Instructions to use sovthpaw/omnistep-12a3b with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use sovthpaw/omnistep-12a3b with Transformers:

# Load model directly
from transformers import AutoProcessor, AutoModel

processor = AutoProcessor.from_pretrained("sovthpaw/omnistep-12a3b")
model = AutoModel.from_pretrained("sovthpaw/omnistep-12a3b")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = processor.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=40)
print(processor.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

llama-cpp-python

How to use sovthpaw/omnistep-12a3b with llama-cpp-python:

# !pip install llama-cpp-python

from llama_cpp import Llama

llm = Llama.from_pretrained(
	repo_id="sovthpaw/omnistep-12a3b",
	filename="omnistep-12a3b-f16.gguf",
)

llm.create_chat_completion(
	messages = "No input example has been defined for this model task."
)

Notebooks
Google Colab
Kaggle
Local Apps Settings

llama.cpp

How to use sovthpaw/omnistep-12a3b with llama.cpp:

Install from brew

brew install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf sovthpaw/omnistep-12a3b:Q4_K_M
# Run inference directly in the terminal:
llama-cli -hf sovthpaw/omnistep-12a3b:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf sovthpaw/omnistep-12a3b:Q4_K_M
# Run inference directly in the terminal:
llama-cli -hf sovthpaw/omnistep-12a3b:Q4_K_M

Use pre-built binary

# Download pre-built binary from:
# https://github.com/ggerganov/llama.cpp/releases
# Start a local OpenAI-compatible server with a web UI:
./llama-server -hf sovthpaw/omnistep-12a3b:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf sovthpaw/omnistep-12a3b:Q4_K_M

Build from source code

git clone https://github.com/ggerganov/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-server llama-cli
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama-server -hf sovthpaw/omnistep-12a3b:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf sovthpaw/omnistep-12a3b:Q4_K_M

Use Docker

docker model run hf.co/sovthpaw/omnistep-12a3b:Q4_K_M

LM Studio
Jan
Ollama
How to use sovthpaw/omnistep-12a3b with Ollama:
```
ollama run hf.co/sovthpaw/omnistep-12a3b:Q4_K_M
```

Unsloth Studio

How to use sovthpaw/omnistep-12a3b with Unsloth Studio:

Install Unsloth Studio (macOS, Linux, WSL)

curl -fsSL https://unsloth.ai/install.sh | sh
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for sovthpaw/omnistep-12a3b to start chatting

Install Unsloth Studio (Windows)

irm https://unsloth.ai/install.ps1 | iex
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for sovthpaw/omnistep-12a3b to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for sovthpaw/omnistep-12a3b to start chatting

Atomic Chat new
Docker Model Runner
How to use sovthpaw/omnistep-12a3b with Docker Model Runner:
```
docker model run hf.co/sovthpaw/omnistep-12a3b:Q4_K_M
```

Lemonade

How to use sovthpaw/omnistep-12a3b with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull sovthpaw/omnistep-12a3b:Q4_K_M

Run and chat with the model

lemonade run user.omnistep-12a3b-Q4_K_M

List all available models

lemonade list

sovthpaw commited on 5 days ago

Commit

964eb36

verified ·

1 Parent(s): 85020ba

Update README.md

Browse files

Files changed (1) hide show

README.md +0 -3

README.md CHANGED Viewed

@@ -289,9 +289,6 @@ r_final(T) = τ · r_MRI(T) + (1 - τ) · r_genome(T)
 r_MRI(T) = MRI_B(T) / (MRI_A(T) + MRI_B(T))
 MRI(T)  = α · Static(T) + (1 - α) · Probe(T),  α = 0.5 (paper-fixed)
 ```
-**The merge result:** the Qwen2.5-Omni text body (Qwen2.5-3B class, hidden=2048, intermediate=11008, num_kv_heads=2, 36L) and the ACE-Step v1.5 XL text encoder (Qwen3 class, hidden=2048, intermediate=6144, num_kv_heads=8, 32L) have **zero tensors with matching shape**. The Architecture Mapper's "skip on dim mismatch, keep parent A" behavior is applied for the entire shared-key set, preserving the Qwen2.5-Omni text body intact and adding the ACE encoder components as a separate module.
 The **fixed starting genome** used (paper-recommended):
 ```json

 r_MRI(T) = MRI_B(T) / (MRI_A(T) + MRI_B(T))
 MRI(T)  = α · Static(T) + (1 - α) · Probe(T),  α = 0.5 (paper-fixed)
 ```
 The **fixed starting genome** used (paper-recommended):
 ```json