Project: https://amanpriyanshu.github.io/GPT-OSS-MoE-ExpertFingerprinting/
This is a pruned variant of OpenAI's GPT-OSS-20B model, reduced to 10 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 | ~7.8B (pruned from 21B) |
| Original Experts per Layer | 32 |
| Pruned Experts per Layer | 10 |
| 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 10 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-7.8b-specialized-all-pruned-moe-only-10-experts",
torch_dtype=torch.bfloat16,
device_map="cpu",
trust_remote_code=True
)
tokenizer = AutoTokenizer.from_pretrained("AmanPriyanshu/gpt-oss-7.8b-specialized-all-pruned-moe-only-10-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-7.8b-specialized-all-pruned-moe-only-10-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-7.8b-specialized-all-pruned-moe-only-10-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 explain AI concept and workings.<|end|><|start|>assistant<|channel|>final<|message|>**Artificial Intelligence (AI)** is a set of computer‑based systems that can *under‑process* data, *reason* about it, and *act* in a way that mimakes human cognition. In short, AI is a machine that can “see” patterns, “learn” from those patterns, and “decate” (or act) based on the information it has.
---
## 1. What is AI?
| **Core Idea** | **Key Components** | **Examples** |
|--------------------------- |--------------------------- |--------------------------- |
| A system that can *recognize* and *use* information to produce useful results | • Sensors (vision, sound, text) • Algorithms (logic, math) • Memory (storage) • Learning (feedback) | • Chatbots (e.g., GPT‑4) • Image‑recognisers (e.g., facial‑detection) • Autonomous‑vehicles (e.g., self‑driving cars) |
---
## 2. How AI Works? (A simplified pipeline)
1. **Input → Data**
The system receives data from a user, a sensor, or a database.
*Example:* a text prompt, a set of images, a sensor‑read of a temperature.
2. **Pre‑Processing → Parsing**
The AI “under‑reads” the data, turning it into a structured format.
*Example:* a language model tokenises a sentence into words; a vision model extracts pixel coordinates.
3. **Pattern‑Detection → Recognition**
The system looks for *similarities* (patterns) across the data.
*Example:* a language model finds that “cat” and “dog” are both nouns; a vision model finds that all pixels share a 3‑pixel offset.
4. **Inference → Reasoning**
Using the patterns, the AI draws *conferences* (hyplications).
*Example:* “cat” and “dog” are both nouns → the model predicts that the user is asking about animals.
5. **Learning → Updating**
The AI stores the new inference and uses it again on new data.
*Example:* the model learns that “cat” is a noun and “dog” is a noun, and will remember this for future questions.
6. **Action → Output**
The AI produces a response (text, image,
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 a way of understanding how objects move in space, especially when they’re far apart. Think of it as a “slap” that makes the universe look different from the way we used to think it.
---
### 1. What we used to think before
- **Newton’s law of inertia**: If you push a ball forward, it keeps moving forward.
- In a world where the universe is a fixed, “centered” place (like a planet), this law works fine.
- But if the universe is a *moving* place (like a star that’s moving toward the center), the ball will actually move backward.
- The old way of describing motion (Newtonian mechanics) assumed the universe was a fixed center.
- **The “center” of the universe**: In the old picture, the universe was a fixed point (the “center”) and all objects moved away from that point. The center was at the top of the universe, so the universe was a *fixed* place.
---
### 2. What Einstein changed
- **The universe is a moving place**: The “center” of the universe is not fixed; it moves.
- Think of a star that’s moving toward the center. The center is moving too, so the universe is a *moving* place.
- **The law of inertia still holds**: If you push a ball forward, it still moves forward *relative to the center*.
- But because the center is moving, the ball’s motion is *relative* to the center, not to the universe as a whole.
- **The “slap”**: The universe is moving, so the center is moving. The relative motion of an object is the *difference* between the center’s motion and the object’s motion.
- That’s the “slap” that Einstein called.
---
### 3. How it works in space
- **Near the center** (e.g., a planet close to the center):
- The center’s motion is very small.
- The relative motion of an object is also very small.
- The universe looks almost the same as the old picture.
- **Far from the center** (e.g., a star far away):
- The center’s motion is large.
- The relative motion of an object
Question: How does democracy function in modern governments?
Response:
<|channel|>analysis<|message|>Need explain role of democracy in modern governments.<|end|><|start|>assistant<|channel|>final<|message|>**How Democracy Functions in Modern Governments**
---
### 1. The Core Idea
Democracy is the idea that *people* (or the *citry*) have a say in the direction of a state. In modern governments, that idea is expressed in a few ways:
| Feature | What it means | How it works in practice |
|---------|---------------------------|---------------------------|
| **Representative** | Politicians are elected by the people | Elections, parliamentary seats, etc. |
| **Participative** | Citizens can influence policy directly | Public hearings, town‑meetings, online polls |
| **Accountable** | Leaders must answer to the public | Media, watchdog NGOs, legal courts |
| **Legit‑law** | Laws must reflect the will of the people | Constitutional checks, democratic courts |
---
### 2. The Main Mechanisms
| Mechanism | Example | Why it matters |
|---------|---------------------------|---------------------------|
| **Elections** | 5‑party elections, 1‑v‑1 voting | Choose a leader who represents the majority |
| **Parpar‑System** | Parliament, Senate, House of Representatives | Legislators debate, vote on laws |
| **Checks & Bal** | Judiciary, Ombudsman, watchdog bodies | Prevent abuse, enforce rule of law |
| **Civil‑Rights** | Freedom of speech, assembly, press | Citizens can voice opinions, hold leaders accountable |
| **Budget & Finance** | Public‑budget committees, transparency portals | Citizens can see how money is spent |
---
### 3. How It Works in a Modern State
1. **Electors** – Citizens vote for a president, prime‑minister, or local mayor.
2. **Parpar** – The elected body debates laws, passes bills, and elects the next leader.
3. **Checks** – Courts and NGOs review laws, expose corruption, and hold leaders in court.
4. **Civil‑Rights** – Citizens write op‑pieces, hold rallies, and use social media to pressure leaders.
5. **Budget** – Citizens can see how money is spent, and can demand better spending.
---
### 4. The Strengths
- **Legitimacy** – Leaders are chosen by the people.
- **Checks** – Power is limited by law and public scrutiny.
- **Adaptability** – Laws
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}
}