You can deploy this model and start earning money today!

AmanPriyanshu/
gpt-oss-15.5b-specialized-science-pruned-moe-only-23-experts

Name: gpt-oss-15.5b-specialized-science-pruned-moe-only-23-experts
Availability: InStock

Views

No views yet

Science GPT-OSS Model (23 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 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 science 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	~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	Science

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 science evaluation tasks
Systematic Reduction: Reduced from 32 to 23 experts per layer
No Retraining: Direct removal without additional training steps

Performance & Applications

Pruning Benefits

Smaller Memory Footprint: 71.9% of original expert parameters
Reduced Computational Load: Fewer routing decisions during inference
Focused Capabilities: Retains experts relevant to science 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 science-specialized model leverages experts that showed high activation patterns during scientific reasoning tasks from GPQA (physics, chemistry, biology) and MMLU science domains. These experts demonstrate superior performance on complex scientific problem-solving and technical knowledge recall.

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 science tasks, we created this specialized model that maintains domain expertise while significantly reducing computational requirements from 32 to 23 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 23 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-15.5b-specialized-science-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-science-pruned-moe-only-23-experts")

# Generate with the model
messages = [
    {"role": "user", "content": "Explain the process of photosynthesis in plants."}
]

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-15.5b-specialized-science-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-science-pruned-moe-only-23-experts")

# Generate with the model
messages = [
    {"role": "user", "content": "Explain the process of photosynthesis in plants."}
]

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 science tasks:

Example 1

Question: Explain the process of photosynthesis in plants.

Response:

<|channel|>analysis<|message|>Need concise explanation.<|end|><|start|>assistant<|channel|>final<|message|>**Photosynthesis in Plants – A Step‑by‑Step Overview**

| **Stage** | **What Happens** | **Key Players** |
|-----------|------------------|----------------|
| 1. Light Capture | Sunlight hits chlorophyll‑rich pigments in the chloroplasts. | Chlorophyll a & b, accessory pigments (carotenoids). |
| 2. Water Splitting (Photolysis) | Light energy splits H₂O into H⁺, e⁻, and O₂. | Photosystem II (PSII), oxygen‑evolving complex. |
| 3. Electron Transport Chain (ETC) | Electrons travel through PSII → plastoquinone → cytochrome b₆f → plastocyanin → Photosystem I (PSI). | PSII, cytochrome b₆f, PSI, ATP synthase. |
| 4. ATP & NADPH Production | Proton gradient drives ATP synthase; electrons reduce NADP⁺ to NADPH. | ATP synthase, ferredoxin, NADP⁺ reductase. |
| 5. Carbon Fixation (Calvin Cycle) | CO₂ is fixed into 3‑phosphoglycerate (3‑PGA) by Rubisco, then converted to glyceraldehyde‑3‑phosphate (G3P). | Rubisco, G3P dehydrogenase, triose phosphate isomerase. |
| 6. Regeneration of Ribulose‑1,5‑Bisphosphate (RuBP) | G3P is split; some is used to regenerate RuBP, allowing the cycle to continue. | ATP, NADPH, various enzymes. |
| 7. Product Formation | G3P exits the cycle to form glucose, fructose, sucrose, starch, cellulose, etc. | Glycolytic enzymes, sucrose‑phosphate synthase, starch synthase. |

### Key Points

1. **Light‑Dependent Reactions** (Stages 1–4) occur in the thylakoid membranes and produce ATP and NADPH.
2. **Light‑Independent Reactions** (Calvin Cycle, Stage 5–7) occur in the stroma and use ATP/NADPH to fix CO₂ into sugars.
3. **Oxygen Evolution**: The splitting of water releases O₂, which is released into the atmosphere.
4.

Example 2

Question: What causes the greenhouse effect and how does it work?

Response:

<|channel|>analysis<|message|>Need explanation.<|end|><|start|>assistant<|channel|>final<|message|>### What Causes the Greenhouse Effect?

The greenhouse effect is a natural process that keeps Earth’s surface warm enough to support life. It is caused by certain gases in the atmosphere—called **greenhouse gases (GHGs)**—that absorb, trap, and re‑radiate infrared (IR) radiation emitted by the planet’s surface. The main greenhouse gases are:

| Gas | Typical Atmospheric Concentration | Primary Source |
|-----|-------------------------------------|----------------|
| **Water vapor (H₂O)** | ~0.3 % (variable) | Evaporation from oceans, lakes, plants |
| **Carbon dioxide (CO₂)** | ~0.04 % (≈ 410 ppm) | Fossil‑fuel combustion, deforestation, industrial processes |
| **Methane (CH₄)** | ~0.00018 % (≈ 1.8 ppm) | Agriculture (especially ruminants), wetlands, natural gas leaks |
| **Nitrous oxide (N₂O)** | ~0.00027 % (≈ 0.27 ppm) | Agricultural fertilizers, combustion, industrial processes |
| **Fluorinated gases (e.g., CFCs, HFCs, SF₆)** | Trace amounts | Refrigeration, air‑conditioning, industrial processes |

These gases are **transparent to visible sunlight** (the short‑wave radiation that reaches Earth) but **opaque to the long‑wave (infrared) radiation** that the Earth emits after absorbing that sunlight. That opacity is what creates the greenhouse effect.

---

### How the Greenhouse Effect Works (Step‑by‑Step)

1. **Solar Radiation Reaches Earth**  
   - The Sun emits a spectrum of electromagnetic radiation.  
   - About 30 % of the incoming solar energy is reflected back to space by clouds, ice, and the atmosphere.  
   - The remaining ~70 % is absorbed by the Earth’s surface (land, oceans, vegetation).

2. **Earth’s Surface Heats Up**  
   - The absorbed solar energy warms the surface.  
   - A warm surface emits thermal radiation (infrared, ~10 µm wavelength) back into the atmosphere.

3. **Greenhouse Gases Absorb Infrared Radiation**  
   - GHG molecules have vibrational modes that resonate with IR wavelengths.  
   - When a GHG molecule absorbs an IR photon, its internal energy increases

Example 3

Question: Describe the structure and function of DNA.

Response:

<|channel|>analysis<|message|>Need concise description.<|end|><|start|>assistant<|channel|>final<|message|>**DNA (Deoxyribonucleic Acid)** is the hereditary material that stores and transmits genetic information in living organisms. Its structure and function can be understood in two complementary layers: the *molecular architecture* and the *biological roles* it plays.

---

## 1. Structural Features

| Feature | Description | Biological Significance |
|---------|-------------|------------------------|
| **Double‑helix** | Two complementary strands wound around each other in a right‑handed helix. | Provides stability and a compact, yet flexible, storage form. |
| **Back‑bone** | Alternating sugar (deoxyribose) and phosphate groups. | Forms a rigid, negatively charged scaffold that protects the bases. |
| **Nucleobases** | Four nitrogenous bases: Adenine (A), Thymine (T), Cytosine (C), Guanine (G). | Encode genetic information; base pairing (A‑T, C‑G) ensures faithful replication. |
| **Base pairing** | Hydrogen bonds: 2 between A‑T, 3 between C‑G. | Drives complementary strand synthesis and accurate base pairing during replication and transcription. |
| **Major & Minor grooves** | Structural pockets that accommodate protein binding. | Allows specific interaction with transcription factors, polymerases, and repair enzymes. |
| **Super‑coiling** | Higher‑level winding of the helix. | Compacts DNA into chromosomes and facilitates processes like replication and transcription. |
| **Chromatin packaging** | DNA wrapped around histone octamers → nucleosomes → higher‑order fibers. | Regulates accessibility of genes and protects DNA from damage. |

---

## 2. Functional Roles

| Function | Mechanism | Outcome |
|----------|-----------|---------|
| **Genetic Information Storage** | Sequence of bases encodes proteins, RNAs, regulatory elements. | Determines phenotype, cellular function, and inheritance. |
| **Replication** | Semi‑conservative duplication via DNA polymerases; helicases unwind the helix; primase lays RNA primers. | Produces identical daughter chromosomes for cell division. |
| **Transcription** | RNA polymerase binds promoter, unwinds DNA, synthesizes mRNA complementary to the template strand. | Generates messenger RNA that directs protein synthesis. |
| **Translation (indirect)** | mRNA is translated by ribosomes into polypeptides. | Produces

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