Downloads · 30 days
48
21% of all-time downloads
AmanPriyanshu/gpt-oss-12.0b-specialized-science-pruned-moe-only-17-experts
gpt-oss-12.0b-specialized-science-pruned-moe-only-17-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
48
21% of all-time downloads
All-time downloads
228
Public
Parameters
12B
23.9 GB on disk
Likes
2
Public
Click a slice to open those files.
.safetensors23.9 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 17 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 science 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 | ~12.0B (pruned from 21B) |
| Original Experts per Layer | 32 |
| Pruned Experts per Layer | 17 |
| 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 | Science |
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 science-specialized model leverages experts that showed high activation patterns during scientific reasoning tasks from GPQA (physics, chemistry, biology) and MMLU science domains. These experts demonstrate superior performance on complex scientific problem-solving and technical knowledge recall.
The expert selection process utilized our comprehensive analysis of router activation patterns across multiple evaluation benchmarks:
By identifying experts that consistently activated for science tasks, we created this specialized model that maintains domain expertise while significantly reducing computational requirements from 32 to 17 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-12.0b-specialized-science-pruned-moe-only-17-experts",
torch_dtype=torch.bfloat16,
device_map="cpu",
trust_remote_code=True
)
tokenizer = AutoTokenizer.from_pretrained("AmanPriyanshu/gpt-oss-12.0b-specialized-science-pruned-moe-only-17-experts")
# Generate with the model
messages = [
{"role": "user", "content": "Explain the process of photosynthesis in plants."}
]
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-12.0b-specialized-science-pruned-moe-only-17-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-12.0b-specialized-science-pruned-moe-only-17-experts")
# Generate with the model
messages = [
{"role": "user", "content": "Explain the process of photosynthesis in plants."}
]
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 science tasks:
Question: Explain the process of photosynthesis in plants.
Response:
<|channel|>analysis<|message|>Need concise explanation.<|end|><|start|>assistant<|channel|>final<|message|>**Photosynthesis – the “making of food” in plants**
| Step | What happens | Key molecules & enzymes | Where it occurs |
|------|--------------|------------------------|----------------|
| 1. **Light absorption** | Chlorophyll‑a, chlorophyll‑b, and accessory pigments in the thylakoid membranes absorb photons (400–700 nm). | • Chlorophyll a & b | Thylakoid membranes of chloroplasts |
| 2. **Excitation of electrons** | Energy excites electrons in chlorophyll‑a → they are passed to the *electron transport chain* (ETC). | • Photosystem II (PSII) → plastoquinone → Photosystem I (PSI) | Thylakoid membrane |
| 3. **Water splitting (photolysis)** | PSII uses light energy to split H₂O → 2 H⁺ + 2 e⁻ + O₂. | • Oxygen‑evolving complex (OEC) | Thylakoid membrane |
| 4. **ATP synthesis** | Electrons travel through ETC, pumping protons into the thylakoid lumen → proton‑gradient drives ATP synthase → ATP. | • ATP synthase | Thylakoid membrane |
| 5. **NADPH formation** | Electrons reduce NADP⁺ to NADPH (via PSI). | • PSI, cytochrome b₆₁ | Thylakoid membrane |
| 6. **Carbon fixation (Calvin cycle)** | In the stroma, CO₂ is fixed into 3‑phosphoglycerate (3‑PG) by the enzyme Rubisco. 3‑PG is then recycled to produce sugars (glucose, fructose). | • Rubisco, GAPDH, PK, etc. | Stroma of chloroplasts |
| 7. **Sugar export** | Glucose and other sugars are transported out of the chloroplast to the cytosol and then to other tissues. | • Hexokinase, sucrose transporters | Cytosol, plasmodesmata |
**Key points**
1. **Energy source** – Light energy drives the entire process.
2. **Electron transport chain** – Generates ATP and NADPH, the “energy currency” for the Calvin cycle.
3. **Water → Oxygen** – Photosynthesis releases O₂
Question: What causes the greenhouse effect and how does it work?
Response:
<|channel|>analysis<|message|>Need explain greenhouse effect causes, greenhouse gases, absorption of IR, trapping heat, radiative balance.<|end|><|start|>assistant<|channel|>final<|message|>### 1. What causes the greenhouse effect?
The “greenhouse effect” (or “greenhouse‑gas effect”) is the natural warming of the Earth’s surface that results from the presence of certain gases in the atmosphere that can absorb and re‑emit infrared (IR) radiation. The effect is named after the way a greenhouse works: a glass roof lets sunlight (visible light) in, but traps the heat that the plants and the roof emit back as IR radiation.
The main greenhouse gases (GHGs) are:
| Gas | Natural concentration (ppm) | Typical source | Radiative effect |
|------|------------------------------|----------------|----------------|
| Water vapor | 2000–3000 | Evaporation, weather | Strong, but variable |
| Carbon dioxide (CO₂) | 400 | Fossil fuels, deforestation, industry | Strong, long‑term |
| Methane (CH₄) | 1.8 | Agriculture, natural gas, wetlands | Very potent, short‑term |
| Nitrous oxide (N₂O) | 0.3 | Agriculture, livestock, industrial | Potent, short‑term |
| Ozone (O₃) | 0.3 | Photochemical reactions | Strong, short‑term |
| Aerosols (e.g., sulfate) | – | Volcanic, industrial | Generally cooling (anti‑greenhouse) |
The atmosphere is transparent to most of the Sun’s visible and ultraviolet (UV) radiation, so that radiation can reach the surface. Once the surface is heated, it emits long‑wave IR radiation. Greenhouse gases absorb a fraction of that IR and then re‑emit it in all directions. A fraction of the re‑emitted IR escapes to space, but a larger fraction is directed back toward the surface, keeping the surface warmer than it would be if the atmosphere were transparent to IR.
### 2. How does it work? (Radiative transfer)
1. **Solar radiation reaches the Earth**
- The Sun emits a spectrum that peaks in the visible/near‑UV.
- The atmosphere is largely transparent to this part of the spectrum, so it reaches the surface.
2. **Surface absorbs and re‑emits**
- The surface absorbs the incoming solar energy and warms.
- A warm surface emits
Question: Describe the structure and function of DNA.
Response:
<|channel|>analysis<|message|>Need to explain DNA structure: double helix, nucleotides, base pairs, sugar-phosphate backbone, major bases, base pairing rules, major vs minor, major groove, minor groove, hydrogen bonding, base composition, sequence, etc. Function: genetic information storage, replication, transcription, regulation, coding for proteins, etc. Also mention roles in evolution, regulation, epigenetics. Provide details.<|end|><|start|>assistant<|channel|>final<|message|>**DNA – the Double‑Helix Blueprint of Life**
---
## 1. The Physical Structure of DNA
| Feature | What it Is | Why It Matters |
|--------|------------|----------------|
| **Double‑helix** | Two strands of nucleic acid that wind around each other like a ladder. | Gives stability, protects the genetic code, and allows proteins to read the sequence. |
| **Back‑to‑back sugar‑phosphate backbone** | 5‑carbon de‑oxyribose sugars linked by phosphates. | Provides a rigid, uniform scaffold that keeps the bases in the right orientation. |
| **Nucleobases** | Four nitrogenous bases: Adenine (A), Thymine (T), Cytosine (C), Guanine (G). | Carry the “letters” of the genetic alphabet. |
| **Base pairing** | A pairs with T, C pairs with G via hydrogen bonds (2 H‑bonds for A‑T, 3 for C‑G). | Ensures complementary strands can bind tightly and be read accurately. |
| **Major vs. Minor groove** | The A‑T and C‑G pairs create a wide, deep major groove and a shallow, narrow minor groove. | The major groove is the site where proteins (e.g., transcription factors, DNA‑binding proteins) dock. |
| **Sequence direction** | 5′ → 3′ (from 5‑phosphate to 3‑phosphate). | Sets the “handedness” of the helix and the direction of replication/transcription. |
| **Length & flexibility** | A typical eukaryotic chromosome is ~10⁶–10⁹ base pairs; DNA can be a few nanometers to centimeters long. | Allows packing of vast amounts of information in a single cell. |
### Visualizing the Ladder
5′-G‑C‑A‑T‑G‑C‑...‑3′ | | | | |
---
## 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}
}