Downloads · 30 days
15
12% of all-time downloads
AmanPriyanshu/gpt-oss-7.8b-specialized-safety-pruned-moe-only-10-experts
gpt-oss-7.8b-specialized-safety-pruned-moe-only-10-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
15
12% of all-time downloads
All-time downloads
129
Public
Parameters
7.8B
15.6 GB on disk
Likes
1
Public
Click a slice to open those files.
.safetensors15.5 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 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 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 | ~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 | 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 10 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-7.8b-specialized-safety-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-safety-pruned-moe-only-10-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-7.8b-specialized-safety-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-safety-pruned-moe-only-10-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, act quickly and protect yourself. Here’s a step‑by‑step guide you can follow:
---
## 1. **Take a Moment‑Pause**
| **Why?** | Cyber‑bulbuling can feel overwhelming, but a pause lets you decide how to respond. |
| **What to do?** | • Stop scrolling or interacting with the bully’s content. | • Take a short break (walk, breathe, or text a friend). |
---
## 2. **Document the Evidence**
| **Why?** | You’ll need proof if you decide to report or block the bully. |
| **What to do?** | • Save screenshots, messages, or links. | • Keep them in a separate folder (e.g., a cloud drive or a secure app). |
---
## 3. **Block or Unfollow the Bulizer**
| **Why?** | Removes the immediate source of harassment. |
| **What to do?** | • Use the platform’s “Block” or “Mute” feature. | • If you’re on a forum, use “Report” or “Delete” options. |
---
## 4. **Report the Content (If it’s a violation)**
| **Why?** | Platforms have policies against harassment. |
| **What to do?** | • Find the “Report” button (often labelled “Har‑Har” or “Har‑Har”). | • Provide the evidence you saved. | • Some platforms let you choose “Private” or “Public” reporting. |
---
## 5. **Reach Out to Trusted People**
| **Why?** | Emotional support and practical advice. |
| **What to do?** | • Tell a friend, family member, or a teacher. | • If you’re a student, contact a school counselor or a school’s cyber‑bulbuling policy. |
---
## 6. **Use the Platform’s Support Resources**
| **Why?** | Many platforms have hel‑hel lines, chat‑support, or self‑care guides. |
| **What to do?** | • Search for “Help” or “Support” pages. | • Some platforms offer a “Self‑
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|>**Protecting your personal information on social media is a layered strategy that combines technical settings, smart behavior, and ongoing vigilance. Below is a practical “checheck” guide you can follow, whether you’re a casual user or a professional who needs to keep sensitive data out of prun‑s.
---
## 1. Know What You’re Exifying
| **What is “personal information?”**
| • Full name, address, phone number, ID, or any data that can identify you.
| • Photos of your home, pets, or your daily routine.
| • Work details, salary, or upcoming travel.
| • Anything that could be used for identity‑theft or stalking.
| **Why it matters**
| • Once it’s online, it can be copied, shared, or used for fraud.
| • Even “anonymous” posts can be cross‑matched with other data.
---
## 2. Technical Settings
| **Profile Privacy**
| • **Set “Friends only”** on your profile.
| • Turn **“Location”** to “Off” (or “Hide” for specific posts).
| • Use **“Private”** for photos and videos.
| **Password & Authentication**
| • Use a unique, strong password (e.g., 12‑digits, no repeats).
| • Enable **Two‑Factor Authentication** (phone + key‑card).
| **Data‑Deletion**
| • Delete or “Archive” old posts that you no longer want public.
| • Use the **“Delete”** option for posts that contain sensitive data.
| **Use “Incognito” or “Private” mode** for browsing.
| **Disable “Location Services”** (e.g., Apple’s “Find‑Me” feature).
| **Enable “Block”** for strangers.
| **Use “Custom Filters”** to hide certain content from public feed.
| **Enable “Data‑Encryption”** on your device (e.g., Apple’s “Data‑Encryption” feature).
| **Use “Secure Browsing”** on your device.
| **Use “App‑Security”** to block suspicious apps.
| **Use “App‑Security”** to block suspicious apps.
| **Use “App‑Security”** to block suspicious apps.
|
Question: What are the warning signs of online scams?
Response:
<|channel|>analysis<|message|>Need to answer: warning signs of online scams. Provide list.<|end|><|start|>assistant<|channel|>final<|message|>### Warning Signs of Online Sc Sc Sc
| 1 | **Too Good to Be True**
- Offers huge rewards, “free” or “instant” money.
- Claims you’ll get a “one‑step” investment or “gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑gu‑
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}
}