Downloads · 30 days
18
4% of all-time downloads
AmanPriyanshu/gpt-oss-16.1b-specialized-all-pruned-moe-only-24-experts
gpt-oss-16.1b-specialized-all-pruned-moe-only-24-experts is a text generation model from AmanPriyanshu. Use it when you need the model to write or continue text. The card lists the license as apache-2.0.
Project: https://amanpriyanshu.github.io/GPT-OSS-MoE-ExpertFingerprinting/
Downloads · 30 days
18
4% of all-time downloads
All-time downloads
473
Public
Parameters
16.1B
32.3 GB on disk
Likes
1
Public
Click a slice to open those files.
.safetensors32.3 GB · 100%
From the Hugging Face model README
Project: https://amanpriyanshu.github.io/GPT-OSS-MoE-ExpertFingerprinting/
<div align="center">This is a pruned variant of OpenAI's GPT-OSS-20B model, reduced to 24 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 | ~16.1B (pruned from 21B) |
| Original Experts per Layer | 32 |
| Pruned Experts per Layer | 24 |
| 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 24 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-16.1b-specialized-all-pruned-moe-only-24-experts",
torch_dtype=torch.bfloat16,
device_map="cpu",
trust_remote_code=True
)
tokenizer = AutoTokenizer.from_pretrained("AmanPriyanshu/gpt-oss-16.1b-specialized-all-pruned-moe-only-24-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-16.1b-specialized-all-pruned-moe-only-24-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-16.1b-specialized-all-pruned-moe-only-24-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, types, how works.<|end|><|start|>assistant<|channel|>final<|message|>## What is Artificial Intelligence?
Artificial Intelligence (AI) is the field of computer science that focuses on building systems capable of performing tasks that, in humans, are considered “intelligent.” These tasks include:
| Task | Human‑like behavior | Typical AI approach |
|------|---------------------|---------------------|
| Recognizing faces | We see a face and instantly know it’s a person | Computer vision + deep‑learning classifiers |
| Translating language | We read a sentence in one language and understand it in another | Natural‑language processing (NLP) + transformer models |
| Playing chess | We evaluate positions and choose a move | Game‑theory + reinforcement learning |
| Driving a car | We sense the road, obey traffic rules, and avoid obstacles | Sensor fusion + control systems + deep‑learning perception |
In short, **AI is about creating machines that can perceive, reason, learn, and act in ways that mimic or surpass human intelligence.**
---
## How Does AI Work? (A High‑Level View)
1. **Data Collection & Representation**
- **Input**: Raw data (images, text, sensor streams, etc.).
- **Feature extraction**: Convert raw data into a structured format (e.g., pixel arrays, word embeddings, sensor vectors).
2. **Model Architecture**
- **Rule‑Based Systems**: Hand‑crafted logic (IF‑THEN rules). Good for narrow, well‑defined domains.
- **Statistical Models**: Probabilistic models (e.g., Bayesian networks, support vector machines).
- **Deep Neural Networks (DNNs)**: Layers of interconnected “neurons” that learn representations automatically. The most common today.
3. **Learning / Training**
- **Supervised Learning**: Model learns from labeled examples (e.g., “cat” vs. “dog”).
- **Unsupervised Learning**: Model discovers patterns without labels (e.g., clustering).
- **Reinforcement Learning**: Model learns by trial‑and‑error, receiving rewards or penalties (e.g., playing Atari games).
- **Transfer Learning**: Re‑use knowledge from one task to accelerate learning on another.
4. **Inference / Decision Making**
- Once trained, the model receives new input and produces an output (classification, prediction, action recommendation).
- In real‑time systems, inference must be fast
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, everyday language**
---
### 1. Two main ideas
| **Special Relativity** | **General Relativity** |
|------------------------|------------------------|
| Deals with objects moving at constant speeds (especially near the speed of light). | Deals with gravity and how it bends space and time. |
---
### 2. Special Relativity (1905)
1. **Speed of light is the same for everyone**
- No matter how fast you’re moving, if you shine a flashlight, the light still travels at about 300 000 km/s.
- This constancy forces the universe to “adjust” how we measure space and time.
2. **Space and time are not absolute**
- If you’re moving fast, you’ll find that:
* **Time slows down** for you compared to someone standing still (time dilation).
* **Lengths shrink** in the direction of motion (length contraction).
- These effects are tiny at everyday speeds but become noticeable as you approach the speed of light.
3. **Mass and energy are interchangeable**
- The famous equation \(E = mc^2\) says that a small amount of mass can be converted into a huge amount of energy (and vice versa).
---
### 3. General Relativity (1915)
1. **Gravity is not a force pulling objects**
- Think of a heavy ball placed on a rubber sheet.
- The ball creates a dip; a smaller ball rolling nearby will spiral toward it.
- In reality, mass and energy “curve” the four‑dimensional fabric of space‑time, and objects follow the curved paths (geodesics).
2. **The shape of space‑time depends on what’s inside it**
- A massive star warps space‑time around it; light passing near it will bend (gravitational lensing).
- The faster an object moves, the more it feels the curvature.
3. **Time runs differently in different gravitational fields**
- Near a massive object (like Earth or a black hole), clocks tick a little slower than far away.
- This is why astronauts in orbit age a fraction of a second less than people on Earth.
---
### 4. Everyday Take‑aways
- **Speed matters**: The faster you go, the more
Question: How does democracy function in modern governments?
Response:
<|channel|>analysis<|message|>Need explanation.<|end|><|start|>assistant<|channel|>final<|message|>**Democracy in Modern Governments: A Practical Overview**
| Key Feature | How It Works Today | Typical Examples |
|-------------|--------------------|------------------|
| **Popular Sovereignty** | Citizens hold ultimate power through voting, referenda, and civic participation. | U.S. presidential elections, Swiss referenda, EU citizens’ initiatives |
| **Rule of Law** | Laws are applied equally; no one is above the law. | Constitutional courts (e.g., U.S. Supreme Court), independent judiciary |
| **Separation of Powers** | Executive, legislative, and judicial branches check each other. | U.S. Congress vs. President vs. Supreme Court; Germany’s Bundestag, Bundesrat, and Bundesverfassungsgericht |
| **Free and Fair Elections** | Regular, competitive elections with transparent processes. | Parliamentary elections in the UK, general elections in India |
| **Political Pluralism** | Multiple parties and interest groups compete for power. | Multiparty systems in France, coalition governments in Italy |
| **Civil Liberties & Human Rights** | Freedom of speech, assembly, press, and minority rights protected. | Charter of Rights in Canada, European Convention on Human Rights |
| **Accountability & Transparency** | Mechanisms to hold officials responsible (audits, ombudsmen, whistle‑blower laws). | Freedom of Information Acts, anti‑corruption agencies |
| **Decentralization & Local Governance** | Power shared with regional or municipal bodies. | Federalism in the U.S., devolved governments in Spain (Catalonia, Basque Country) |
| **Citizen Engagement** | Beyond voting: public consultations, participatory budgeting, digital platforms. | e‑democracy tools in Estonia, participatory budgeting in Brazil |
---
### 1. **The Core Mechanisms**
1. **Elections**
* **Universal suffrage** (age‑based, sometimes gender‑based) ensures every adult can vote.
* **Electoral systems** (first‑past‑the‑post, proportional representation, mixed systems) shape how votes translate into seats.
* **Campaign finance laws** and **media regulation** aim to level the playing field.
2. **Constitutional Framework**
* A written constitution (or set of foundational laws) defines the structure, powers, and limits of government.
* Constitutional courts interpret the constitution and can strike down unconstitutional laws.
3. **Checks &
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}
}