Instructions to use ornith-ai/Ornith-1.5-397B-NVFP4 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use ornith-ai/Ornith-1.5-397B-NVFP4 with Transformers:

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

pipe = pipeline("text-generation", model="ornith-ai/Ornith-1.5-397B-NVFP4")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"},
            {"type": "text", "text": "What animal is on the candy?"}
        ]
    },
]
pipe(text=messages)

# Load model directly
from transformers import AutoProcessor, AutoModelForMultimodalLM

processor = AutoProcessor.from_pretrained("ornith-ai/Ornith-1.5-397B-NVFP4")
model = AutoModelForMultimodalLM.from_pretrained("ornith-ai/Ornith-1.5-397B-NVFP4", device_map="auto")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"},
            {"type": "text", "text": "What animal is on the candy?"}
        ]
    },
]
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]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use ornith-ai/Ornith-1.5-397B-NVFP4 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "ornith-ai/Ornith-1.5-397B-NVFP4"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "ornith-ai/Ornith-1.5-397B-NVFP4",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/ornith-ai/Ornith-1.5-397B-NVFP4

SGLang

How to use ornith-ai/Ornith-1.5-397B-NVFP4 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 "ornith-ai/Ornith-1.5-397B-NVFP4" \
    --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": "ornith-ai/Ornith-1.5-397B-NVFP4",
		"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 "ornith-ai/Ornith-1.5-397B-NVFP4" \
        --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": "ornith-ai/Ornith-1.5-397B-NVFP4",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use ornith-ai/Ornith-1.5-397B-NVFP4 with Docker Model Runner:
```
docker model run hf.co/ornith-ai/Ornith-1.5-397B-NVFP4
```

Can 2x PRO 6000 handle it?

by shsky - opened 3 days ago

Discussion

shsky

3 days ago

Can 2x PRO 6000 handle it?

mwyborski

about 1 hour ago

•

edited about 1 hour ago

I had to place 17 experts (each about 3.5GB) on my 4x 3090s to have space for 1M context and mmproj. I had good performance somehow with this placement around 40-50 tok/s. This placement probably isn't the optimal, but a starting point.

LLAMA_CPP_SERVER_ARGS="-m /mnt/models/Ornith-1.5-397B-NVFP4/Ornith-1.5-397B-NVFP4-nomtp.gguf -mm /mnt/models/Ornith-1.5-397B-NVFP4/mmproj-Ornith-1.5-397B-BF16.gguf --alias ornith15-nvfp4 --host 127.0.0.1 --port 30014 --device CUDA0,CUDA1,CUDA2,CUDA3,CUDA4,CUDA5 -ngl all -sm layer -ts 1,1,0,0,0,0 --fit off -ot blk\.(0|36|42|54)\.ffn_.*_exps.*=CUDA2,blk\.(3|12|21|33)\.ffn_.*_exps.*=CUDA3,blk\.(6|15|24|39|45)\.ffn_.*_exps.*=CUDA4,blk\.(9|18|51|57)\.ffn_.*_exps.*=CUDA5 -c 1048576 -ctk q8_0 -ctv q8_0 -fa on --rope-scaling yarn --rope-scale 4 --yarn-orig-ctx 262144 -b 2048 -ub 512 -t 8 -tb 32 --parallel 4 --kv-unified -cram 0 --no-cache-idle-slots --temp 1.0 --top-p 0.95 --top-k 20 --jinja"

CUDA0, CUDA1 are the RTX Pro 6000
CUDA2,CUDA3,CUDA4,CUDA5 are RTX 3090s

Fitting the 17 experts on only 3x3090 also worked btw.
6,6,5 split
--device CUDA0,CUDA1,CUDA2,CUDA3,CUDA4

-ot blk\.(0|9|18|36|45|54)\.ffn_.*_exps.*=CUDA2,blk\.(3|12|21|33|42|51)\.ffn_.*_exps.*=CUDA3,blk\.(6|15|24|39|48)\.ffn_.*_exps.*=CUDA4

Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.

Tap or paste here to upload images

· Sign up or log in to comment