Instructions to use h2oai/h2ovl-mississippi-800m with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use h2oai/h2ovl-mississippi-800m with Transformers:

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

pipe = pipeline("text-generation", model="h2oai/h2ovl-mississippi-800m", trust_remote_code=True)
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoModel
model = AutoModel.from_pretrained("h2oai/h2ovl-mississippi-800m", trust_remote_code=True, device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use h2oai/h2ovl-mississippi-800m with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "h2oai/h2ovl-mississippi-800m"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "h2oai/h2ovl-mississippi-800m",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/h2oai/h2ovl-mississippi-800m

SGLang

How to use h2oai/h2ovl-mississippi-800m 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 "h2oai/h2ovl-mississippi-800m" \
    --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": "h2oai/h2ovl-mississippi-800m",
		"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 "h2oai/h2ovl-mississippi-800m" \
        --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": "h2oai/h2ovl-mississippi-800m",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use h2oai/h2ovl-mississippi-800m with Docker Model Runner:
```
docker model run hf.co/h2oai/h2ovl-mississippi-800m
```

fix: compute stochastic-depth rates in pure Python for transformers 5.x meta-device init

by brandenh2oai - opened Jul 15

base: refs/heads/main

←

from: refs/pr/4

Discussion Files changed

+53

-5

brandenh2oai

H2O.ai org Jul 15

Fix `Tensor.item() cannot be called on meta tensors` under transformers 5.x

InternVisionEncoder.__init__ computes the stochastic-depth decay rule with
torch.linspace(...).item(). As of transformers 5.x, from_pretrained runs
model __init__ inside a with torch.device("meta") context, so
torch.linspace(...) returns a meta tensor and .item() raises
RuntimeError: Tensor.item() cannot be called on meta tensors. This makes the
model fail to load via AutoModel.from_pretrained(..., trust_remote_code=True)
on transformers>=5.

This computes the identical decay values in pure Python (no device-dependent
tensor), so init works regardless of the ambient device context. The values
match torch.linspace(0, drop_path_rate, num_hidden_layers) to within float32
rounding (~5e-9), which is inconsequential for drop-path probabilities.

fix: compute stochastic-depth rates in pure Python for transformers 5.x meta-device inite53a7407

fix: complete transformers 5.x compatibility (backward compatible with 4.x)a60a0b23

ilu000 changed pull request status to merged Jul 16

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

fix: compute stochastic-depth rates in pure Python for transformers 5.x meta-device init

Fix Tensor.item() cannot be called on meta tensors under transformers 5.x

Fix `Tensor.item() cannot be called on meta tensors` under transformers 5.x