Project: https://amanpriyanshu.github.io/GPT-OSS-MoE-ExpertFingerprinting/
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 instruction following 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 | ~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 | Instruction Following |
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 instruction-following model leverages experts that excelled at constraint satisfaction tasks from Tulu3 Persona Instruction Following dataset. These experts specialize in precise adherence to user specifications and formatting requirements.
The expert selection process utilized our comprehensive analysis of router activation patterns across multiple evaluation benchmarks:
By identifying experts that consistently activated for instruction following tasks, we created this specialized model that maintains domain expertise while significantly reducing computational requirements from 32 to 22 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-14.9b-specialized-instruction_following-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-instruction_following-pruned-moe-only-22-experts")
# Generate with the model
messages = [
{"role": "user", "content": "Write a formal email to a professor requesting a meeting, including: subject line, greeting, purpose, proposed times, and professional closing."}
]
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-14.9b-specialized-instruction_following-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-instruction_following-pruned-moe-only-22-experts")
# Generate with the model
messages = [
{"role": "user", "content": "Write a formal email to a professor requesting a meeting, including: subject line, greeting, purpose, proposed times, and professional closing."}
]
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 instruction following tasks:
Question: Write a formal email to a professor requesting a meeting, including: subject line, greeting, purpose, proposed times, and professional closing.
Response:
<|channel|>analysis<|message|>Need formal email.<|end|><|start|>assistant<|channel|>final<|message|>**Subject:** Request for Meeting – Discussion of Thesis Progress
Dear Professor [Last Name],
I hope this message finds you well. I am writing to request a brief meeting with you to discuss the current status of my thesis and to seek your guidance on the next steps.
Could we meet at one of the following times next week?
- Tuesday, August 20 at 10:00 a.m.
- Wednesday, August 21 at 2:00 p.m.
- Thursday, August 22 at 11:30 a.m.
If any of these slots are inconvenient, I would be happy to adjust to a time that better fits your schedule.
Thank you for your time and consideration. I look forward to your response.
Sincerely,
[Your Full Name]
[Your Program / Department]
[University]
[Phone Number] | [Email Address]<|return|>
Question: Create a numbered list of steps for making scrambled eggs, using exactly 6 steps.
Response:
<|channel|>analysis<|message|>Need 6 steps numbered.<|end|><|start|>assistant<|channel|>final<|message|>1. **Gather ingredients and tools** – You’ll need eggs, a splash of milk or water, butter or oil, salt, pepper, and a non‑stick skillet or frying pan.
2. **Heat the pan** – Place the skillet over medium heat and add a small pat of butter or a drizzle of oil, letting it melt and coat the surface.
3. **Whisk the eggs** – Crack the eggs into a bowl, add a tablespoon of milk or water, a pinch of salt and pepper, then whisk until the mixture is uniform.
4. **Pour the mixture into the pan** – Once the butter/oil is hot and shimmering, pour the whisked eggs into the skillet, letting them spread evenly.
5. **Cook and stir gently** – Allow the eggs to set for a few seconds, then use a spatula to stir gently, scraping the bottom and sides of the pan to prevent sticking. Continue cooking until the eggs are softly set but still slightly runny.
6. **Serve immediately** – Remove the pan from heat, transfer the scrambled eggs to a plate, and enjoy while warm.<|return|>
Question: Summarize the water cycle in exactly 50 words, no more, no less.
Response:
<|channel|>analysis<|message|>Need 50 words. Let's craft: "Water evaporates from oceans, lakes, and soil, forming clouds. Precipitation returns as rain or snow, filling rivers and groundwater. Plants absorb water, releasing vapor through transpiration. Groundwater moves through aquifers, emerging at springs. The cycle repeats, sustaining ecosystems and human life." Count words. Let's count: Water(1) evaporates2 from3 oceans,4 lakes,5 and6 soil,7 forming8 clouds.9 Precipitation10 returns11 as12 rain13 or14 snow,15 filling16 rivers17 and18 groundwater.19 Plants20 absorb21 water,22 releasing23 vapor24 through25 transpiration.26 Groundwater27 moves28 through29 aquifers,30 emerging31 at32 springs.33 The34 cycle35 repeats,36 sustaining37 ecosystems38 and39 human40 life41. Need 9 more words. Add: "and45 climate45? Wait count. Let's add "and45 climate46 balance." Need 9 words: "and climate balance, sustaining biodiversity, agriculture, and civilization." Count: and(1) climate(2) balance,(3) sustaining(4) biodiversity,(5) agriculture,(6) and(7) civilization(8). Need one more word: "forever"(9). Now total 41+9=50. Let's recount full sentence.<|end|><|start|>assistant<|channel|>final<|message|>Water evaporates from oceans, lakes, and soil, forming clouds. Precipitation returns as rain or snow, filling rivers and groundwater. Plants absorb water, releasing vapor through transpiration. Groundwater moves through aquifers, emerging at springs. The cycle repeats, sustaining ecosystems and human life, and climate balance, sustaining biodiversity, agriculture, and civilization forever.<|return|>
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}
}