You can deploy this model and start earning money today!

AmanPriyanshu/
gpt-oss-18.5b-specialized-all-pruned-moe-only-28-experts

Name: gpt-oss-18.5b-specialized-all-pruned-moe-only-28-experts
Availability: InStock

Views

No views yet

All GPT-OSS Model (28 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 28 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 all 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	~18.5B (pruned from 21B)
Original Experts per Layer	32
Pruned Experts per Layer	28
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:

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 all evaluation tasks
Systematic Reduction: Reduced from 32 to 28 experts per layer
No Retraining: Direct removal without additional training steps

Performance & Applications

Pruning Benefits

Smaller Memory Footprint: 87.5% 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 28 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 28 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-18.5b-specialized-all-pruned-moe-only-28-experts", 
    torch_dtype=torch.bfloat16, 
    device_map="cpu", 
    trust_remote_code=True
)
tokenizer = AutoTokenizer.from_pretrained("AmanPriyanshu/gpt-oss-18.5b-specialized-all-pruned-moe-only-28-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-18.5b-specialized-all-pruned-moe-only-28-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-18.5b-specialized-all-pruned-moe-only-28-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 the field of computer science that seeks to build systems capable of performing tasks that normally require human intelligence. These tasks include:

| Task | Human‑like AI capability | Typical AI technique |
|------|--------------------------|-----------------------|
| Recognizing faces | “I can see a face” | Convolutional Neural Networks (CNNs) |
| Understanding language | “I can read and respond” | Natural Language Processing (NLP) models |
| Playing chess | “I can plan moves” | Search algorithms + reinforcement learning |
| Driving a car | “I can navigate safely” | Sensor fusion + deep reinforcement learning |
| Diagnosing disease | “I can suggest a diagnosis” | Probabilistic models + deep learning |

In short, **AI is about creating machines that can perceive, reason, learn, and act in ways that mimic or surpass human cognition.**

---

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

1. **Data Collection & Representation**
   - **Input**: Raw data (images, text, sensor readings, etc.).
   - **Pre‑processing**: Cleaning, normalizing, augmenting.
   - **Feature extraction**: Turning raw data into useful representations (e.g., pixel grids, word embeddings).

2. **Model Selection**
   - Choose an algorithm that best fits the problem:
     - **Rule‑based systems** (if‑then logic) for simple, deterministic tasks.
     - **Statistical models** (e.g., logistic regression, Naïve Bayes) for probabilistic inference.
     - **Machine learning** (supervised, unsupervised, reinforcement) for pattern discovery.
     - **Deep learning** (neural networks) for high‑dimensional, complex data.

3. **Training / Learning**
   - **Supervised learning**: Feed labeled examples; the model learns to map inputs → outputs.
   - **Unsupervised learning**: Discover structure without labels (clustering, dimensionality reduction).
   - **Reinforcement learning**: Learn by trial‑and‑error, receiving rewards for good actions.
   - **Transfer learning**: Fine‑tune a pre‑trained model on a new task.

4. **Inference / Decision Making**
   - Once trained, the model processes new inputs and produces outputs (classification, regression, generation, etc.).
   -

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 two postulates

1. **The laws of physics are the same for everyone who is moving at a constant speed in a straight line (in “inertial” frames).**  
   – No experiment can tell you whether you’re at rest or moving at a steady speed, as long as you’re not turning or accelerating.

2. **The speed of light in a vacuum is always the same (≈ 299 792 km/s), no matter how fast you’re moving or how fast the light source is moving.**  
   – Light is a universal speed limit.

### What that means

- **Time slows down for fast‑moving clocks** (time dilation).  
  If you travel in a spaceship at 99 % the speed of light, your clock ticks slower than a clock on Earth.

- **Lengths contract in the direction of motion** (length contraction).  
  The spaceship would appear shorter to an observer on Earth.

- **Mass increases with speed** (relativistic mass).  
  A particle needs more energy to accelerate as it approaches light speed.

- **Energy and mass are interchangeable**: \(E = mc^2\).  
  A tiny amount of mass can be converted into a huge amount of energy (nuclear power, bombs).

---

## 2. General Relativity (1915)

### The key idea

**Gravity is not a force pulling objects together; it is the curvature of spacetime caused by mass and energy.**  
Think of spacetime as a stretchy sheet. A heavy ball (the Sun) sits on the sheet and creates a dip. A smaller ball (Earth) rolls around that dip, following a curved path that we call an orbit.

### Consequences

- **Light bends around massive objects** (gravitational lensing).  
  This was confirmed during a 1911 solar eclipse.

- **Time runs slower near massive bodies** (gravitational time dilation).  
  GPS satellites must correct their clocks for this effect to stay accurate.

- **The universe is dynamic**: space can expand or contract, leading to cosmology (Big Bang

Example 3

Question: How does democracy function in modern governments?

Response:

<|channel|>analysis<|message|>Need explain democratic functioning.<|end|><|start|>assistant<|channel|>final<|message|>### 1.  The Core Idea of Democracy

At its heart, democracy is a system of government in which **power ultimately resides with the people**.  
Modern democracies translate that principle into a set of institutions, processes, and norms that:

| Element | What it Looks Like in Practice | Why It Matters |
|---------|---------------------------------|----------------|
| **Popular sovereignty** | Citizens vote for representatives or decide on laws directly (referenda). | Ensures that the government reflects the will of the populace. |
| **Rule of law** | Laws are applied equally; no one is above the law. | Prevents arbitrary rule and protects individual rights. |
| **Separation of powers** | Executive, legislative, and judicial branches are distinct and can check one another. | Avoids concentration of power and promotes accountability. |
| **Free and fair elections** | Regular, transparent, competitive elections with universal adult suffrage. | Guarantees that leaders are chosen by the people. |
| **Civil liberties** | Freedom of speech, assembly, press, religion, etc. | Enables citizens to express dissent and hold leaders accountable. |
| **Pluralism & minority rights** | Protection of diverse groups and their interests. | Prevents majoritarian tyranny. |
| **Transparency & accountability** | Open government, public records, independent media, watchdog institutions. | Keeps officials answerable to the public. |

---

### 2.  How These Elements Operate in Modern Governments

| Institutional Layer | Typical Mechanisms | Example |
|---------------------|-------------------|---------|
| **Legislature** | Bills are drafted, debated, amended, and voted on. Committees scrutinize details. | U.S. Congress, German Bundestag, Indian Lok Sabha |
| **Executive** | Head of state (president or monarch) and head of government (prime minister) implement laws, manage day‑to‑day governance, and represent the country internationally. | U.S. President, German Chancellor, Japanese Prime Minister |
| **Judiciary** | Courts interpret laws, adjudicate disputes, and can strike down unconstitutional acts. | U.S. Supreme Court, UK Supreme Court, Constitutional Court of South Africa |
| **Electoral Commission / Election Authority** | Organizes elections, ensures fairness, counts votes, resolves disputes. | Election Commission of India, Federal Election Commission (US) |
| **Civil Society & Media** | NGOs, think‑tanks

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