You can deploy this model and start earning money today!

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

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

Views

No views yet

All 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 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	~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	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 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 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 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-all-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-all-pruned-moe-only-26-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-17.3b-specialized-all-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-all-pruned-moe-only-26-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:

| Human Skill | AI Equivalent |
|-------------|---------------|
| Perception (seeing, hearing) | Computer vision, speech recognition |
| Reasoning & Decision‑Making | Rule‑based systems, probabilistic inference |
| Learning & Adaptation | Machine learning, deep learning |
| Language Understanding | Natural language processing (NLP) |
| Physical Interaction | Robotics, autonomous vehicles |

In short, **AI is about giving machines the ability to observe, understand, learn, and act in ways that mimic or surpass human cognition.**

---

## How AI Works – The Core Ingredients

| Component | What It Does | Typical Algorithms / Models |
|-----------|--------------|------------------------------|
| **Data** | The raw material. AI learns patterns from examples. | Image pixels, text corpora, sensor streams |
| **Representation** | How information is encoded. | Feature vectors, embeddings, graphs |
| **Learning** | Adjusting internal parameters to fit data. | Gradient descent, back‑propagation, reinforcement learning |
| **Inference** | Using the learned model to make predictions or decisions. | Forward pass in a neural net, Bayesian inference |
| **Feedback Loop** | Improving the model over time. | Online learning, active learning, human‑in‑the‑loop |

### 1. **Data Collection & Pre‑processing**

- **Gather**: Raw data from sensors, text, images, etc.
- **Clean**: Remove noise, handle missing values.
- **Transform**: Normalize, encode categorical variables, augment data.

### 2. **Feature Engineering (or Representation Learning)**

- **Manual Features**: Domain experts craft descriptors (e.g., edge detectors in images).
- **Automatic Features**: Deep learning learns hierarchical representations directly from raw data.

### 3. **Model Selection**

| Task | Typical Models |
|------|----------------|
| Classification | Logistic regression, SVM, Random Forest, CNN |
| Regression | Linear regression, Gradient Boosting, RNN |
| Sequence Modeling | RNN, LSTM, Transformer |
| Decision Making | Markov Decision Process, Q‑learning, Policy Gradient |
| Generative Tasks | GANs, VAEs, Diffusion Models |

### 4. **Training (Learning)**

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 – in plain English**

---

### 1. Two main ideas

| **Special Relativity** | **General Relativity** |
|------------------------|------------------------|
| Deals with objects moving at constant speeds (especially close to the speed of light). | Deals with gravity and how it bends space and time. |

---

### 2. Special Relativity (1905)

| Concept | What it means |
|---------|---------------|
| **Speed of light is constant** | No matter of how fast you’re moving, light always travels at the same speed (≈ 300 000 km/s). |
| **Time and space are relative** | How fast time ticks and how long distances are depends on how fast you’re moving relative to something else. |
| **Mass grows with speed** | As you approach light speed, you become heavier and need more energy to keep accelerating. |
| **E = mc²** | Energy (E) and mass (m) are interchangeable; a tiny amount of mass can become a huge amount of energy. |

**Why it matters:**  
- Explains why a fast‑moving clock runs slower (time dilation).  
- Explains why a fast‑moving object’s length appears shortened (length contraction).  
- Gives the foundation for modern physics, GPS, particle accelerators, etc.

---

### 3. General Relativity (1915)

| Concept | What it means |
|---------|---------------|
| **Gravity is geometry** | Massive objects (planets, stars, black holes) curve the “fabric” of space‑time. |
| **Objects follow the curves** | A planet orbits the Sun because it’s following the curved path created by the Sun’s mass. |
| **Time runs slower near mass** | Clocks closer to a massive body tick more slowly than clocks farther away. |
| **Light bends around mass** | Light follows the curved space‑time, so it bends when passing near a massive object (gravitational lensing). |

**Why it matters:**  
- Predicts the bending of starlight by the Sun (confirmed during the 1917 eclipse).  
- Explains why GPS satellites need relativistic corrections.  
- Describes black holes, the expansion of the universe, and gravitational waves.

---

### 4. Everyday Takeaway

- **Speed matters:** The faster you go, the more time slows and

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 in Practice  
Modern democracies are not a single, monolithic system; they are a collection of institutions, norms, and processes that together create a government that is **accountable to its citizens**.  The core features that most contemporary democracies share are:

| Feature | Practical Manifestation |
|---------|------------------------|
| **Popular sovereignty** | Citizens elect representatives or vote directly on key issues. |
| **Rule of law** | Laws apply equally; no one is above the law. |
| **Separation of powers** | Legislative, executive, and judicial branches are distinct and can check one another. |
| **Free and fair elections** | Regular, transparent, and competitive elections with universal adult suffrage. |
| **Political pluralism** | Multiple parties, interest groups, and civil society organizations can compete and influence policy. |
| **Civil liberties** | Freedom of speech, assembly, press, and religion are protected. |
| **Accountability mechanisms** | Media, courts, audits, and public oversight bodies hold officials responsible. |

### 2.  How the System Works Day‑to‑Day  

| Stage | Key Actors | Typical Actions | Checks & Balances |
|-------|------------|-----------------|-------------------|
| **Policy agenda setting** | Citizens, NGOs, lobbyists, media | Raise issues, mobilize public opinion | Media scrutiny, public debate |
| **Legislative process** | Parliament/Consentus | Draft, debate, amend, vote on bills | Committee reviews, opposition scrutiny, public hearings |
| **Executive implementation** | President/Prime Minister, ministries | Enact laws, administer programs, negotiate treaties | Parliamentary oversight, budget approvals |
| **Judicial review** | Courts | Interpret laws, adjudicate disputes, strike down unconstitutional acts | Judicial independence, appellate review |
| **Election cycle** | Electoral commissions, political parties, voters | Campaigns, voting, vote counting | Observation missions, recounts, legal challenges |
| **Public accountability** | Media, watchdog NGOs, audit institutions | Investigate corruption, expose misconduct | Freedom of information laws, whistleblower protections |

### 3.  Modern Democratic Innovations  

| Innovation | How It Enhances Democracy |
|------------|---------------------------|
| **Direct democracy tools** (referenda, citizen initiatives) | Gives citizens a direct voice on specific policies. |
| **Digital platforms

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