in practice |
|---------|--------------------------------|
| **Popular sovereignty** | Citizens elect representatives or vote directly on key issues. |
| **Rule of law** | Laws apply equally; courts can strike down unconstitutional acts. |
| **Separation of powers** | Executive, legislature, judiciary operate independently. |
| **Protection of rights** | Constitutional guarantees (speech, assembly, due process). |
| **Pluralism & competition** | Multiple parties, interest groups, and civil society. |
| **Transparency & accountability** | Open government, audits, whistle‑blower protections. |

---

### 2.  How the system works in a typical modern state

| Stage | What happens | Typical institutions |
|-------|--------------|----------------------|
| **Election cycle** | Citizens vote for representatives (parliament, congress, local councils). | Electoral commissions, polling stations, ballot counting. |
| **Legislative process** | Bills are drafted, debated, amended, and voted on. | Parliament, committees, public hearings. |
| **Executive action** | The head of state/government implements laws, manages policy, and represents the country. | President/Prime Minister, cabinet, ministries. |
| **Judicial review** | Courts interpret laws and can invalidate unconstitutional acts. | Supreme Court, appellate courts, constitutional courts. |
| **Oversight & audit** | Independent bodies scrutinize government spending and performance. | Auditors General, anti‑corruption agencies, ombudsmen. |
| **Civil society & media** | Citizens, NGOs, and journalists hold power to account. | NGOs, watchdogs, press, social media. |

---

### 3.  Democratic mechanisms in action

| Mechanism | How it functions | Example |
|-----------|------------------|---------|
| **Direct democracy** | Citizens vote on specific policies (referenda, initiatives). | Swiss referenda, U.S. ballot initiatives in California. |
| **Representative democracy** | Elected officials make decisions on behalf of constituents. | U.S. Congress, UK Parliament. |
| **Deliberative democracy** | Structured discussion to reach consensus (citizen assemblies, deliberative polls). | Irish Citizens’ Assembly
```

---

## 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](https://openai.com/index/introducing-gpt-oss/)
- **Model Hub**: [GPT-OSS-20B on Hugging Face](https://huggingface.co/openai/gpt-oss-20b)
- **Expert Analysis Dataset**: [GPT-OSS-20B MoE Expert Activations](https://huggingface.co/datasets/AmanPriyanshu/GPT-OSS-20B-MoE-expert-activations)
- **Project Page**: [GPT-OSS MoE Expert Fingerprinting](https://amanpriyanshu.github.io/GPT-OSS-MoE-ExpertFingerprinting/)
- **GitHub Repository**: [OpenAI GPT-OSS](https://github.com/openai/gpt-oss)