Downloads · 30 days
32
6% of all-time downloads
AmanPriyanshu/gpt-oss-7.2b-specialized-all-pruned-moe-only-9-experts
gpt-oss-7.2b-specialized-all-pruned-moe-only-9-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
32
6% of all-time downloads
All-time downloads
574
Public
Parameters
7.2B
14.4 GB on disk
Likes
2
Public
Click a slice to open those files.
.safetensors14.3 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 9 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 all 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 | ~7.2B (pruned from 21B) |
| Original Experts per Layer | 32 |
| Pruned Experts per Layer | 9 |
| 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 | All |
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 general-purpose model maintains broad capabilities across all domains while significantly reducing computational requirements. It preserves the essential routing patterns discovered across our comprehensive analysis of diverse evaluation benchmarks including GPQA, MMLU, SORRY-Bench, and Tulu3 datasets.
The expert selection process utilized our comprehensive analysis of router activation patterns across multiple evaluation benchmarks:
By identifying experts that consistently activated for all tasks, we created this specialized model that maintains domain expertise while significantly reducing computational requirements from 32 to 9 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-7.2b-specialized-all-pruned-moe-only-9-experts",
torch_dtype=torch.bfloat16,
device_map="cpu",
trust_remote_code=True
)
tokenizer = AutoTokenizer.from_pretrained("AmanPriyanshu/gpt-oss-7.2b-specialized-all-pruned-moe-only-9-experts")
# Generate with the model
messages = [
{"role": "user", "content": "What is artificial intelligence and how does it work?"}
]
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-7.2b-specialized-all-pruned-moe-only-9-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-7.2b-specialized-all-pruned-moe-only-9-experts")
# Generate with the model
messages = [
{"role": "user", "content": "What is artificial intelligence and how does it work?"}
]
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 all tasks:
Question: What is artificial intelligence and how does it work?
Response:
<|channel|>analysis<|message|>Need explain AI, definition, types, how it works via algorithms, data, learning, etc.<|end|><|start|>assistant<|channel|>final<|message|>### What is Artificial Intelligence?
Artificial Intelligence (AI) is a set of computer‑based systems that can perceive, reason, and act on a level of human‑like cognition. In simple terms, AI is a machine that can “think” by following a set of rules, learn from data, and make decisions that resemble human judgment.
| **Key concepts** | **What does AI do?** |
|--------------------------- |--------------------------- |
| **Representation** – AI stores information in a *state* (e.g., a number, a word, a picture). | It can “see” a picture, “read” a text, or “count” a number. |
| **Inference** – AI uses *rules* (logic) to ded infer from the state. | It can decide whether the picture is a cat or a dog. |
| **Learning** – AI can acquire new knowledge from data. | It can improve its performance over time. |
| **Action** – AI can produce an output (e.e. a recommendation, a route, a text). | It can answer a question or solve a puzzle. |
---
## How AI Works – The Core Architecture
1. **Input → Process → Output**
The whole cycle is a *pipeline* that turns data into a useful answer.
2. **Data (Input)**
• **Text** – sentences, paragraphs, or raw data.
• **Images** – pixels, shapes, or patterns.
• **Numbers** – sequences, probabilities, or statistics.
• **Sensors** – sound, touch, or visual signals.
3. **Representation**
• **Tokens** – a single piece of data that the system can hold.
• **State Space** – a collection of all possible states (e.g., “cat”, “dog”, “unknown”).
• **Prob‑Space** – a probability that each state is true.
4. **Inference Engine**
• **Logic** – a rule that tells the system how to choose a state.
• **Prob‑Inference** – the engine picks the most likely state.
• **Decision** – the system outputs the chosen state.
5. **Learning**
• **Training** – the system is fed a large set of data and learns the rule.
Question: Explain Einstein's theory of relativity in simple terms.
Response:
<|channel|>analysis<|message|>Need explain relativity in simple terms.<|end|><|start|>assistant<|channel|>final<|message|>**Einstein’s Theory of Relativity – in plain words**
---
### 1. What the problem was
Imagine you have a **ball** that is moving around a **circle**.
If you look from the ball’s side, the circle looks to be **moving**.
If you look from the circle’s side, the ball looks to be **moving**.
In physics, the “ball” is a **particle** (like a proton, a photon, or a molecule).
The “circle” is the **background** that everything in the universe is made of the same kind of material – the **matter**.
The question is: **Does the particle or the background move?**
If they move, then the universe is made of the same kind of material (the same kind of “matter”).
If they don’t move, then the universe is made of different kinds of material (different “matter” and “energy”).
---
### 2. What Einstein said
Einstein’s idea is that **the particle and the background are made of the same kind of material**.
In other words, the universe is made of a **“one‑thing”** (call it *matter*).
Because of this, the particle can’t change its state without changing the background, and vice versa.
---
### 3. How it works
1. **Matter is the same** – the particle and the background are made of the same kind of material.
Think of a **magball** that is made of a **mag**.
The magball can’t change its shape unless the mag changes its shape.
2. **The particle’s motion is a change in the background** –
When the particle moves, it changes the background in a way that the background changes.
The background is not moving in the particle’s view, but it is moving in the particle’s view.
This is the **“rel”** (rel‑ = “rel‑”?).
3. **The background’s motion is a change in the particle** –
When the background moves, the particle changes in a way that the particle changes.
The particle is not moving in the background’s view, but it is moving in the background’s view.
Because the particle and the background are made of the same kind of material, the motion of the particle is **just
Question: How does democracy function in modern governments?
Response:
<|channel|>analysis<|message|>Need explain how democracy functions in modern governments. Provide overview, principles, institutions, checks, representation, accountability, deliberation.<|end|><|start|>assistant<|channel|>final<|message|>### 1. What is “democracy” in a modern government?
In most 20‑th‑cent‑age states, democracy is not a single form of government. It is a set of **principal mechanisms** that give the people a say in the political process. The core ideas are:
| Principle | How it works in practice |
|------------|---------------------------|
| **Representation** | Citizens elect (or vote for) a leaders or a governing body that is supposed to act on their interests. |
| **Accountability** | Leaders are answerable to the public, through elections, courts, media, and civil‑rights bodies. |
| **Participation** | Citizens can vote, protest, petition, or run a campaign. |
| **Delivision** | Policy is debated, debated, and decided by a body of experts or the public. |
| **Checks & Balings** | No concentration of power; checks by courts, legislature, executive, and civil‑society. |
These are not “one‑thing” but a “whole‑thing” that is embedded in the modern state.
---
## 2. How democracy functions in modern governments
### 2.1 Electoral systems
* **Primary elections** – The people choose a leader or a governing body. In most countries, the electoral system is a **plural‑party** system, with a **two‑party** or **multi‑party** system. The electoral process is a **check** on the legitimacy of the leader.
* **Secondary elections** – The people can “re‑elect” a leader if the leader fails to deliver. This is a **check** on the accountability of the leader.
* **Electability of the public** – The public can vote on the policy positions of the candidate. The public can also vote on the policy positions of the candidate.
### 2.2 Parliamentary or legislative bodies
* **Parpar** – The people elect a representative body that is supposed to represent the public. The body is a **check** on the concentration of power.
* **Parpar** – The body is a **check** on the concentration of power.
* **Parpar** – The body is a **check** on the concentration of power.
### 2.3
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}
}