Instructions to use llmvision/glimpse-v1 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use llmvision/glimpse-v1 with Transformers:

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

pipe = pipeline("image-text-to-text", model="llmvision/glimpse-v1")

# Load model directly
from transformers import AutoProcessor, AutoModelForImageTextToText

processor = AutoProcessor.from_pretrained("llmvision/glimpse-v1")
model = AutoModelForImageTextToText.from_pretrained("llmvision/glimpse-v1")

llama-cpp-python

How to use llmvision/glimpse-v1 with llama-cpp-python:

# !pip install llama-cpp-python

from llama_cpp import Llama

llm = Llama.from_pretrained(
	repo_id="llmvision/glimpse-v1",
	filename="glimpse-v1.BF16-mmproj.gguf",
)

output = llm(
	"Once upon a time,",
	max_tokens=512,
	echo=True
)
print(output)

Notebooks
Google Colab
Kaggle
Local Apps Settings

llama.cpp

How to use llmvision/glimpse-v1 with llama.cpp:

Install from brew

brew install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf llmvision/glimpse-v1:BF16
# Run inference directly in the terminal:
llama-cli -hf llmvision/glimpse-v1:BF16

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf llmvision/glimpse-v1:BF16
# Run inference directly in the terminal:
llama-cli -hf llmvision/glimpse-v1:BF16

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 llmvision/glimpse-v1:BF16
# Run inference directly in the terminal:
./llama-cli -hf llmvision/glimpse-v1:BF16

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 llmvision/glimpse-v1:BF16
# Run inference directly in the terminal:
./build/bin/llama-cli -hf llmvision/glimpse-v1:BF16

Use Docker

docker model run hf.co/llmvision/glimpse-v1:BF16

LM Studio
Jan

vLLM

How to use llmvision/glimpse-v1 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "llmvision/glimpse-v1"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "llmvision/glimpse-v1",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/llmvision/glimpse-v1:BF16

SGLang

How to use llmvision/glimpse-v1 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 "llmvision/glimpse-v1" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "llmvision/glimpse-v1",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

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 "llmvision/glimpse-v1" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "llmvision/glimpse-v1",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Ollama
How to use llmvision/glimpse-v1 with Ollama:
```
ollama run hf.co/llmvision/glimpse-v1:BF16
```

Unsloth Studio

How to use llmvision/glimpse-v1 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 llmvision/glimpse-v1 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 llmvision/glimpse-v1 to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for llmvision/glimpse-v1 to start chatting

Docker Model Runner
How to use llmvision/glimpse-v1 with Docker Model Runner:
```
docker model run hf.co/llmvision/glimpse-v1:BF16
```

Lemonade

How to use llmvision/glimpse-v1 with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull llmvision/glimpse-v1:BF16

Run and chat with the model

lemonade run user.glimpse-v1-BF16

List all available models

lemonade list

Recommended hardware

by Test107 - opened 6 days ago

Discussion

Test107

6 days ago

What would be the recommended hardware to run this model? does a Coral TPU help?

valentinfrlch

LLM Vision org 6 days ago

The easiest way to set up glimpse is by using Ollama. Unfortunately, since Ollama is based on llama.cpp, it does not support TPUs. Glimpse was tested on a mac mini, but I think other mini PCs will work as well. I'd be interested to hear what hardware you end up using, so I can collect some recommended hardware options.

enoquelights

4 days ago

I am using a GTX1660, 6G VRAM and am getting cudaMalloc failed: out of memory during CLIP init. I think this is to do with the mmproj, as I can start it with the base. I have tried ollama and llamacpp, both OOM. In llamacpp, it starts fine with the base model.

valentinfrlch

LLM Vision org 4 days ago

6GB VRAM should be plenty. But if you get OOM, you can give the quantization a try: https://ollama.com/llmvision/glimpse-v1:q4_k_m

There are even more quantizations available here: https://huggingface.co/mradermacher/glimpse-v1-GGUF

enoquelights

3 days ago

I got the Q4_K_M, thanks. No more OOM, but from Image Analyzer:
A woman standing in a driveway holding a black backpack and wearing a grey
tank top and striped pants. A blue car is parked nearby and a green trash can
is on the other side of the driveway.

<|im_start|>user

Image 2:<|im_end|>

<|im_start|>assistant

A man standing in a driveway holding a black object and wearing a dark shirt
and shorts. A blue car is parked nearby and a green trash can is on the other
side of the driveway.

<|im_start|>user

Image 3:<|im_end|>

<|im_start|>assistant
This outputs 16 times and never ends, just runs until out of tokens. I also needed to run it with --chat-template chatml to get it to output the "A man standing...." blocks.
Then for ollama(Open Webui):CDCl2
CDCl2 стоковое изображение
CDCl2
CDCl2 стоковое изображение
CDCl2
CDCl2 стоковое изображение
CDCl2
CDCl2 стоковое изображение
CDCl2
CDCl2 стоковое изображение
CDCl2
CDCl2 стоковое изображение
CDCl2
CDCl2 стоковое изображение
CDCl2
CDCl2 стоковое изображение
CDCl2
CDCl2 стоковое изображение
CDCl2
CDCl2 стоковое изображение
CDCl2
CDCl2 стоковое изображение
CDCl2
Seems like something is wrong with the stop tokens

valentinfrlch

LLM Vision org 3 days ago

I should probably put this information in the model card too, but be sure to use the original training instructions as prompt:
https://ollama.com/llmvision/glimpse-v1

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