Model Card for Pixtral-12B-2409
The Pixtral-12B-2409 is a Multimodal Model of 12B parameters plus a 400M parameter vision encoder.
For more details about this model please refer to our release
blog post .
Key features
Natively multimodal, trained with interleaved image and text data
12B parameter Multimodal Decoder + 400M parameter Vision Encoder
Supports variable image sizes
Leading performance in its weight class on multimodal tasks
Maintains state-of-the-art performance on text-only benchmarks
Sequence length: 128k
License: Apache 2.0
Benchmarks
The performance of Pixtral-12B-2409 compared to multimodal models.
Multimodal Benchmarks
Pixtral 12B Qwen2 7B VL LLaVA-OV 7B Phi-3 Vision Phi-3.5 Vision MMMU (CoT) 52.5 47.6 45.1 40.3 38.3 Mathvista (CoT) 58.0 54.4 36.1 36.4 39.3 ChartQA (CoT) 81.8 38.6 67.1 72.0 67.7 DocVQA (ANLS) 90.7 94.5 90.5 84.9 74.4 VQAv2 (VQA Match) 78.6 75.9 78.3 42.4 56.1
Instruction Following
Pixtral 12B Qwen2 7B VL LLaVA-OV 7B Phi-3 Vision Phi-3.5 Vision MM MT-Bench 6.05 5.43 4.12 3.70 4.46 Text MT-Bench 7.68 6.41 6.94 6.27 6.31 MM IF-Eval 52.7 38.9 42.5 41.2 31.4 Text IF-Eval 61.3 50.1 51.4 50.9 47.4
Text Benchmarks
Pixtral 12B Qwen2 7B VL LLaVA-OV 7B Phi-3 Vision Phi-3.5 Vision MMLU (5-shot) 69.2 68.5 67.9 63.5 63.6 Math (Pass@1) 48.1 27.8 38.6 29.2 28.4 Human Eval (Pass@1) 72.0 64.6 65.9 48.8 49.4
Comparison with Closed Source and Larger Models
Pixtral 12B Claude-3 Haiku Gemini-1.5 Flash 8B (0827) . LLaVA-OV 72B GPT-4o Claude-3.5 Sonnet MMMU (CoT) 52.5 50.4 50.7 54.4 68.6 68.0 Mathvista (CoT) 58.0 44.8 56.9 57.2 64.6 64.4 ChartQA (CoT) 81.8 69.6 78.0 66.9 85.1 87.6 DocVQA (ANLS) 90.7 74.6 79.5 91.6 88.9 90.3 VQAv2 (VQA Match) 78.6 68.4 65.5 83.8 77.8 70.7
Usage Examples
vLLM (recommended)
We recommend using Pixtral with the
vLLM library
to implement production-ready inference pipelines with Pixtral.
Installation
Make sure you install vLLM >= v0.6.2:
pip install --upgrade vllmAlso make sure you have mistral_common >= 1.4.4 installed:
pip install --upgrade mistral_commonYou can also make use of a ready-to-go
docker image .
Simple Example
1 from vllm import LLM
2 from vllm . sampling_params import SamplingParams
3
4 model_name = "mistralai/Pixtral-12B-2409"
5
6 sampling_params = SamplingParams ( max_tokens = 8192 )
7
8 llm = LLM ( model = model_name , tokenizer_mode = "mistral" )
9
10 prompt = "Describe this image in one sentence."
11 image_url = "https://picsum.photos/id/237/200/300"
12
13 messages = [
14 {
15 "role" : "user" ,
16 "content" : [ { "type" : "text" , "text" : prompt } , { "type" : "image_url" , "image_url" : { "url" : image_url } } ]
17 } ,
18 ]
19
20 outputs = llm . chat ( messages , sampling_params = sampling_params )
21
22 print ( outputs [ 0 ] . outputs [ 0 ] . text )
Advanced Example
You can also pass multiple images per message and/or pass multi-turn conversations
1 from vllm import LLM
2 from vllm . sampling_params import SamplingParams
3
4 model_name = "mistralai/Pixtral-12B-2409"
5 max_img_per_msg = 5
6
7 sampling_params = SamplingParams ( max_tokens = 8192 , temperature = 0.7 )
8
9 # Lower max_num_seqs or max_model_len on low-VRAM GPUs.
10 llm = LLM ( model = model_name , tokenizer_mode = "mistral" , limit_mm_per_prompt = { "image" : max_img_per_msg } , max_model_len = 32768 )
11
12 prompt = "Describe the following image."
13
14 url_1 = "https://huggingface.co/datasets/patrickvonplaten/random_img/resolve/main/yosemite.png"
15 url_2 = "https://picsum.photos/seed/picsum/200/300"
16 url_3 = "https://picsum.photos/id/32/512/512"
17
18 messages = [
19 {
20 "role" : "user" ,
21 "content" : [ { "type" : "text" , "text" : prompt } , { "type" : "image_url" , "image_url" : { "url" : url_1 } } , { "type" : "image_url" , "image_url" : { "url" : url_2 } } ] ,
22 } ,
23 {
24 "role" : "assistant" ,
25 "content" : "The images shows nature." ,
26 } ,
27 {
28 "role" : "user" ,
29 "content" : "More details please and answer only in French!."
30 } ,
31 {
32 "role" : "user" ,
33 "content" : [ { "type" : "image_url" , "image_url" : { "url" : url_3 } } ] ,
34 }
35 ]
36
37 outputs = llm . chat ( messages = messages , sampling_params = sampling_params )
38 print ( outputs [ 0 ] . outputs [ 0 ] . text )
You can find more examples and tests directly in vLLM.
Server
You can also use pixtral in a server/client setting.
Spin up a server:
vllm serve mistralai/Pixtral-12B-2409 --tokenizer_mode mistral --limit_mm_per_prompt 'image=4'
And ping the client:
curl --location 'http://<your-node-url>:8000/v1/chat/completions' \
--header 'Content-Type: application/json' \
--header 'Authorization: Bearer token' \
--data '{
"model": "mistralai/Pixtral-12B-2409",
"messages": [
{
"role": "user",
"content": [
{"type" : "text", "text": "Describe this image in detail please."},
{"type": "image_url", "image_url": {"url": "https://s3.amazonaws.com/cms.ipressroom.com/338/files/201808/5b894ee1a138352221103195_A680%7Ejogging-edit/A680%7Ejogging-edit_hero.jpg"}},
{"type" : "text", "text": "and this one as well. Answer in French."},
{"type": "image_url", "image_url": {"url": "https://www.wolframcloud.com/obj/resourcesystem/images/a0e/a0ee3983-46c6-4c92-b85d-059044639928/6af8cfb971db031b.png"}}
]
}
]
}'Mistral-inference
We recommend using
mistral-inference to quickly try out / "vibe-check" Pixtral.
Install
Make sure to have mistral_inference >= 1.4.1 installed.
pip install mistral_inference --upgradeDownload
1 from huggingface_hub import snapshot_download
2 from pathlib import Path
3
4 mistral_models_path = Path . home ( ) . joinpath ( 'mistral_models' , 'Pixtral' )
5 mistral_models_path . mkdir ( parents = True , exist_ok = True )
6
7 snapshot_download ( repo_id = "mistralai/Pixtral-12B-2409" , allow_patterns = [ "params.json" , "consolidated.safetensors" , "tekken.json" ] , local_dir = mistral_models_path )
Chat
After installing mistral_inference, a mistral-chat CLI command should be available in your environment.
You can pass text and images or image urls to the model in instruction-following mode as follows:
mistral-chat $HOME/mistral_models/Pixtral --instruct --max_tokens 256 --temperature 0.35E.g. Try out something like:
Text prompt: What can you see on the following picture?
[You can input zero, one or more images now.]
Image path or url [Leave empty and press enter to finish image input]: https://picsum.photos/id/237/200/300
Image path or url [Leave empty and press enter to finish image input]:
I see a black dog lying on a wooden surface. The dog appears to be looking up, and its eyes are clearly visible.Python
You can also run the model in a Python shell as follows.
1 from mistral_inference . transformer import Transformer
2 from mistral_inference . generate import generate
3
4 from mistral_common . tokens . tokenizers . mistral import MistralTokenizer
5 from mistral_common . protocol . instruct . messages import UserMessage , TextChunk , ImageURLChunk
6 from mistral_common . protocol . instruct . request import ChatCompletionRequest
7
8 tokenizer = MistralTokenizer . from_file ( f" { mistral_models_path } /tekken.json" )
9 model = Transformer . from_folder ( mistral_models_path )
10
11 url = "https://huggingface.co/datasets/patrickvonplaten/random_img/resolve/main/yosemite.png"
12 prompt = "Describe the image."
13
14 completion_request = ChatCompletionRequest ( messages = [ UserMessage ( content = [ ImageURLChunk ( image_url = url ) , TextChunk ( text = prompt ) ] ) ] )
15
16 encoded = tokenizer . encode_chat_completion ( completion_request )
17
18 images = encoded . images
19 tokens = encoded . tokens
20
21 out_tokens , _ = generate ( [ tokens ] , model , images = [ images ] , max_tokens = 256 , temperature = 0.35 , eos_id = tokenizer . instruct_tokenizer . tokenizer . eos_id )
22 result = tokenizer . decode ( out_tokens [ 0 ] )
23
24 print ( result )
Limitations
The Pixtral model does not have any moderation mechanisms. We're looking forward to engaging with the community on ways to
make the model finely respect guardrails, allowing for deployment in environments requiring moderated outputs.
The Mistral AI Team
Albert Jiang, Alexandre Sablayrolles, Alexis Tacnet, Alok Kothari, Antoine Roux, Arthur Mensch, Audrey Herblin-Stoop, Augustin Garreau, Austin Birky, Bam4d, Baptiste Bout, Baudouin de Monicault, Blanche Savary, Carole Rambaud, Caroline Feldman, Devendra Singh Chaplot, Diego de las Casas, Diogo Costa, Eleonore Arcelin, Emma Bou Hanna, Etienne Metzger, Gaspard Blanchet, Gianna Lengyel, Guillaume Bour, Guillaume Lample, Harizo Rajaona, Henri Roussez, Hichem Sattouf, Ian Mack, Jean-Malo Delignon, Jessica Chudnovsky, Justus Murke, Kartik Khandelwal, Lawrence Stewart, Louis Martin, Louis Ternon, Lucile Saulnier, Lélio Renard Lavaud, Margaret Jennings, Marie Pellat, Marie Torelli, Marie-Anne Lachaux, Marjorie Janiewicz, Mickaël Seznec, Nicolas Schuhl, Niklas Muhs, Olivier de Garrigues, Patrick von Platen, Paul Jacob, Pauline Buche, Pavan Kumar Reddy, Perry Savas, Pierre Stock, Romain Sauvestre, Sagar Vaze, Sandeep Subramanian, Saurabh Garg, Soham Ghosh, Sophia Yang, Szymon Antoniak, Teven Le Scao, Thibault Schueller, Thibaut Lavril, Thomas Wang, Théophile Gervet, Timothée Lacroix, Valera Nemychnikova, Wendy Shang, William El Sayed, William Marshall
