You can deploy this model and start earning money today!

AmanPriyanshu/
gpt-oss-6.6b-specialized-instruction_following-pruned-moe-only-8-experts

Name: gpt-oss-6.6b-specialized-instruction_following-pruned-moe-only-8-experts
Availability: InStock

Views

No views yet

Instruction Following GPT-OSS Model (8 Experts)

Project: https://amanpriyanshu.github.io/GPT-OSS-MoE-ExpertFingerprinting/

👥 Follow the Authors

Aman Priyanshu

Supriti Vijay

Introduction

This is a pruned variant of OpenAI's GPT-OSS-20B model, reduced to 8 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.

Model Architecture & Statistics

Metric	Value
Base Model	openai/gpt-oss-20b
Architecture	Mixture-of-Experts Transformer
Total Parameters	~6.6B (pruned from 21B)
Original Experts per Layer	32
Pruned Experts per Layer	8
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

Pruning Methodology

What is Expert Pruning?

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:

Analyzing Usage Patterns: Tracking which experts activate most frequently for specific tasks
Removing Underutilized Experts: Discarding experts with low activation rates for the target domain
Preserving Router Functionality: Maintaining the routing mechanism with fewer available experts

Our Approach

Data-Driven Selection: Used activation patterns from instruction following evaluation tasks
Systematic Reduction: Reduced from 32 to 8 experts per layer
No Retraining: Direct removal without additional training steps

Performance & Applications

Pruning Benefits

Smaller Memory Footprint: 25.0% of original expert parameters
Reduced Computational Load: Fewer routing decisions during inference
Focused Capabilities: Retains experts relevant to instruction following tasks

Use Cases

Speculative Decoding: Draft model for full GPT-OSS-20B
Resource-Constrained Deployment: Edge devices, mobile applications
Research: Study expert specialization in MoE models
Fine-tuning: Smaller base model for domain adaptation

Note: Performance may vary depending on how well the pruned experts match your specific use case.

Motivation & Expert Selection

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:

GPQA: Graduate-level questions in physics, chemistry, biology (Diamond & Expert subsets)
MMLU/MMLU-Pro: Comprehensive knowledge across 57+ subjects including science, medicine, law
SORRY-Bench: Safety evaluation across harmful content categories
Tulu3: Persona-driven instruction following with verifiable constraints
Polyglot-or-Not: Multilingual factual completion tasks

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 8 experts per layer.

Dataset & Analysis Foundation

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.

Pruning Methodology

Our approach involves:

Activation Analysis: Comprehensive evaluation of expert usage patterns across domain-specific tasks
Expert Ranking: Identification of the most frequently activated experts for target domains
Systematic Pruning: Reduction from 32 to 8 experts while preserving router functionality
Quality Validation: Testing to ensure maintained performance on target tasks

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.

Usage

CPU Inference

python

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

# Load the specialized model on CPU
model = AutoModelForCausalLM.from_pretrained(
    "AmanPriyanshu/gpt-oss-6.6b-specialized-instruction_following-pruned-moe-only-8-experts", 
    torch_dtype=torch.bfloat16, 
    device_map="cpu", 
    trust_remote_code=True
)
tokenizer = AutoTokenizer.from_pretrained("AmanPriyanshu/gpt-oss-6.6b-specialized-instruction_following-pruned-moe-only-8-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)

Apple Silicon (MPS) Inference

python

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-6.6b-specialized-instruction_following-pruned-moe-only-8-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-6.6b-specialized-instruction_following-pruned-moe-only-8-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)

GPU Inference

For CUDA GPU inference, the models work with minimal modifications from the CPU example above. Simply change:

python

device_map="auto"  # Will automatically use GPU if available
torch_dtype=torch.bfloat16  # or torch.float16

Model Examples

The following examples demonstrate the model's performance on instruction following tasks:

Example 1

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 to produce formal email.<|end|><|start|>assistant<|channel|>final<|message|>**Subject:** Request for Meeting – Availability for Discussion on [Topic]  

Dear Professor [Last Name],  

I hope you are doing well. I am writing to request a brief meeting to discuss [briefly describe the topic or project you need to discuss]. Your insights would be invaluable for [specific purpose, e.g., refining the research proposal, reviewing the manuscript, or clarating the course material].  

I would be available on the following days:  
- **Monday, 2024‑08‑13,  30 – 45 ** (after the lecture)  
- **Tuesday, 2024‑08‑13,  30 – 45 ** (after the lecture)  
- **Wednesday, 2024‑08‑13,  30 – 45 ** (after the lecture)  

If you have a different preference, please let me know and I will adjust accordingly.  

Thank you for your time and consideration. I look forward to our conversation.  

Kind regards,  

[Your Full Name]  
[Your Program / Department]  
[University / Institution]  
[Phone] | [Email]  

---  

**Professional Closing**:  
Sincerely,  
[Your Name]  <|return|>

Example 2

Question: Create a numbered list of steps for making scrambled eggs, using exactly 6 steps.

Response:

<|channel|>analysis<|message|>We need to give a numbered list of steps for making scrambled eggs, exactly 6 steps. So 1- 2- etc.<|end|><|start|>assistant<|channel|>final<|message|>**How to make scrambled eggs (six steps)**  

1. **Gather the ingredients** – Grab a handful of eggs (at least 1–3 or 4–6) and a clean, slightly‑cleaned kitchen counter.  
2. **Open the eggs** – Crack the eggs into a small, sealed container or a plain, small, clean, slightly‑cleaned counter.  
3. **Heat the eggs** – Warm the eggs to a slightly‑cleaned, slightly‑cleaned, slightly‑cleaned, slightly‑cleaned, slightly‑cleaned, slightly‑cleaned, slightly‑cleaned, slightly‑cleaned, slightly‑cleaned, slightly‑cleaned, slightly‑cleaned, slightly‑cleaned, slightly‑cleaned, slightly‑cleaned, slightly‑cleaned, slightly‑cleaned, slightly‑cleaned, slightly‑cleaned, slightly‑cleaned, slightly‑cleaned, slightly‑cleaned, slightly‑cleaned, slightly‑cleaned, slightly‑cleaned, slightly‑cleaned, slightly‑cleaned, slightly‑cleaned, slightly‑cleaned, slightly‑cleaned, slightly‑cleaned, slightly‑cleaned, slightly‑cleaned, slightly‑cleaned, slightly‑cleaned, slightly‑cleaned, slightly‑cleaned, slightly‑cleaned, slightly‑cleaned, slightly‑cleaned, slightly‑cleaned, slightly‑cleaned, slightly‑cleaned, slightly‑cleaned, slightly‑cleaned, slightly‑cleaned, slightly‑cleaned, slightly‑cleaned, slightly‑cleaned, slightly‑cleaned, slightly‑cleaned, slightly‑cleaned, slightly‑cleaned, slightly‑cleaned, slightly‑cleaned, slightly‑cleaned, slightly‑cleaned, slightly‑cleaned, slightly‑cleaned, slightly‑cleaned, slightly‑cleaned, slightly‑cleaned, slightly‑cleaned, slightly‑cleaned, slightly‑cleaned, slightly‑cleaned, slightly‑cleaned, slightly‑cleaned, slightly‑cleaned, slightly‑cleaned, slightly‑cleaned, slightly‑cleaned, slightly‑cleaned, slightly‑cleaned, slightly‑cleaned, slightly‑cleaned, slightly‑cleaned, slightly

Example 3

Question: Summarize the water cycle in exactly 50 words, no more, no less.

Response:

<|channel|>analysis<|message|>Need 50 words exactly. Count words. Let's craft: "Water begins as a liquid, flows through rivers, evaporates into clouds, condates into ice, and returns to the surface. It circulates in ecosystems, sustaining life, then recagates to the atmosphere, forming a continuous cycle of birth, use, and renewal." Count words: 1-1. Let's count: "Water(1) begins(2) as(3) a(4) liquid,(5) flows(6) in(7) rivers,(8) evaporates(9) into(10) clouds,(11) condates(12) into(13) ice,(14) and(15) recagates(16) to(17) the(18) surface,(19) it(20) circulates(21) in(22) ecosystems,(23) sustaining(24) life,(25) then(26) recagates(27) to(28) the(29) atmosphere,(30) forming(31) a(34) continuous(5) cycle(5) of(5) birth,(5) use,(5) and(5) renewal." Too many. Need exactly 50 words. Let's craft: "Water starts as a liquid, flows through rivers, evaporates into clouds, condates into ice, and returns to the surface. It circulates in ecosystems, sustaining life, then returns to the atmosphere, forming a continuous cycle of birth, use, and renewal." Count words: Water(1) starts(2) as(3) a(4) liquid,(5) flows(6) through(7) rivers,(8) evaporates(9) into(10) clouds,(11) condates(12) into(13) ice,(14) and(15) returns(16) to(17) the(18) surface.(19) It(20) circulates(21) in(22) ecosystems,(23) sustaining(24) life,(25) then(26) returns(27) to(28) the(29) atmosphere,(30) forming(31) a(34) continuous(5) cycle(5) of(5) birth,(5) use,(5) and(5) renewal. That's 34 words. Need 50. Let's add: "Water starts as a liquid, flows through rivers, evaporates into clouds, condates into ice, and returns to the surface. It circulates

Citation

If you use this model in your research, please cite:

bibtex

@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}
}

References & Resources

Original Model: OpenAI GPT-OSS Model Card
Model Hub: GPT-OSS-20B on Hugging Face
Expert Analysis Dataset: GPT-OSS-20B MoE Expert Activations
Project Page: GPT-OSS MoE Expert Fingerprinting
GitHub Repository: OpenAI GPT-OSS