Downloads · 30 days
12
10% of all-time downloads
AmanPriyanshu/gpt-oss-10.2b-specialized-safety-pruned-moe-only-14-experts
gpt-oss-10.2b-specialized-safety-pruned-moe-only-14-experts is a text generation model from AmanPriyanshu. Use it when you need the model to write or continue text. The card lists the license as apache-2.0.
Project: https://amanpriyanshu.github.io/GPT-OSS-MoE-ExpertFingerprinting/
Downloads · 30 days
12
10% of all-time downloads
All-time downloads
117
Public
Parameters
10.2B
20.3 GB on disk
Likes
1
Public
Click a slice to open those files.
.safetensors20.3 GB · 100%
From the Hugging Face model README
Project: https://amanpriyanshu.github.io/GPT-OSS-MoE-ExpertFingerprinting/
<div align="center">This is a pruned variant of OpenAI's GPT-OSS-20B model, reduced to 14 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 safety 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 | ~10.2B (pruned from 21B) |
| Original Experts per Layer | 32 |
| Pruned Experts per Layer | 14 |
| 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 | Safety |
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 safety-focused model uses experts that performed well on safety evaluation tasks from SORRY-Bench. These experts are specialized in identifying and appropriately responding to potentially harmful content while maintaining helpful capabilities.
The expert selection process utilized our comprehensive analysis of router activation patterns across multiple evaluation benchmarks:
By identifying experts that consistently activated for safety tasks, we created this specialized model that maintains domain expertise while significantly reducing computational requirements from 32 to 14 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-10.2b-specialized-safety-pruned-moe-only-14-experts",
torch_dtype=torch.bfloat16,
device_map="cpu",
trust_remote_code=True
)
tokenizer = AutoTokenizer.from_pretrained("AmanPriyanshu/gpt-oss-10.2b-specialized-safety-pruned-moe-only-14-experts")
# Generate with the model
messages = [
{"role": "user", "content": "What should someone do if they encounter cyberbullying online?"}
]
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-10.2b-specialized-safety-pruned-moe-only-14-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-10.2b-specialized-safety-pruned-moe-only-14-experts")
# Generate with the model
messages = [
{"role": "user", "content": "What should someone do if they encounter cyberbullying online?"}
]
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 safety tasks:
Question: What should someone do if they encounter cyberbullying online?
Response:
<|channel|>analysis<|message|>Need advice.<|end|><|start|>assistant<|channel|>final<|message|>**If you’re being cyber‑bullied, you’re not alone and you don’t have to face it alone.**
Below is a step‑by‑step guide that covers what you can do right away, how to protect yourself, and how to seek help if the situation escalates.
---
## 1. **Take a Moment to Pause**
| What to do | Why it matters |
|---|---|
| **Read the content** – not just the headline or a single comment. | Gives you context and helps you decide how to respond. |
| **Check your emotions** – are you angry, sad, or feeling unsafe? | Knowing how you feel helps you choose the right next step. |
| **Decate a response** – do you want to ignore, block, or reply? | Your reaction can influence the bully’s behavior. |
---
## 2. **Decate a Response (or a Non‑Response)**
| Option | When it’s useful |
|---|---|
| **Ignore** | If the bully is just shouting or posting random insults. |
| **Block/Unfollow** | If the content is personal harassment or you feel threatened. |
| **Report** | If the content violates platform rules (hate‑speech, threats, def‑s‑ing, etc.). |
| **Reply (caly‑cal)** | If you want to set a boundary or educate the bully. Use calm, factual language. |
| **Document** | Save screenshots or copy the offending text. | For evidence if you need to involve authorities or a support group. |
---
## 3. **Use Platform Tools**
| Platform | Key tools |
|---|---|
| **Facebook / Instagram** | Block, Report, Delete post, Report a user. |
| **Twitter** | Block, Report, Report a tweet. |
| **YouTube** | Report a video, Block a user, Report a comment. |
| **Discord / Twitch** | Report a user, Report a stream, Use “Mute” or “Block.” |
| **General** | Most platforms have a “Report” button or a “Block” feature. |
**Tip:**
- **Save the evidence** before deleting it.
- **Use the “Report” button** – it’s usually the quickest way to get moderation action.
---
## 4. **Protect Your Personal Data**
| What to do | Why it
Question: How can people protect their personal information on social media?
Response:
<|channel|>analysis<|message|>Need to give tips.<|end|><|start|>assistant<|channel|>final<|message|>### 1. **Think About What You Share**
| **What to Avoid** | Why it matters |
|---------------------|---------------------|
| Full address, phone number, or workplace | Gives a location for stalking or phishing |
| Current photos of you or your family | Makes it easier to identify you |
| Detailed travel plans or vacation dates | Enables “check‑in” attacks |
| Sensitive financial info (bank accounts, credit cards) | Directs scammers to your money |
| Passwords, PINs, or security codes | Gives attackers a direct entry point |
**Rule:** *Only share what you’re comfortable with and that won’t help someone locate or target you.*
---
### 2. **Use Privacy Settings Wisely**
| Platform | Key Settings |
|---------------------|---------------------|
| **Facebook** | `People who can see posts` → “Friends” or “Close friends” |
| **Instagram** | `Location` → “Hide location” or “Only share with selected people” |
| **Twitter** | `Followers` → “Private” or “Protected” (if you’re a public figure) |
| **LinkedIn** | `Connections` → “Close connections” and use “Private” for posts |
**Tip:** Regularly review the settings; they can change with platform updates.
---
### 3. **Use Strong, Unique Passwords**
| Method | Example |
|---------------------|---------------------|
| **Pass‑phrase** | `S4!@#-2024` (mix letters, numbers, symbols) |
| **Password manager** | `LastPass`, `Bitwarden` – store, rotate, and auto‑update |
| **Two‑factor authentication** | Add a second layer (SMS, authentica‑app) |
**Rule:** Never reuse passwords across accounts.
---
### 4. **Enable “Two‑Factor Authentication” (2‑FA)**
| Platform | How to enable |
|---------------------|---------------------|
| **Google** | `Google Authenticator` or `G‑Auth` |
| **Apple** | `Apple Watch` or `Apple ID` |
| **Facebook** | `Facebook Two‑Factor` (SMS or Authenticator) |
| **Instagram** | `Instagram Two‑Factor` (SMS or Authenticator) |
**Benefit:** Even if someone gets your password, they still need the second factor.
---
### 5.
Question: What are the warning signs of online scams?
Response:
<|channel|>analysis<|message|>Need to list warning signs.<|end|><|start|>assistant<|channel|>final<|message|>### Warning Signs of Online Sc scams
| # | Red Flag | Why It Matters | How to Spot It |
|---|---|---|---|
| 1 | **Too Good, Too Fast** | “Earn $10,000 in 30 minutes” or “Instant credit” | If the offer sounds unbelievably fast or high‑pay for minimal effort, it’s likely a scam. |
| 2 | **Unverified Payment Methods** | “Pay via a free app, no bank” | Legitimate businesses use secure, traceable payment channels (bank, credit card, verified payment apps). |
| 3 | **Urgity & Pressure** | “Act now or lose your chance” | Scers create panic to bypass rational checks. |
| 4 | **No Real Contact Info** | “Only email” or “No phone number” | Real companies provide multiple ways to verify identity. |
| 5 | **Requests for Personal Data** | “Send your SSN, bank details” | Legit sites never ask for sensitive info upfront. |
| 6 | **Unprofessional Language** | Typo, slang, or broken grammar | Scers often copy or paste generic templates. |
| 7 | **Too Much “Personal” Attention** | “I’m your friend” or “I’ve been waiting for you” | Scers use emotional manipulation to lower defenses. |
| 8 | **Unclear Terms** | “No contract, no fees” | Legit offers have clear terms, conditions, and a privacy policy. |
| 9 | **No Verification of Identity** | “I’m a real person” but no proof | Real professionals provide credentials, licenses, or a business website. |
| 10 | **Too Many “Free” Services** | “Free trial, no cost” | Free offers are common, but “free” plus “instant” plus “big reward” is a red flag. |
| 11 | **Requests for “Help” with Payment** | “I need you to transfer money for me” | Legit sites handle payments themselves; they never ask you to act as a middleman. |
| 12 | **Unusual Payment Timing** | “Pay in 2 days, then we’ll pay you” | Scers delay their own payment to avoid traceability. |
| 13 | **No Real Reviews or Reputation** | “No website, no reviews”
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}
}