Instructions to use Minachist/Ornith-1.0-35B-INT8-AutoRound with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Minachist/Ornith-1.0-35B-INT8-AutoRound with Transformers:

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

pipe = pipeline("text-generation", model="Minachist/Ornith-1.0-35B-INT8-AutoRound")
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("Minachist/Ornith-1.0-35B-INT8-AutoRound")
model = AutoModelForMultimodalLM.from_pretrained("Minachist/Ornith-1.0-35B-INT8-AutoRound")
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 Minachist/Ornith-1.0-35B-INT8-AutoRound with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Minachist/Ornith-1.0-35B-INT8-AutoRound"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Minachist/Ornith-1.0-35B-INT8-AutoRound",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/Minachist/Ornith-1.0-35B-INT8-AutoRound

SGLang

How to use Minachist/Ornith-1.0-35B-INT8-AutoRound 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 "Minachist/Ornith-1.0-35B-INT8-AutoRound" \
    --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": "Minachist/Ornith-1.0-35B-INT8-AutoRound",
		"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 "Minachist/Ornith-1.0-35B-INT8-AutoRound" \
        --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": "Minachist/Ornith-1.0-35B-INT8-AutoRound",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use Minachist/Ornith-1.0-35B-INT8-AutoRound with Docker Model Runner:
```
docker model run hf.co/Minachist/Ornith-1.0-35B-INT8-AutoRound
```

Ornith-1.0-35B-INT8-AutoRound / quantize.py

Minachist

Update quantize.py

9974f95 verified 13 days ago

Raw

History Blame Contribute Delete

2.24 kB

	import torch
	from transformers import AutoTokenizer
	from auto_round import AutoRound

	# You need to use datasets 3.6.0 since >=4.0 doesn't support codeparrot/github-code-clean dataset
	# You also need torchvision for some reason even if it's not used, otherwise it will throw an error "Unrecognized image processor ...".
	# TORCHVISION_AVAILABLE = False is there because datasets 3.6.0 uses the old torchvision api, and isn't compatible w/ this torchvision.'(And it will throw ImportError: cannot import name 'VideoReader' from 'torchvision.io')
	# Downgrading torchvision might solve the issue, but there's also a chance that doing that breaks other deps. Since I don't want to damage my brain anymore, I decided to just simply told datasets not to use torchvision.
	# To replicate the setup:
	# uv venv --python 3.13 .venv
	# source .venv/bin/activate
	# uv pip install "torch==2.12.1" "torchvision==0.27.1" \
	# --index-url https://download.pytorch.org/whl/cu130
	# uv pip install "transformers==5.12.1" "auto-round-nightly==0.14.0.dev20260625" \
	# "accelerate==1.14.0" "datasets==3.6.0" "pillow"
	# This is why I don't like Python, it's dependency hell.
	import datasets
	datasets.config.TORCHVISION_AVAILABLE = False

	model_name_or_path = "."
	output_dir = "./Ornith-1.0-35B-INT8-AutoRound/"

	tokenizer = AutoTokenizer.from_pretrained(model_name_or_path, trust_remote_code=True)

	# auto-round doesn't support quantizing linear_attn, but I wanted to be sure.
	ignore_keywords = [
	"embed_tokens",
	"linear_attn",
	"shared_expert",
	"mlp.gate",
	"visual",
	"lm_head",
	]

	layer_config = {}
	for keyword in ignore_keywords:
	layer_config[keyword] = {"bits": 16}

	seqlen = 2048
	nsamples = 1024

	dataset = "NeelNanda/pile-10k:num=256,codeparrot/github-code-clean:num=768"

	ar = AutoRound(
	model=model_name_or_path,
	tokenizer=tokenizer,
	scheme="W8A16",
	enable_torch_compile=False,
	group_size=-1,
	sym=True,
	layer_config=layer_config,

	dataset=dataset,

	device_map="0,1",
	batch_size=8,
	seqlen=seqlen,
	iters=1000,
	nsamples=nsamples,
	low_gpu_mem_usage=True,
	)

	ar.quantize_and_save(output_dir, format="auto_round")