You can deploy this model and start earning money today!

AmanPriyanshu/
gpt-oss-14.9b-specialized-all-pruned-moe-only-22-experts

Name: gpt-oss-14.9b-specialized-all-pruned-moe-only-22-experts
Availability: InStock

Views

No views yet

All GPT-OSS Model (22 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 22 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	~14.9B (pruned from 21B)
Original Experts per Layer	32
Pruned Experts per Layer	22
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 22 experts per layer
No Retraining: Direct removal without additional training steps

Performance & Applications

Pruning Benefits

Smaller Memory Footprint: 68.8% 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 22 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 22 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-14.9b-specialized-all-pruned-moe-only-22-experts", 
    torch_dtype=torch.bfloat16, 
    device_map="cpu", 
    trust_remote_code=True
)
tokenizer = AutoTokenizer.from_pretrained("AmanPriyanshu/gpt-oss-14.9b-specialized-all-pruned-moe-only-22-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-14.9b-specialized-all-pruned-moe-only-22-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-14.9b-specialized-all-pruned-moe-only-22-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 it works.<|end|><|start|>assistant<|channel|>final<|message|>## What Is Artificial Intelligence?

Artificial Intelligence (AI) is the field of computer science that builds systems capable of performing tasks that normally require human intelligence. These tasks include:

| Human‑like ability | Typical AI task |
|--------------------|----------------|
| Perception (seeing, hearing) | Computer vision, speech recognition |
| Reasoning (solving problems, making decisions) | Planning, game‑playing, medical diagnosis |
| Learning (improving from experience) | Machine learning, deep learning |
| Language (understanding, generating text) | Natural‑language processing |
| Interaction (communicating, collaborating) | Chatbots, virtual assistants |

AI is **not** a single technology but a collection of methods, algorithms, and architectures that together enable machines to act “intelligently.”

---

## How AI Works – The Big Picture

1. **Data Collection & Representation**  
   - AI systems need data: images, text, sensor readings, etc.  
   - Data is transformed into a format the algorithm can use (vectors, tensors, graphs).

2. **Model Selection**  
   - Choose a computational model that can learn from the data.  
   - Common families:  
     * **Statistical models** (regression, Bayesian networks)  
     * **Symbolic models** (logic, rule‑based systems)  
     * **Connectionist models** (neural networks)  
     * **Hybrid models** (combining the above)

3. **Training / Learning**  
   - The model adjusts internal parameters to minimize a *loss function* (error between prediction and truth).  
   - **Supervised learning**: labeled data (e.g., “cat” vs. “dog”).  
   - **Unsupervised learning**: discover patterns without labels (e.g., clustering).  
   - **Re‑inforcement learning**: learn by trial‑and‑error with rewards.

4. **Inference / Decision Making**  
   - Once trained, the model receives new input and produces an output (classification, prediction, action).  
   - In some systems, the output is a *policy* that tells the system what to do next.

5. **Evaluation & Feedback**  
   - Measure performance using metrics (accuracy, precision, recall, F1, etc.).  
   - If performance is unsatisfactory, iterate: gather more data, tweak the model, or try a

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, everyday 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 – “Time and Space are Flexible”

1. **Speed of Light is a Cosmic Speed‑Limit**  
   - Light always travels at the same speed (≈ 300 000 km/s) no matter who’s looking at it.  
   - Nothing can go faster than light.

2. **Time Slows Down When You Move Fast**  
   - Imagine you’re on a very fast train. To someone standing on the platform, your watch ticks slower.  
   - The farther you get to the speed of light, the more your time “stretches.”  
   - This is called **time dilation**.

3. **Length Shrinks in the Direction of Motion**  
   - A moving object looks shorter in the direction it’s moving.  
   - This is **length contraction**.

4. **No Absolute “Now”**  
   - Two observers moving relative to each other can disagree about when an event happens.  
   - There isn’t a single, universal “now” that everyone shares.

**Bottom line:** If you’re moving very fast, your clock runs slower and distances in front of you shrink. But you’ll still see the same physics – the same laws of motion, the same speed of light.

---

### 3. General Relativity – “Gravity is Curved Space‑Time”

1. **Space‑Time is a Fabric**  
   - Think of space and time as a stretchy sheet.  
   - Massive objects (planets, stars, black holes) dent this sheet.

2. **Gravity is the Result of the Dents**  
   - A ball rolls toward the dent because the sheet is curved.  
   - In the universe, objects “roll” toward each other because the space‑time around them is curved.

3. **Light Follows the Curved Sheet**  
   - Light rays bend around massive objects (gravitational lensing).  
   - This is why distant galaxies can appear magnified or distorted.

4. **Time Runs Differently Where Gravity is Strong**

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  
Modern democracies are not a single, one‑style system; they are a family of institutions and practices that share a few core principles:

| Core Principle | What It Looks Like in Practice |
|----------------|------------------------------------|
| **Popular Sovereignty** | The people are the ultimate source of political legitimacy. |
| **Rule of Law** | All citizens, including leaders, are bound by the same laws. |
| **Political Pluralism** | Multiple parties, interest groups, and civil‑society actors compete for influence. |
| **Free and Fair Elections** | Regular, secret, and inclusive voting that genuinely reflects the will of the electorate. |
| **Separation of Powers** | Executive, legislative, and judicial branches are distinct and can check one another. |
| **Civil Liberties** | Freedom of speech, assembly, press, and religion are protected. |
| **Accountability & Transparency** | Governments must explain their actions and be open to scrutiny. |

These principles are the scaffolding that modern democratic governments use to organize themselves.

---

### 2.  Institutional Architecture

| Branch | Key Functions | Typical Checks |
|--------|---------------|----------------|
| **Executive** (President/Prime‑Minister & Cabinet) | Implements laws, runs day‑to‑day administration, foreign policy | Veto power, impeachment, parliamentary confidence |
| **Legislature** (Parliament/ Congress) | Drafts, debates, and passes laws | Oversight hearings, budget control, confirmation of appointments |
| **Judiciary** (Courts) | Interprets laws, protects rights, resolves disputes | Judicial review, appellate review, independence from politics |
| **Electoral Bodies** (Independent commissions) | Organize elections, enforce campaign finance rules | Audits, dispute resolution |
| **Civil‑Society & Media** | Represent interests, hold power accountable | Investigative journalism, NGOs, public protests |

---

### 3.  The Everyday Life of a Democratic State

1. **Policy Formation**  
   * Legislators propose bills → committee reviews → public hearings → debate → vote.  
   * The executive can propose legislation, but it must be approved by the legislature.

2. **Checks & Balances**  
   * A law passed by the legislature can be vetoed by the executive (often overridden by a super‑majority

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