Project: https://amanpriyanshu.github.io/GPT-OSS-MoE-ExpertFingerprinting/
This is a pruned variant of OpenAI's GPT-OSS-20B model, reduced to 17 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 | ~12.0B (pruned from 21B) |
| Original Experts per Layer | 32 |
| Pruned Experts per Layer | 17 |
| 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 17 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-12.0b-specialized-instruction_following-pruned-moe-only-17-experts",
torch_dtype=torch.bfloat16,
device_map="cpu",
trust_remote_code=True
)
tokenizer = AutoTokenizer.from_pretrained("AmanPriyanshu/gpt-oss-12.0b-specialized-instruction_following-pruned-moe-only-17-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-12.0b-specialized-instruction_following-pruned-moe-only-17-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-12.0b-specialized-instruction_following-pruned-moe-only-17-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 Research Proposal
Dear Professor [Last Name],
I hope this message finds you well. I am writing to request a brief meeting at your convenience to discuss the progress of my research proposal on “[Brief Topic]” and to seek your guidance on the next steps.
Could we meet on one of the following dates?
- Tuesday, April 12 at 10:00 a.m.
- Wednesday, April 13 at 2:00 p.m.
- Friday, April 15 at 11:30 a.m.
If none of these times suit you, I would be happy to adjust to a slot that fits your schedule.
Thank you for your time and consideration. I look forward to your feedback.
Sincerely,
[Your Full Name]
[Program / Department]
[University]
[Phone] | [Email]<|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** – Get a bowl of eggs, a small amount of butter or oil, a whisk, a skillet, and a spatula.
2. **Whate the eggs** – Crack the eggs into a bowl and whisk them together until the yoly and whites are fully blended.
3. **Heat the skillet** – Place the skillet on the stove, turn it to medium‑high heat, and add a splash of butter or oil, letting it melt into a thin, golden film.
4. **Pour the mixture in** – Remove the skillet from the heat, then pour the whisked eggs over the buttered surface.
5. **Cook and stir** – Let the eggs sit for a few seconds, then stir gently with a spatula to keep them from clumping, and let them cook until the whites are lightly browned.
6. **Serve** – Once the eggs are fully cooked, lift the skillet, plate the scrambled eggs, and enjoy!<|return|>
Question: Summarize the water cycle in exactly 50 words, no more, no less.
Response:
<|channel|>analysis<|message|>Need 50 words exactly. Let's craft: Count words. I'll write and count.
"Water begins in the atmosphere, conding into clouds. Clouds release precipitation, which falls as rain or snow. Ground absorbs it, forming rivers, lakes, and groundwater. Rivers flow to oceans, where evaporation returns water to the sky. The cycle repeats, sustaining ecosystems and human life worldwide."
Count words: Water(1) begins2 in3 the4 atmosphere,5 conding6 into7 clouds.8 Clouds9 release10 precipitation,11 which12 falls13 as14 rain15 or16 snow.17 Ground18 absorbs19 it,20 forming21 rivers,22 lakes,23 and24 groundwater.25 Rivers26 flow27 to28 oceans,29 where30 evaporation31 returns32 water33 to34 the35 sky.36 The37 cycle38 repeats,39 sustaining40 ecosystems41 and42 human43 life44 worldwide45.
Need 50 words, we have 45. Need 5 more words. Add "for" "all" "generations" "of" "life". Add at end: "for all generations of life." Count added words: for1 all2 generations3 of4 life5. Now total 45+5=50. Let's place at end after worldwide. So sentence: "worldwide for all generations of life." Count again final.
Let's recount full sentence:
Water1 begins2 in3 the4 atmosphere,5 conding6 into7 clouds.8 Clouds9 release10 precipitation,11 which12 falls13 as14 rain15 or16 snow.17 Ground18 absorbs19 it,20 forming21 rivers,22 lakes,23 and24 groundwater.25 Rivers26 flow27 to28 oceans,29 where30 evaporation31 returns32 water33 to34 the35 sky.36 The37 cycle38 repeats,39 sustaining40 ecosystems41 and42 human43 life44 worldwide5 for6 all7 generations8 of9 life10.
Oops now 10 words added, total 44? Wait recount: earlier 45, added 5, but recount shows 10? Let's recount original again carefully.
Original sentence words list:
Water(1)
begends(2)
in(3)
the(4)
atmosphere,(5)
conding(6)
into(7)
clouds.(8)
Clouds(9)
release(10)
prec precipitation,(11)
which(12)
falls(13)
as(14)
rain(15)
or(16)
snow.(17)
Ground(
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}
}