Project: https://amanpriyanshu.github.io/GPT-OSS-MoE-ExpertFingerprinting/
This is a pruned variant of OpenAI's GPT-OSS-20B model, reduced to 23 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.
| Metric | Value |
|---|---|
| Base Model | openai/gpt-oss-20b |
| Architecture | Mixture-of-Experts Transformer |
| Total Parameters | ~15.5B (pruned from 21B) |
| Original Experts per Layer | 32 |
| Pruned Experts per Layer | 23 |
| 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 |
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:
Note: Performance may vary depending on how well the pruned experts match your specific use case.
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:
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 23 experts per layer.
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.
Our approach involves:
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.
from transformers import AutoModelForCausalLM, AutoTokenizer
import torch
# Load the specialized model on CPU
model = AutoModelForCausalLM.from_pretrained(
"AmanPriyanshu/gpt-oss-15.5b-specialized-all-pruned-moe-only-23-experts",
torch_dtype=torch.bfloat16,
device_map="cpu",
trust_remote_code=True
)
tokenizer = AutoTokenizer.from_pretrained("AmanPriyanshu/gpt-oss-15.5b-specialized-all-pruned-moe-only-23-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)
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-15.5b-specialized-all-pruned-moe-only-23-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-15.5b-specialized-all-pruned-moe-only-23-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)
For CUDA GPU inference, the models work with minimal modifications from the CPU example above. Simply change:
device_map="auto" # Will automatically use GPU if available
torch_dtype=torch.bfloat16 # or torch.float16
The following examples demonstrate the model's performance on all tasks:
Question: What is artificial intelligence and how does it work?
Response:
<|channel|>analysis<|message|>Need concise explanation.<|end|><|start|>assistant<|channel|>final<|message|>**Artificial Intelligence (AI)** is a field of computer science that builds systems capable of performing tasks that normally require human intelligence. These tasks include:
| Task | Typical Human Skill | AI Approach |
|------|---------------------|-------------|
| Recognizing faces | Visual perception | Pattern‑matching, deep‑learning CNNs |
| Translating language | Linguistic understanding | Statistical models, neural language models |
| Driving a car | Decision‑making, sensor fusion | Reinforcement learning, rule‑based control |
| Diagnosing disease | Medical knowledge, reasoning | Probabilistic inference, deep‑learning classifiers |
---
### 1. What AI *does*
- **Perception** – interprets raw data (images, sounds, text).
- **Reasoning** – draws conclusions, plans, or solves problems.
- **Learning** – improves performance over time by adjusting internal parameters.
- **Acting** – produces outputs (text, actions, predictions).
---
### 2. How AI *works* – the core ingredients
| Component | Key Idea | Example |
|-----------|----------|---------|
| **Data** | The “fuel” that trains models. | Thousands of labeled images for face recognition. |
| **Model** | A mathematical representation of knowledge. | Neural networks, decision trees, Bayesian networks. |
| **Learning algorithm** | Optimizes the model’s parameters. | Gradient descent, back‑propagation, gradient‑boosted trees. |
| **Loss / objective function** | Measures how far the model is from the desired output. | Mean‑squared error, cross‑entropy. |
| **Optimization** | Finds parameter values that minimize loss. | Stochastic gradient descent, Adam, RMSprop. |
| **Evaluation** | Tests generalization on unseen data. | Cross‑validation, ROC curves. |
---
### 3. The typical AI workflow
1. **Problem definition** – decide what the system should do.
2. **Data collection & preprocessing** – clean, label, and format data.
3. **Model selection** – choose a suitable architecture (e.g., CNN for vision).
4. **Training** – run the learning algorithm on the data.
5. **Validation & hyper‑parameter tuning** – adjust learning rates, regularization, etc.
6. **Deployment** – integrate the trained model into an application.
7. **Monitoring & retraining**
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 language**
---
### 1. Two main ideas
| **Special Relativity** | **General Relativity** |
|------------------------|------------------------|
| Deals with objects moving at high speeds (close to the speed of light). | Deals with gravity and how it bends space and time. |
---
### 2. Special Relativity (1905)
| Concept | What it means |
|---------|---------------|
| **Time slows down for fast‑moving objects** | If you’re on a very fast spaceship, your clock ticks a little slower than a clock on Earth. |
| **Lengths shrink in the direction of motion** | A ruler moving at high speed looks shorter to an outside observer. |
| **Speed of light is the same for everyone** | No matter how fast you’re moving, light always travels at the same speed (≈ 300 000 km/s). |
| **Mass increases with speed** | A fast‑moving object behaves as if it’s heavier. |
**Why?**
Einstein realized that the usual rules of space and time (what we call “Newtonian physics”) break down when you go near light speed. He rewrote the laws so that the speed of light stays constant for everyone, and the weird effects (time dilation, length contraction, mass increase) come out naturally.
---
### 3. General Relativity (1918)
| Concept | What it means |
|---------|---------------|
| **Gravity is not a force but a curvature of space‑time** | Think of space‑time as a stretchy sheet. Heavy objects (planets, stars) dent the sheet; other objects follow the curved paths. |
| **“The closer you are to a massive body, the slower your clock runs”** | Time near the Earth runs a tiny bit slower than far away. |
| **Light follows curved paths** | Light bends around massive objects (gravitational lensing). |
| **The universe can be described by one set of equations** | Einstein’s equations tell you how mass and energy shape space‑time and how that shape tells matter how to move. |
**Why?**
Einstein noticed that the way clocks tick and how objects move near massive bodies could be explained if we let space‑time itself be flexible. Instead of a “force” pulling things together, we say that mass tells space‑time to curve, and
Question: How does democracy function in modern governments?
Response:
<|channel|>analysis<|message|>Need to explain democratic functioning.<|end|><|start|>assistant<|channel|>final<|message|>**How Democracy Functions in Modern Governments**
Modern democracies are built on a set of interlocking institutions, processes, and norms that together create a system in which power ultimately rests with the people. While the specific arrangements vary from country to country, most contemporary democracies share a core set of features that enable them to function effectively. Below is a step‑by‑step overview of how these features work together in practice.
---
## 1. **Foundational Principles**
| Principle | What It Means | How It Operates in Practice |
|-----------|---------------|------------------------------|
| **Popular Sovereignty** | The people are the ultimate source of political authority. | Elections, referendums, and public consultations give citizens a direct say in who governs and what laws are adopted. |
| **Rule of Law** | All actions of government must be grounded in law, and everyone is subject to the same legal standards. | Constitutions, independent courts, and legal codes prevent arbitrary rule. |
| **Separation of Powers** | Legislative, executive, and judicial branches are distinct and can check one another. | Parliament enacts laws, the president or prime minister executes them, and courts interpret and enforce them. |
| **Political Pluralism** | Multiple political parties and interest groups compete for influence. | Elections, campaign finance laws, and media access allow diverse voices to be heard. |
| **Civil Liberties** | Freedom of speech, assembly, religion, and privacy are protected. | Constitutional guarantees, human‑rights commissions, and watchdog NGOs safeguard individual rights. |
---
## 2. **Institutional Architecture**
### 2.1 The Legislature
| Feature | Function |
|---------|----------|
| **Bicameral or Unicameral** | Two chambers (e.g., House of Commons & Senate) or one chamber debate and pass laws. |
| **Committee System** | Specialized committees scrutinize bills, conduct hearings, and oversee agencies. |
| **Quorum & Voting Rules** | Ensure that decisions reflect a legitimate majority. |
### 2.2 The Executive
| Feature | Function |
|---------|----------|
| **Head of State** (President, Monarch, etc.) | Symbolic or substantive role in representing the nation. |
| **Head of Government** (Prime Minister, Chancellor) | Leads the cabinet, sets policy agenda, and implements laws. |
| **Cabinet** | Ministers oversee specific portfolios
If you use this model in your research, please cite:
@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}
}