Model Follower.

Up to date

Rebuilt
2026-10-25 06:54 UTC
Release radar
117 on watch
Hub snapshot
2026-10-25
Re-checked
daily · every number sourced and dated

Models › idefics2 family

idefics2-8b

Hugging Face ↗ Follow changes

HuggingFaceM4 Apache 2.0 out 3 years ago

8.40B parameters, images and text in, text out, 32K tokens of context.

Verified 2026-10-25 Rank #504 in LLMs · score 76.3 method v1 Sources 3 Downloads 30d 11.9k · #497 of 619 tracked

Can I run it?5.0GB at INT4Estimatefits a 16 GB cardmethod
Can I use it?Apache 2.0Permissive; patent grantlicence file2026-10-24
Do people use it?11.9kdownloads, 30 days#497 of 619 we track · below the median+0 likes, downloads −49% this week627 likesHugging Face2026-10-24

Worth reading

Picked by rule from 1 tracked piece: the lab's own post, the most shared, a field report, the latest.

Latest · 2 years ago

Idefics2: A Powerful 8B Vision-Language Model for the Community

Idefics2 is an 8B parameter multimodal model released under an Apache 2.0 license that supports native image resolutions up to 980 x 980.

  • Idefics2 departs from Idefics1's gated cross-attentions by using a learned Perceiver pooling and an MLP modality projection.
  • The model allows optional sub-image splitting in 4 following the strategy from SPHINX and LLaVa-NeXT.
huggingface.co 2024-05-14 6 HN pts 1.1k words about the idefics2 family family

Where it's used

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: 1 quantized builds and 100 Spaces.

  • As a service0no API provider listed
  • On own hardware1quantized builds
  • Built on192167 fine-tunes · 25 adapters · 0 merges
  • In apps100+Hugging Face Spaces

Top Spaces running it: HuggingFaceM4/idefics2_playground ♥ 170 · HuggingFaceM4/idefics-8b ♥ 146 · thobuiq/GPT-4o ♥ 44 · EPFL-VILAB/ViPer ♥ 13

Will it fit?

Weights only, an estimate: leave room for the KV cache. Method

Fits a 16 GB card at INT4

About 5.0 GB at INT4, 10 GB at FP8, 20 GB at BF16: weights only, so leave room for the KV cache.

16 GB card64 GB Mac2×80 GB8×80 GBINT4: about 5.0 GB of weights (estimate; KV cache and activations extra)INT45.0 GBFP8: about 10 GB of weights (estimate; KV cache and activations extra)FP810 GBBF16: about 20 GB of weights (estimate; KV cache and activations extra)BF1620 GB
16 GB card64 GB Mac2×80 GB8×80 GBINT4: about 5.0 GB of weights (estimate; KV cache and activations extra)INT45.0 GBFP8: about 10 GB of weights (estimate; KV cache and activations extra)FP810 GBBF16: about 20 GB of weights (estimate; KV cache and activations extra)BF1620 GB

Estimate params × bytes per param × 1.2 · how we estimate

The story so far

Weights, every dated article, and repo commits on one rail.

1 dated article and 24 repo commits so far

Teal is the weights (2024-04-09), amber triangles are 1 dated article (each links to it), grey ticks are repo commits.

MayJunJulAugSepOctNovDecJanFebMarAprMayJunJulAugSepOctNovDecJanFebMarAprMayJunJulAugSepOct2024-10-14: 1 repo commit2024-07-30: 1 repo commit2024-07-28: 1 repo commit2024-05-30: 1 repo commit2024-05-06: 2 repo commits2024-04-29: 2 repo commits2024-04-19: 1 repo commit2024-04-18: 1 repo commit2024-04-17: 1 repo commit2024-04-16: 2 repo commits2024-04-15: 4 repo commits2024-04-14: 2 repo commits2024-04-13: 1 repo commit2024-04-12: 1 repo commit2024-04-10: 2 repo commits2024-04-09: 1 repo commitIdefics2: A Powerful 8B Vision-Language Model for the Community · 2024-05-14Weights 2024-04-09 · 2024-04-09Weights 2024-04-09now
MayJunJulAugSepOctNovDecJanFebMarAprMayJunJulAugSepOctNovDecJanFebMarAprMayJunJulAugSepOct2024-10-14: 1 repo commit2024-07-30: 1 repo commit2024-07-28: 1 repo commit2024-05-30: 1 repo commit2024-05-06: 2 repo commits2024-04-29: 2 repo commits2024-04-19: 1 repo commit2024-04-18: 1 repo commit2024-04-17: 1 repo commit2024-04-16: 2 repo commits2024-04-15: 4 repo commits2024-04-14: 2 repo commits2024-04-13: 1 repo commit2024-04-12: 1 repo commit2024-04-10: 2 repo commits2024-04-09: 1 repo commitIdefics2: A Powerful 8B Vision-Language Model for the Community · 2024-05-14Weights 2024-04-09 · 2024-04-09Weights 2024-04-09now

What people built on it

193 derivatives on the Hub. Which build to run, by runtime: the most downloaded in each class.

193 derivatives on the Hub, most of them fine-tunes

From the Hugging Face model tree; quantized builds in teal, fine-tunes, merges and adapters in greys.

1 quantized (Hugging Face model tree)167 fine-tunes (Hugging Face model tree)25 adapters (Hugging Face model tree)1 quantized167 fine-tunes25 adapters
1 quantized (Hugging Face model tree)167 fine-tunes (Hugging Face model tree)25 adapters (Hugging Face model tree)1 quantized167 fine-tunes25 adapters

All variants →

Reference

Everything else, folded: the full tables, every source.

What the card says

1 cited sentence

Summary generated, cited

The model accepts arbitrary sequences of image and text inputs to produce text outputs.[1]

  1. “accepts arbitrary sequences of image and text inputs and …”

Generated 2026-10-07 15:09 UTC by parasail-gemma-4-26b-a4b-it; each numeral points at the fact it rests on. Not a verdict.

Use-case evidence

4 use cases

What it is good at 4 use cases

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.

Facts

21 fields

Listed in LLMs

Facts

Pipeline tagimage-text-to-text
Librarytransformers
ArchitecturesIdefics2ForConditionalGeneration
Model typeidefics2
Dtypefloat32
Params by dtypeF32: 8,402,768,112
Size bandM
MoEno
Context length32,768 (config.json max_position_embeddings)
Languagesen
PapersarXiv:2306.16527, arXiv:2405.02246, arXiv:2307.06304, arXiv:2311.07575, arXiv:2103.03206
LicenseApache 2.0 (apache-2.0)
Hub snapshot2026-10-25

Servability

vLLM · SGLang

Servability

vLLM: not in registry · 7050ef881ff0835f6890a47db1859a24c65481d6SGLang: supported · Idefics2ForConditionalGeneration · 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.

Hardware fit

3 estimates

Hardware fit ESTIMATE

Weights-only memory for idefics2-8b: params × bytes/param × 1.2 overhead. KV cache and activations are extra. How this is computed.

BF16 (2 bytes/param)20 GB · Single GPU (<24 GB)
FP8 (1 byte/param)10 GB · Single GPU (<24 GB)
INT4 (0.5 bytes/param)5.0 GB · Single GPU (<24 GB)

Weights only (params × bytes/param × 1.2 overhead); KV cache and activations extra.

All variants

40 variants

Lineage

Original model Family: HuggingFaceM4/idefics2 193 derivatives on the Hub

finetune: 167adapter: 25quantized: 1

Top variants (40 shown, ranked by downloads)

Relation
40 rows
Variant Relation Quant Params Band License Author Likes Downloads
TIGER-Lab/Mantis-8B-Idefics2 finetune — 8.40B M apache-2.0 TIGER-Lab 14 23k
Mantis-VL/mantis-8b-idefics2_8192 finetune — 8.40B M apache-2.0 Mantis-VL 0 776
cmarkea/idefics2-8b-ft-docvqa-lora adapter — — — apache-2.0 cmarkea 2 248
VictorSanh/test-victor adapter — — — apache-2.0 VictorSanh 0 342
zesquirrelnator/idefics2-8b-docvqa-finetuned-tutorial finetune — — — apache-2.0 zesquirrelnator 0 259
Mantis-VL/idefics2-8b-video-eval-refined-40k_4096_generation finetune — 8.40B M apache-2.0 Mantis-VL 0 210
Mantis-VL/mantis-8b-idefics2-video-eval_6144_regression finetune — 8.27B M apache-2.0 Mantis-VL 0 204
edbeeching/vsft-idefics2 finetune — 8.40B M apache-2.0 edbeeching 0 190
Mantis-VL/mantis-8b-idefics2-video-eval_5184_regression finetune — 8.27B M apache-2.0 Mantis-VL 0 173
Mantis-VL/mantis-8b-idefics2-video-eval-20k_2048 finetune — 8.40B M apache-2.0 Mantis-VL 0 171
Mantis-VL/mantis-8b-idefics2-video-eval-95k_4096 finetune — 8.40B M apache-2.0 Mantis-VL 0 169
mqliu/mantis-8b-idefics2_1024 finetune — 8.40B M apache-2.0 mqliu 0 167
Mantis-VL/mantis-8b-idefics2-video-eval-50k_4096 finetune — 8.40B M apache-2.0 Mantis-VL 0 155
Mantis-VL/idefics2-8b-video-eval-refined-40k_4096_regression finetune — 8.27B M apache-2.0 Mantis-VL 0 152
pallavibiswas/idefics2-finetuned-re-id finetune — — — apache-2.0 pallavibiswas 0 144
andrew-together/idefics2-8b-finetune-combined-160k_8192 finetune — 8.40B M apache-2.0 andrew-together 0 141
Mantis-VL/mantis-8b-idefics2-video-eval-95k-2epoch_4096 finetune — 8.40B M apache-2.0 Mantis-VL 0 133
Mantis-VL/mantis-8b-idefics2-video-eval-50k-2epoch_4096 finetune — 8.40B M apache-2.0 Mantis-VL 0 130
Taehyun34/sftwithanswer adapter — — — apache-2.0 Taehyun34 0 113
DatBui/idefics2-8b-finetuned-multimodal12 finetune — — — apache-2.0 DatBui 0 108
vctmk/mantis-8b-idefics2-classification-tedEDself_8g_4096_regression finetune — 8.27B M apache-2.0 vctmk 0 97
vctmk/mantis-8b-idefics2-classification-tedED-8g_4096_regression finetune — 8.27B M apache-2.0 vctmk 0 84
vctmk/mantis-8b-idefics2-classification-example_2048_regression finetune — 8.27B M apache-2.0 vctmk 0 78
vctmk/mantis-8b-idefics2-classification-tedEDself_v2_3_8g_4096_regression finetune — 8.27B M apache-2.0 vctmk 0 72
Bharanidharan07/idefics_2_finetuned_copy quantized — 8.40B M apache-2.0 Bharanidharan07 0 69
Pakawat-Phasook/Ailone_8B_GOD finetune — — — apache-2.0 Pakawat-Phasook 1 0
de-Rodrigo/idefics2-merit finetune — — — mit de-Rodrigo 1 0
kapsb2171/IDEFICS2-DocVQA-fine-tuned123 finetune — — — apache-2.0 kapsb2171 1 0
andrew-together/idefics2-8b-finetune-combined-50k_8192 finetune — 8.40B M apache-2.0 andrew-together 0 46
andrew-together/idefics2-8b-finetune-combined-10k_8192 finetune — 8.40B M apache-2.0 andrew-together 0 43
Mantis-VL/mantis-8b-idefics2-classification-example_4096_regression finetune — 8.27B M apache-2.0 Mantis-VL 0 42
Mantis-VL/mantis-8b-idefics2-video-eval-debug_4096_regression finetune — 8.27B M apache-2.0 Mantis-VL 0 41
Taehyun34/output_gptreasoning adapter — — — apache-2.0 Taehyun34 0 41
dawoz/idefics2-frozenlake adapter — — — apache-2.0 dawoz 0 40
dyliu/vist_ft adapter — — — apache-2.0 dyliu 0 37
morizon/dpo_idefics_rlaif-v-50 adapter — — — apache-2.0 morizon 0 37
Bharanidharan07/Idefics_2_8b_OG finetune — 8.40B M apache-2.0 Bharanidharan07 0 35
matbee/test adapter — — — apache-2.0 matbee 0 33
andrew-together/idefics2-8b-finetune-combined-5k_8192 finetune — 8.40B M apache-2.0 andrew-together 0 33
debenoist/test_push adapter — — — — debenoist 0 32

Repository changelog

24 commits

Repository changelog 24 commits

DateCommitSHA
2024-10-14 Format modelcard better (#75) 2c42686
2024-07-30 Update README.md a70c0c0
2024-07-28 Update README.md 633869c
2024-05-30 Update README.md 317fdf2
2024-05-06 tgi 2c031da
2024-05-06 update readme 87d5d88
2024-04-29 Add chat template to processor_config.json (#36) 8e6111a
2024-04-29 Update README.md (#35) da7999c
2024-04-19 gpu memory / inference speed - tradeoffs of quantization 1fa1cbd
2024-04-18 Update README.md a44e7e3
2024-04-17 being explicit with transformers version bc7e459
2024-04-16 Update README.md 6e13295
2024-04-16 Comment on image splitting (#4) 2c25217
2024-04-15 typo 59e3081
2024-04-15 Update README.md (#1) 12d27ed
2024-04-15 Update README.md 84bc60f
2024-04-15 readme 335ef98
2024-04-14 Update generation_config.json dc2d28f
2024-04-14 Upload folder using huggingface_hub bc364a7
2024-04-13 This commit corrects the mistake 3392312
2024-04-12 add flowchart 036a316
2024-04-10 fix 8a0f3fa
2024-04-10 idefics2-8b 44328af
2024-04-09 initial commit c828c05

Articles and coverage

1 piece

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.

Official1

Idefics2: A Powerful 8B Vision-Language Model for the Community
huggingface.co2024-05-14officialauthorityabout the idefics2 familyHN 6 pts · 0 comments1,069 words
Idefics2 is an 8B parameter multimodal model released under an Apache 2.0 license that supports native image resolutions up to 980 x 980. It processes 64 tokens per image without splitting or 320 tokens with optional sub-image splitting and competes with larger models on benchmarks like MMMU and MathVista. The model was instruction fine-tuned on The Cauldron, an open compilation of 50 manually-curated datasets formatted for multi-turn conversations.written, reviewed, cited
  • Idefics2 departs from Idefics1's gated cross-attentions by using a learned Perceiver pooling and an MLP modality projection.
  • The model allows optional sub-image splitting in 4 following the strategy from SPHINX and LLaVa-NeXT.
  • Idefics2 is built on top of Mistral-7B-v0.1 and siglip-so400m-patch14-384, both released under Apache-2.0 license.

Papers

5 papers

Papers 5

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).

Similar models

3 models

Similar models

  • Qwen3-VL-8B-Instruct · 8.77B
    same size band (M), same task (image-text-to-text), more providers, permissive license
  • Qwen2.5-VL-7B-Instruct · 8.29B
    same size band (M), same task (image-text-to-text), more providers, permissive license
  • Qwen3.5-9B · 9.65B
    same task (image-text-to-text), more providers, permissive license

Sources

3 sources

Sources

Hub snapshot: 2026-10-25 · HF API: 2026-10-24 21:38 UTC · Providers: 2026-10-24 21:44 UTC · Articles: 2026-10-24 21:46 UTC · Summary: 2026-10-07 15:09 UTC · Last verified 2026-10-25

Follow idefics2-8b

Watch this model

Email alerts are not open yet; the feeds above are the same data. See feeds for the item schema.