Up to date
meta-llama Llama 3.2 Community License out 2 years ago
88.6B parameters, images and text in, text out.
Verified 2026-10-25 Rank #648 in LLMs · score 73.1 method v1 Sources 3 Downloads 30d 508 · #607 of 619 tracked
Picked by rule from 25 tracked pieces: the lab's own post, the most shared, a field report, the latest.
The project trained a Sparse Autoencoder on 6.5 Billion activations from Llama 3.2-3B layer 23 over 10 epochs.
A minimal implementation of Llama 3.2 Instruct inference runs in ~100 lines of pure Numpy code.
The 2 picks above, plus these 23. Top is the radar's score: discussion, source authority and the lab's own posts, fading with age.
Hugging Face does not say who downloads a model or with what. These are the proxies: API providers listing it, quantized builds, fine-tunes/adapters/merges, and Spaces that load it.
No API provider serves it, so its downloads are people running it themselves: 6 quantized builds and 100 Spaces.
Top Spaces running it: algorithmicsuperintelligence/OptiLLM ♥ 27 · NiansuhAI/Main ♥ 10 · davehusk/meta-llama-Llama-3.2-90B-Vision-Instruct ♥ 3 · tavli/anne-kalori-asistani ♥ 2
Weights only, an estimate: leave room for the KV cache. Method
About 53 GB at INT4, 106 GB at FP8, 213 GB at BF16: weights only, so leave room for the KV cache.
Estimate params × bytes per param × 1.2 · how we estimate
Weights, every dated article, and repo commits on one rail.
Teal is the weights (2024-09-19), amber triangles are 10 dated articles (each links to it).
15 derivatives on the Hub. Which build to run, by runtime: the most downloaded in each class.
From the Hugging Face model tree; quantized builds in teal, fine-tunes, merges and adapters in greys.
Everything else, folded: the full tables, every source.
Labels are evidence, not verdicts: measured a third-party leaderboard row for this model; author-reported a number the model author published in the card or eval files; structural a deterministic signal from the name, config or chat template, or a decision-model estimate. Each chip links to its source. All use cases.
Listed in LLMs
| Pipeline tag | image-text-to-text |
|---|---|
| Library | transformers |
| Architectures | MllamaForConditionalGeneration |
| Model type | mllama |
| Params by dtype | BF16: 88,593,355,323 |
| Size band | XL |
| MoE | no |
| Gated | manual |
| Papers | arXiv:2204.05149 |
| License | Llama 3.2 Community License (llama3.2) |
| Hub snapshot | 2026-10-25 |
vLLM: not in registry · MllamaForConditionalGeneration · 7050ef881ff0835f6890a47db1859a24c65481d6SGLang: supported · MllamaForConditionalGeneration · 773cb938fe9484a8f7e7e46fb1b1fe96412c59ed
vLLM registry (7050ef881ff0835f6890a47db1859a24c65481d6) · SGLang registry (773cb938fe9484a8f7e7e46fb1b1fe96412c59ed) · Matched by architecture class name against each project's registry on main; 'supported' means the architecture is registered, not that this checkpoint was tested.
Weights-only memory for Llama-3.2-90B-Vision-Instruct: params × bytes/param × 1.2 overhead. KV cache and activations are extra. How this is computed.
| BF16 (2 bytes/param) | 213 GB · Multi-GPU node (<640 GB) |
|---|---|
| FP8 (1 byte/param) | 106 GB · Multi-GPU node (<640 GB) |
| INT4 (0.5 bytes/param) | 53 GB · Single high-end GPU (<80 GB) |
Weights only (params × bytes/param × 1.2 overhead); KV cache and activations extra.
quantized: 6finetune: 6adapter: 3
| Variant | Relation | Quant | Params | Band | License | Author | Likes | Downloads |
|---|---|---|---|---|---|---|---|---|
| RedHatAI/Llama-3.2-90B-Vision-Instruct-FP8-dynamic | quantized | -FP8 | 88.6B | XL | llama3.2 | RedHatAI | 12 | 445.5k |
| unsloth/Llama-3.2-90B-Vision-Instruct-bnb-4bit | quantized | -bnb | 91.3B | XL | llama3.2 | unsloth | 19 | 57.7k |
| unsloth/Llama-3.2-90B-Vision-Instruct | finetune | — | 88.6B | XL | llama3.2 | unsloth | 18 | 40.7k |
| mlx-community/Llama-3.2-90B-Vision-Instruct-4bit | finetune | -4bit | 88.6B | XL | llama3.2 | mlx-community | 5 | 16.6k |
| OPEA/Llama-3.2-90B-Vision-Instruct-int4-sym-inc | quantized | int4 | 14.5B | L | llama3.2 | OPEA | 0 | 2.9k |
| amd/Llama-3.2-90B-Vision-Instruct-FP8-KV | quantized | -FP8 | 88.6B | XL | llama3.2 | amd | 1 | 1.7k |
| unsloth/Llama-3.2-90B-Vision-Instruct-unsloth-bnb-4bit | quantized | -bnb | 88.6B | XL | llama3.2 | unsloth | 2 | 514 |
| osllmai-community/Llama-3.2-90B-Vision-Instruct | finetune | — | — | — | llama3.2 | osllmai-community | 0 | 580 |
| kiddobellamy/Llama_Vision | finetune | — | — | — | mit | kiddobellamy | 1 | 341 |
| gghfez/Llama-3.3-90B-Vision-merged | finetune | — | 88.6B | XL | llama3.3 | gghfez | 5 | 50 |
| itsanurag/Llama-3.2-90B-Vision-Instruct-bnb-4bit | quantized | -bnb | 91.3B | XL | llama3.2 | itsanurag | 0 | 74 |
| haniu/vision | adapter | — | — | — | llama3.2 | haniu | 0 | 70 |
| haniu/textvision | adapter | — | — | — | llama3.2 | haniu | 0 | 56 |
| haniu/text | adapter | — | — | — | llama3.2 | haniu | 0 | 54 |
| SaiChamakura/fine-tuned-visionllama90B | finetune | — | — | — | — | SaiChamakura | 0 | 33 |
Ranked by source signals (HN, Lobsters, authority domains, official posts) and recency. Summaries are generated from a stored copy of the article text; every sentence cites a span of that text, and nothing is shown that failed the citation check.
arXiv links found in the model card and its tags. "Peer-reviewed" is shown only with evidence: a published venue in OpenAlex (measured) or the authors' own journal reference on arXiv (reported).
Hub snapshot: 2026-10-25 · HF API: 2026-10-24 21:40 UTC · Providers: 2026-10-24 21:44 UTC · Articles: 2026-10-24 21:46 UTC · Summary: 2026-10-07 15:12 UTC · Last verified 2026-10-25
Email alerts are not open yet; the feeds above are the same data. See feeds for the item schema.
