Instructions to use daedalus2027/Ouro-1.4B-bucket with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use daedalus2027/Ouro-1.4B-bucket with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="daedalus2027/Ouro-1.4B-bucket", trust_remote_code=True)
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoModelForCausalLM
model = AutoModelForCausalLM.from_pretrained("daedalus2027/Ouro-1.4B-bucket", trust_remote_code=True, device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use daedalus2027/Ouro-1.4B-bucket with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "daedalus2027/Ouro-1.4B-bucket"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "daedalus2027/Ouro-1.4B-bucket",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/daedalus2027/Ouro-1.4B-bucket

SGLang

How to use daedalus2027/Ouro-1.4B-bucket with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "daedalus2027/Ouro-1.4B-bucket" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "daedalus2027/Ouro-1.4B-bucket",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "daedalus2027/Ouro-1.4B-bucket" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "daedalus2027/Ouro-1.4B-bucket",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use daedalus2027/Ouro-1.4B-bucket with Docker Model Runner:
```
docker model run hf.co/daedalus2027/Ouro-1.4B-bucket
```

daedalus2027 commited on Jun 12

Commit

69e7537

verified ·

1 Parent(s): 03d35ba

Upload README.md with huggingface_hub

Browse files

Files changed (1) hide show

README.md +4 -11

README.md CHANGED Viewed

@@ -10,7 +10,9 @@ tags:
 # Ouro-1.4B
-📚 [Paper](https://huggingface.co/papers/2510.25741) • 🏠 [Project Page](https://ouro-llm.github.io/) • ⚡ [HF Hub (fork)](https://huggingface.co/daedalus2027/Ouro-1.4B-bucket)
 ## Model Description
@@ -19,6 +21,7 @@ tags:
 **Ouro-1.4B** is a 1.4 billion parameter Looped Language Model (LoopLM) that achieves exceptional parameter efficiency through iterative shared-weight computation.
 ## Key Features
@@ -112,16 +115,6 @@ print(tokenizer.decode(outputs[0], skip_special_tokens=True))
 ## Acknowledgments
 We thank [@Antizana](https://github.com/Antizana) for the KV cache fix merged from [ouro-cache-fix](https://github.com/Antizana/ouro-cache-fix), which resolved a critical compatibility issue with transformers>=4.56.0.
-## Fixes Applied
-This fork includes the following fixes over the original ByteDance release:
-### RoPE Initialization Safety
-The original `OuroRotaryEmbedding.__init__` called `self.rope_init_fn(self.config, device)` directly, which could fail when the target device wasn't ready (e.g., CUDA not yet initialized). The fix defers device-specific tensor creation: `inv_freq` is computed on CPU first, then moved to the target device in `forward()`. A fallback is also provided for `None` `rope_init_fn`.
-### `pad_token_id` in Config
-Added `"pad_token_id": 0` to `config.json`, required by the `nn.Embedding(padding_idx=...)` layer. Without it, the embedding layer could misbehave under certain padding conditions.
 ## Citation
 ```bibtex

 # Ouro-1.4B
+📚 [Paper](https://huggingface.co/papers/2510.25741) • 🏠 [Project Page](https://ouro-llm.github.io/)
+![Ouro Logo](assets/logo.png)
 ## Model Description
 **Ouro-1.4B** is a 1.4 billion parameter Looped Language Model (LoopLM) that achieves exceptional parameter efficiency through iterative shared-weight computation.
+![Model Performance](assets/benchmark.png)
 ## Key Features
 ## Acknowledgments
 We thank [@Antizana](https://github.com/Antizana) for the KV cache fix merged from [ouro-cache-fix](https://github.com/Antizana/ouro-cache-fix), which resolved a critical compatibility issue with transformers>=4.56.0.
 ## Citation
 ```bibtex