Science GPT-OSS Model (10 Experts)
Project: https://amanpriyanshu.github.io/GPT-OSS-MoE-ExpertFingerprinting/
Introduction
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 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 | ~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 | 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 10 experts per layer
- No Retraining: Direct removal without additional training steps
Performance & Applications
Pruning Benefits
- Smaller Memory Footprint: 31.2% 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 10 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 10 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
from transformers import AutoModelForCausalLM, AutoTokenizer
import torch
# Load the specialized model on CPU
model = AutoModelForCausalLM.from_pretrained(
"AmanPriyanshu/gpt-oss-7.8b-specialized-science-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-science-pruned-moe-only-10-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
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-science-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-science-pruned-moe-only-10-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:
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 to explain photosynthesis process.<|end|><|start|>assistant<|channel|>final<|message|>**Photosynthesis in plants**
Plants convert light energy from sunlight into chemical energy (organic molecules) by using the chlorophyll‑pigred system. The process can be broken into three steps:
| Step | What happens | Key molecules | Key enzymes |
|---|---|---|---|
| 1. **Light absorption** | Chlorophyll‑pigred complexes (chlorophyll‑a, chlorophyll‑b, phe‑flin) absorb photons (400–700 nm) and convert excited electrons to a higher energy state (S₁). | • Chlorophyll‑a (a‑t‑chro) | • Light‑dependent enzymes (none) |
| 2. **Electron transfer** | Excited electrons are transferred from the S₁ state to the Q₀ state (via the “S₁ → Q₀” transition). The Q₀ state is a “ground” state but still higher in the energy level. | • Chlorophyll‑b (b‑t‑chro) | • Light‑dependent enzymes (none) |
| 3. **Chemical synthesis** | The Q₀ state is oxidated to form a “reduced” state (Q₀⁻). The reduced state is used to produce a “referred” state (Q₀⁺). The Q₀⁺ state is used to produce a “referred” state (Q₀⁺). The Q₀⁺ state is used to produce a “referred” state (Q₀⁺). The Q₀⁺ state is used to produce a “referred” state (Q�₀⁺). | • Chlorophyll‑a (a‑t‑chro) | • Light‑dependent enzymes (none) |
**Key overall steps**
1. **Light absorption** – Chlorophyll‑pigred complexes absorb light and convert excited electrons to a higher energy state.
2. **Electron transfer** – Excated electrons are transferred from the S₁ state to the Q₀ state.
3. **Chemical synthesis** – The Q₀ state is oxidated to form a “reduced” state (Q₀⁻). The reduced state is used to produce a “referred” state (Q₀⁺).
**Result**
The plant uses the light energy to produce a chemical energy
Example 2
Question: What causes the greenhouse effect and how does it work?
Response:
<|channel|>analysis<|message|>Need to explain greenhouse effect: greenhouse gases trap heat, cause warming. Provide explanation.<|end|><|start|>assistant<|channel|>final<|message|>### What causes the greenhouse effect and how it works?
---
## 1. What is the “greenhouse effect”?
The **greenhouse effect** (or **greenhouse warming**) is the phenomenon that the Earth’s surface temperature rises because the atmosphere traps the heat that the Earth emits.
It is a result of the **greenhouse gases** (mainly CO₂, CH₄, N₂O, and other trace gases) that absorb infrared radiation from the Sun and re‑emit it back to space, keeping the Earth warmer.
---
## 2. Why it happens – the “greenhouse” analogy
1. **Greenhouse** – a container that traps heat.
2. **Greenhouse gases** – molecules that absorb and re‑emit infrared radiation.
3. **Greenhouse effect** – the Earth’s surface temperature rises because the atmosphere behaves like a greenhouse.
---
## 3. The basic physics
| **Solar radiation** | The Sun emits a spectrum of visible and infrared light.
| **Atmospheric absorption** | Greenhouse gases absorb most of the infrared part of the spectrum.
| **Re‑emission** | The absorbed energy is re‑emitted as a black‑body radiation at the same temperature (≈ 300 K).
| **Net warming** | Because the atmosphere is not transparent to the Sun’s radiation, the Earth’s surface receives more energy than it loses, so the temperature rises.
---
## 4. The key steps
1. **Solar energy reaches the Earth** – the Sun’s radiation is absorbed by the surface.
2. **Some part of the energy is lost to space** – the Earth emits as a black‑body radiation.
3. **Greenhouse gases absorb** – the atmosphere absorbs a large fraction of the infrared radiation that the Earth emits.
4. **Re‑emission** – the absorbed energy is re‑emitted back to space, but because the atmosphere is opaque, it is trapped.
5. **Net warming** – the Earth’s surface temperature rises because the net energy absorbed exceeds the energy lost.
---
## 5. What causes the greenhouse effect
| **CO₂** | 40 % of of the greenhouse effect.
| **CH₄ (methane)** | 10 % of the effect.
|
Example 3
Question: Describe the structure and function of DNA.
Response:
<|channel|>analysis<|message|>Need to give explanation of DNA structure: double heline, 5 base pairs, 3 sugar, 2 phosphate, 1 nitrogen. Function: genetic code, replication, transcription, regulation.<|end|><|start|>assistant<|channel|>final<|message|>**DNA – the double‑heline de‑nucle‑acid**
| **Structure** | **Function** |
|---|---|
| **Double‑heline** | Two strands of a single‑strand DNA that wind around a hel‑turn. |
| **5 base pairs** | 5 nucle‑pairs (A‑T, C‑G, G‑T, A‑C, T‑G) that make the “base‑pair” pattern. |
| **3 sugar** | Three sugar‑bases (C, G, T) that are the “s‑s‑s” code. |
| **2 phosphate** | Two phosphate‑bases (P, P) that are the “P‑P” signal. |
| **1 nitrogen** | One nitrogen‑base (N) that is the “N” anchor. |
| **Total 10 nucle‑bases** | The 10‑base “code” that is the “DNA” itself. |
---
### What does DNA do?
1. **Genetic code** – The 10‑base pattern is the “gen‑code” that is the “gen‑code” that is the “gen‑code” that is the “gen‑code” that is the “gen‑code” that is the “gen‑code” that is the “gen‑code” that is the “gen‑code” that is the “gen‑code” that is the “gen‑code” that is the “gen‑code” that is the “gen‑code” that is the “gen‑code” that is the “gen‑code” that is the “gen‑code” that is the “gen‑code” that is the “gen‑code” that is the “gen‑code” that is the “gen‑code” that is the “gen‑code” that is the “gen‑code” that is the “gen‑code” that is the “gen‑code” that is the “gen‑code” that is the “gen‑code” that is the “gen‑code” that is the “gen‑code” that is the “gen‑code” that is the “gen‑code” that is the “gen‑code”
Citation
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}
}
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
- Downloads last month
- 4