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

Law 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 law 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	Law

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:

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

Our Approach

• Data-Driven Selection: Used activation patterns from law 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 law 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 legal domain model employs experts that demonstrated expertise during law-related tasks from MMLU legal subjects. These experts excel at legal reasoning, jurisprudence, and understanding of legal frameworks and procedures.

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 law 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:

1. Activation Analysis: Comprehensive evaluation of expert usage patterns across domain-specific tasks
2. Expert Ranking: Identification of the most frequently activated experts for target domains
3. Systematic Pruning: Reduction from 32 to 8 experts while preserving router functionality
4. 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

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

# Load the specialized model on CPU
model = AutoModelForCausalLM.from_pretrained(
    "AmanPriyanshu/gpt-oss-6.6b-specialized-law-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-law-pruned-moe-only-8-experts")

# Generate with the model
messages = [
    {"role": "user", "content": "What is the difference between civil and criminal law?"}
]

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

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-law-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-law-pruned-moe-only-8-experts")

# Generate with the model
messages = [
    {"role": "user", "content": "What is the difference between civil and criminal law?"}
]

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:

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 law tasks:

Example 1

Question: What is the difference between civil and criminal law?

Response:

<|channel|>analysis<|message|>Need to explain difference: civil law (disrights, contracts, torts) vs criminal law (punative). Provide examples.<|end|><|start|>assistant<|channel|>final<|message|>**Civil law** and **criminal law** are two branches of the legal system that deal with different kinds of disputes and the purposes of the law.  Below is a quick‑table of the main differences:

| | | | | |

| **Purpose** | Civil law is about resolving disputes between private parties (e.g. contracts, torts, property, family matters). | Criminal law is about protecting society and punating conduct that is considered harmful or dangerous to the public. | |

| **Who** | The parties in a civil case are usually two or more private individuals or entities. | The state (or the government) is the plaintiff in a criminal case. | |

| **Outcome** | The court orders one party to do something (e. pay damages, perform a contract, or stop a certain action). | The court may order a person to pay a fine, serve a jail term, or be put in a probation program. | |

| **Bur burden** | The plaintiff in a civil case must prove that the defendant’s conduct caused harm. | The state in a criminal case must prove that the defendant’s conduct was a crime, and that the defendant had the intent or knowledge to do it. | |

| **Standard of proof** | “Pre‑beyond a reasonable doubt” is not used. The plaintiff must prove “pre‑balance of evidence” or “pre‑probical evidence” (i.e. “more than 50% of the evidence” or “more than 50% of the evidence”). | The state must prove “pre‑beyond a reasonable doubt” (i. “more than 50% of the evidence” or “more than 50% of the evidence”). | |

| **Pen** | The court can order a civil damages award, a injunction, or a specific order. | The court can order a jail term, a fine, or a probation program. | |

| **Re‑s** | The civil case is usually a “private” dispute. | The criminal case is a “public” dispute. | |

| **Rem** | The civil case is usually a “private” remedy. | The criminal case is usually a “public” remedy. | |

| **Rem** | The civil case is usually a “private” remedy. | The criminal case is usually

---

Example 2

Question: Explain the concept of due process in legal systems.

Response:

<|channel|>analysis<|message|>Need to explain due process (due dueness? Actually due process = due dueness? It's due dueness? Probably due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's due dueness? It's

---

Example 3

Question: What are the basic rights protected by the First Amendment?

Response:

<|channel|>analysis<|message|>User asks: "What are the basic rights protected by the First Amendment?" They want explanation of rights. Provide overview: freedom of religion, speech, press, assembly, etc. Provide context.<|end|><|start|>assistant<|channel|>final<|message|>The First Amendment of the U‑C‑Con Constitution (the U‑C‑Con Constitution is the U‑C‑Con Constitution) is the most famous and the most famous. It is the most famous. The First Amendment is the most famous. The First Amendment is the most famous. The First Amendment is the most famous. The First Amendment is the most famous. The First Amendment is the most famous. The First Amendment is the most famous. The First Amendment is the most famous. The First Amendment is the most famous. The First Amendment is the most famous. The First Amendment is the most famous. The First Amendment is the most famous. The First Amendment is the most famous. The First Amendment is the most famous. The First Amendment is the most famous. The First Amendment is the most famous. The First Amendment is the most famous. The First Amendment is the most famous. The First Amendment is the most famous. The First Amendment is the most famous. The First Amendment is the most famous. The First Amendment is the most famous. The First Amendment is the most famous. The First Amendment is the most famous. The First Amendment is the most famous. The First Amendment is the most famous. The First Amendment is the most famous. The First Amendment is the most famous. The First Amendment is the most famous. The First Amendment is the most famous. The First Amendment is the most famous. The First Amendment is the most famous. The First Amendment is the most famous. The First Amendment is the most famous. The First Amendment is the most famous. The First Amendment is the most famous. The First Amendment is the most famous. The First Amendment is the most famous. The First Amendment is the most famous. The First Amendment is the most famous. The First Amendment is the most famous. The First Amendment is the most famous. The First Amendment is the most famous. The First Amendment is the most famous. The First Amendment is the most famous. The First Amendment is the most famous. The First Amendment is the most famous. The First Amendment is the most famous. The First Amendment is the most famous. The First Amendment is the most famous. The First Amendment is the most famous. The First Amendment is the most famous. The First Amendment is the most famous. The First Amendment is the

---

Citation

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}
}

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