Skip to content
Local AI Wizard
Your machine
Theme

Perception-LM-8B

Meta · facebook/Perception-LM-8B

Weights license: Publisher terms: FAIR Noncommercial Research · code and components may use other terms

Reported by the publisher40 resultsMeta's model card · read 2026-10-25 · not ranked · see them
Memory needed~8 GBQ4 weights + working room, before the machine's reserve
Parameters9.8 BHub safetensors total
Repository updated2025-07-14A commit is not a release date

Quality

No comparable score

We track no public, comparable benchmark that covers this model yet. No score is inferred.

Reported by the publisher

40 results · Meta's model card · read 2026-10-25
DocVQA
94.6PLM Image Benchmark Results
ChartQA
85.5PLM Image Benchmark Results
TextVQA
86.5PLM Image Benchmark Results
InfoQA
80.9PLM Image Benchmark Results
AI2D
92.7PLM Image Benchmark Results
OCRBench
870PLM Image Benchmark Results
COCO
146.7PLM Image Benchmark Results
Nocap
129.9PLM Image Benchmark Results
Every reported result40
Flickr
105.6PLM Image Benchmark Results
MMMU
46.1PLM Image Benchmark Results
VQAv2
85.6PLM Image Benchmark Results
OKVQA
69.6PLM Image Benchmark Results
VizWiz
67.0PLM Image Benchmark Results
MME
1,989PLM Image Benchmark Results
SEED
79.3PLM Image Benchmark Results
BLINK
56.0PLM Image Benchmark Results
CVBench
81.3PLM Image Benchmark Results
RealWorldQA
75.0PLM Image Benchmark Results
VSR
82.8PLM Image Benchmark Results
POPE
89.9PLM Image Benchmark Results
VATEX
99.7PLM Video Benchmark Results
DREAM 1K
35.9PLM Video Benchmark Results
How2QA
90.7PLM Video Benchmark Results
MVBench
77.1PLM Video Benchmark Results
NExTQA
84.1PLM Video Benchmark Results
PerceptionTest (test)
82.7PLM Video Benchmark Results
STAR
84.9PLM Video Benchmark Results
TVQA
59.3PLM Video Benchmark Results
VideoMME
58.3PLM Video Benchmark Results
TVBench
63.5PLM Video Benchmark Results
ActivityNetQA
67.3PLM Video Benchmark Results
EgoSchema (test)
68.8PLM Video Benchmark Results
TemporalBench
28.3PLM Video Benchmark Results
TOMATO
33.2PLM Video Benchmark Results
MotionBench (dev)
61.4PLM Video Benchmark Results
TempCompass (MCQ)
72.7PLM Video Benchmark Results
CGBench (clue)
46.4PLM Video Benchmark Results
Charades STA
58.6PLM Video Benchmark Results
VideoHallucer
57.7PLM Video Benchmark Results
Halluc. EventHallusion
77.3PLM Video Benchmark Results

Values as printed on the card, with the publisher's own settings (card section: PLM Image Benchmark Results; PLM Video Benchmark Results); column “PLM8B”. Not comparable across publishers and not used in any ranking. Model card

Runs on

Fast memory holds it 90

The planning estimate fits the memory the model can use quickly: graphics card VRAM, or unified memory within the macOS GPU share.

Show all 90

Within memory, outside the fast part 26

Within the machine's memory, outside its fast part: on a graphics card the layers that do not fit run on the CPU from PC RAM (the listed RAM, or 32 GB assumed); on a Mac, above the macOS default GPU share. Not a speed figure.

Show all 26

One memory tier short 3

The same machines with the next memory size up would hold it in fast memory.

Hardware profiles grouped by the Q4 planning estimate on each machine (dedicated GPUs reserve 2 GB; unified and system memory 25%, 2 to 8 GB): the fast memory holds it, it is within memory but outside the fast part (layers offloaded to PC RAM, or above the macOS GPU share), or the machine is one memory tier short. Not a measured fit or speed.

Consider it for

Meta released Perception-LM-8B on 2025-04-17: a vision-language model that answers about images, with 9.79B parameters on the Hugging Face count and the Publisher terms: FAIR Noncommercial Research license. Languages the publisher declares: en.

Check first

License: commercial use not allowed. At Q4 the weights load about 7.3 GB before each machine's reserve; image resolution, video frames and context add memory, and the vision encoder needs support in the runtime you pick. Downloading requires accepting the publisher's terms on Hugging Face.

Measured speed

No published measurement for this checkpoint

Tokens per second are shown only when a source publishes them with hardware, build and setup. They are never estimated.

How to run

Runtime compatibility and tokens per second have not been measured for this entry. Published metadata and a memory estimate do not prove a successful local run.

Documented ways to run it

Copied as printed in the linked source (read 2026-10-18); not run by this site.

Perception Models · README · PLM Resources
python apps/plm/generate.py \
--ckpt facebook/Perception-LM-8B \
--media_type video \
--media_path apps/plm/dummy_datasets/video/videos/GUWR5TyiY-M_000012_000022.mp4 \
--question 'What is happening in the video?'

Choose my computer

Technical details

What the source and the catalog store for this checkpoint. Missing fields are left out, never zero.

Checkpoint / source
facebook/Perception-LM-8B
Task
image-text-to-text
Architecture
PerceptionLMForConditionalGeneration
Weight license
Publisher terms: FAIR Noncommercial ResearchLICENSE.PLM (facebookresearch/perception_models, GitHub) · read 2026-10-18source
Commercial use
Not allowed FAIR Noncommercial Research License v1 (facebookresearch) · read 2026-10-18license text
Languages
enModel card metadata · read 2026-10-25source
Released
facebookresearch/perception_models README updates (GitHub, Meta FAIR) · read 2026-10-18source
Every stored field12 fields
Revision checked
969497c228e880aacb82693813f3169330bec9c8
Architecture
PerceptionLMForConditionalGeneration
Hub task tag
image-text-to-text
Formats
safetensors, pytorch
Access
Gated (manual)
Publisher kind
Official
Hub parameter count
9,794,260,736safetensors.total
Repository created
2025-04-10creation is not the model's release
Repository updated
2025-07-14a commit is not a release date
Hub downloads (30 days) · likes
355 · 67this repository only · read 2026-10-25
Read from the Hub page
task tagsHub overview and file listing · read 2026-10-18
Catalog state
Hub config or weights found; no inventory benchmark link

Sources

Identity and published specifications come from the checkpoint page. Benchmark scores belong to the named release and do not measure speed on your computer.

Curated facts checked: 2026-10-18. The license and release date cite the publisher's sources; parameters, files and task come from the Hugging Face record, and memory is a labelled estimate from its parameter count. A value no source states stays empty; a repository commit is not a model release date.