You can deploy this model and start earning money today!

AmanPriyanshu/
gpt-oss-13.7b-specialized-all-pruned-moe-only-20-experts

Name: gpt-oss-13.7b-specialized-all-pruned-moe-only-20-experts
Availability: InStock

Views

No views yet

All GPT-OSS Model (20 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 20 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	~13.7B (pruned from 21B)
Original Experts per Layer	32
Pruned Experts per Layer	20
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 20 experts per layer
No Retraining: Direct removal without additional training steps

Performance & Applications

Pruning Benefits

Smaller Memory Footprint: 62.5% 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 20 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 20 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-13.7b-specialized-all-pruned-moe-only-20-experts", 
    torch_dtype=torch.bfloat16, 
    device_map="cpu", 
    trust_remote_code=True
)
tokenizer = AutoTokenizer.from_pretrained("AmanPriyanshu/gpt-oss-13.7b-specialized-all-pruned-moe-only-20-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-13.7b-specialized-all-pruned-moe-only-20-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-13.7b-specialized-all-pruned-moe-only-20-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 field of computer science that focuses on building systems—usually software running on computers—that can perform tasks that normally require human intelligence. These tasks include:

| Task | Human Intelligence Needed | AI Equivalent |
|------|---------------------------|--------------|
| Recognizing faces | Visual perception | Computer vision |
| Translating languages | Linguistic understanding | Natural‑language processing |
| Playing chess | Strategic planning | Game‑playing algorithms |
| Diagnosing disease | Medical knowledge | Clinical decision support |
| Driving a car | Perception, decision‑making | Autonomous driving |

In short, **AI is the study and creation of machines that can learn, reason, and act in ways that mimic or surpass human cognition.**

---

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

AI systems are built from a few core ingredients that work together:

| Ingredient | What it does | Typical Technology |
|------------|--------------|---------------------|
| **Data** | The raw “experience” the system learns from | Images, text, sensor readings, etc. |
| **Representation** | A way to encode data so a computer can manipulate it | Vectors, graphs, tensors |
| **Learning Algorithm** | Adjusts the system’s internal parameters to fit the data | Gradient descent, back‑propagation |
| **Inference Engine** | Uses the learned model to answer new questions | Forward‑propagation, rule‑based engines |
| **Evaluation** | Measures how well the system performs | Accuracy, loss functions, cross‑validation |

Below is a step‑by‑step outline of how most modern AI systems (especially deep learning models) are built and used.

---

### 1. **Collecting and Preparing Data**

- **Data Collection**: Gather a large, representative set of examples (e.g., thousands of labeled images of cats and dogs).
- **Pre‑processing**: Clean the data, normalize it, and transform it into a format the algorithm can use (e.g., converting pixel values to a 0‑1 range).
- **Feature Extraction** (optional): Identify useful attributes (e.g., edges, colors) that help the model focus on relevant information.

### 2. **Choosing a Model Architecture**

- **Rule‑Based Systems**: Hand‑crafted logic (e.g., if‑then rules). Good for narrow tasks with clear rules.
- **Stat

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 words**

Imagine you’re on a train that’s moving at a constant speed. If you look at the world outside, it seems to be moving *against* you. But if you’re inside the train, you feel at rest. The same idea applies to how we measure space and time, but on a much bigger scale.

---

### 1. **Two parts of the theory**

| Part | What it says | Everyday picture |
|------|--------------|-------------------|
| **Special Relativity** | *“The laws of physics look the same no matter how fast you’re moving.”* | If you’re on a fast jet, the clocks inside the jet tick the same way as on the ground. |
| **General Relativity** | *“Gravity is just a way that massive objects bend the fabric of space‑time.”* | A heavy planet warps the space around it, making objects near it move differently than far away. |

---

### 2. **Key ideas**

| Idea | Simple explanation |
|------|---------------------|
| **Space and time are one thing (space‑time)** | Think of a 4‑dimensional sheet. Moving forward in time is like moving along the sheet. |
| **Speed changes how you see the world** | The faster you go, the more “slanted” the sheet looks. Things that are normally simultaneous can appear out of order. |
| **Light moves at a fixed speed (c)** | Nothing can go faster than light. That speed is the same for everyone, no matter where you are. |
| **Mass curves space‑time** | A big mass (planet, star) pulls the sheet around it, making paths bend. That’s what we feel as gravity. |

---

### 3. **Why it matters**

* **Time dilation** – A clock on a fast spaceship ticks *slower* than a clock on Earth.  
* **Gravitational redshift** – Light that climbs away from a massive body gets *slower* (redder).  
* **Black holes** – Extremely massive objects curve space‑time so much that even light can’t escape.  

---

### 4. **A quick “thought experiment”**

1. **Twin Paradox** – Two identical twins: one stays on Earth, the other travels at near‑light speed. When the traveler returns, they’re

Example 3

Question: How does democracy function in modern governments?

Response:

<|channel|>analysis<|message|>Need explain democratic functions, institutions, processes.<|end|><|start|>assistant<|channel|>final<|message|>### 1.  What “Democracy” Means Today  
In contemporary politics “democracy” is usually understood as a **plurality‑based, rule‑by‑law system** in which the people are the ultimate source of political authority.  Modern democracies combine several core features:

| Feature | What It Looks Like in Practice | Why It Matters |
|--------|---------------------------------|----------------|
| **Free, fair elections** | Regular, competitive contests for office, with universal adult sufficiency, secret ballots, and independent electoral bodies | Gives the populace a direct, periodic voice |
| **Rule of law** | Constitution, statutes, and courts that bind everyone, including the government | Prevents arbitrary rule and protects minorities |
| **Separation of powers** | Distinct branches (executive, legislative, judicial) with checks and balances | Keeps power from concentrating in one office |
| **Civil liberties** | Freedom of speech, press, assembly, religion, and privacy | Enables dissent, debate, and accountability |
| **Independent institutions** | Electoral commissions, judiciary, ombudsman, audit offices | Provides impartial oversight |
| **Pluralism & minority rights** | Legal protection for ethnic, religious, gender, and other minorities | Ensures that majority rule does not become tyranny |
| **Transparency & accountability** | Open records, freedom of information, anti‑corruption laws | Keeps officials answerable to citizens |

These elements are not a checklist but a **framework** that modern governments adapt to their own histories, cultures, and institutions.

---

### 2.  How the Framework Works in Practice  

#### 2.1  The Legislative Branch  
* **Parliament/ Congress** – elected representatives debate, amend, and pass laws.  
* **Committee System** – specialized groups scrutinize bills, budgets, and executive actions.  
* **Budget & Oversight** – the legislature controls public spending and can hold the executive accountable through hearings, audits, and, in some systems, impeachment.

#### 2.2  The Executive Branch  
* **Head of State / Government** – elected (presidential) or chosen by the legislature (parliamentary).  
* **Cabinet & Ministries** – implement laws, manage public services, and propose new legislation.  
* **Executive Accountability** – must answer to the legislature, submit annual reports, and can be removed by a vote of no confidence or impeachment.

#### 2.

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