The story so far
Weights, every dated article, and repo commits on one rail.
30 repo commits since the weights landed
Teal is the weights (2024-08-05), grey ticks are repo commits.
Idefics3-8B-Llama3
HuggingFaceM4 Apache 2.0 out 2 years ago
8.46B parameters, images and text in, text out, 128K tokens of context.
Verified 2026-10-26 07:27 UTC Rank #581 in LLMs · score 74.4 method v1 Sources 3 Downloads 30d 55k · #374 of 619 tracked
Do people use it?55kdownloads, 30 days#374 of 619 we track · below the median+0 likes, downloads −15% this week307 likesHugging Face2026-10-26
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: 5 quantized builds and 18 Spaces.
- As a service0no API provider listed
- On own hardware5quantized builds
- Built on269 fine-tunes · 17 adapters · 0 merges
- In apps18Hugging Face Spaces
Top Spaces running it: HuggingFaceM4/idefics3 ♥ 102 · TIGER-Lab/MEGA-Bench ♥ 51 · eltorio/IDEFICS3_ROCO ♥ 12 · hexgrad/IDEFICS3_ROCO_ZeroGPU ♥ 7
Will it fit?
Weights only, an estimate: leave room for the KV cache. Method
Fits a 16 GB card at INT4
About 5.1 GB at INT4, 10 GB at FP8, 20 GB at BF16: weights only, so leave room for the KV cache.
Estimate params × bytes per param × 1.2 · how we estimate
What people built on it
31 derivatives on the Hub. Which build to run, by runtime: the most downloaded in each class.
31 derivatives on the Hub, most of them adapters
From the Hugging Face model tree; quantized builds in teal, fine-tunes, merges and adapters in greys.
Reference
Everything else, folded: the full tables, every source.
Benchmarks
5 rows
Benchmarks 5 rows
measured third-party leaderboard rowauthor-reported number from the model card or eval files (5)No confidence intervals unless the source gave them; settings differ between rows, so compare within a benchmark, not across. Date = the benchmark's own date when it gives one, else when we read it.
| Benchmark | Value | Setting / notes | Reported by | Date |
|---|---|---|---|---|
| author-reported DocVQA · Extract text from documents | 87.7 | Idefics3-8B | model author | 2026-10-07 |
| author-reported MMMU · Understand images and video | 46.6 | Idefics3-8B | model author | 2026-10-07 |
| author-reported MMStar · Understand images and video | 55.9 | Idefics3-8B | model author | 2026-10-07 |
| author-reported MathVista · Understand images and video | 58.4 | Idefics3-8B | model author | 2026-10-07 |
| author-reported TextVQA · Understand images and video | 74.9 | Idefics3-8B | model author | 2026-10-07 |
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]
Generated 2026-10-07 15:12 UTC by parasail-gemma-4-26b-a4b-it; each numeral points at the fact it rests on. Not a verdict.
Use-case evidence
5 use cases
What it is good at 5 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.
- Understand images and videoauthor-reportedAuthor reports MMMU <br>(val) 46.6Author reports MathVista <br>(test) 58.4Author reports MMStar <br>(val) 55.9+5 more
- Extract text from documentsauthor-reportedAuthor reports DocVQA <br>(test) 87.7decision-model estimate p=0.96
- Chat and follow instructionsstructuraldecision-model estimate p=0.80tag: conversationalchat template present
- Run on a laptop or phonestructuraldecision-model estimate p=0.78fits: single-gpu
- Read long documentsstructuralconfig: 131,072 context
Facts
21 fields
Listed in LLMs
Facts
- Idefics3-8B-Llama3 has 8.46B parameters. (safetensors metadata, observed 2026-10-26)
- Context length is 131,072 tokens (128K). (config.json max_position_embeddings, observed 2026-10-26)
- License: Apache 2.0 (friction: none). Permissive; patent grant. (license file, observed 2026-10-26)
- Created 2024-08-05, last modified 2024-12-02. (HF API, observed 2026-10-26)
| Pipeline tag | image-text-to-text |
|---|---|
| Library | transformers |
| Architectures | Idefics3ForConditionalGeneration |
| Model type | idefics3 |
| Dtype | bfloat16 |
| Params by dtype | BF16: 8,462,086,384 |
| Size band | M |
| MoE | no |
| Context length | 131,072 (config.json max_position_embeddings) |
| Languages | en |
| Papers | arXiv:2306.16527, arXiv:2405.02246, arXiv:2408.12637 |
| License | Apache 2.0 (apache-2.0) |
| Hub snapshot | 2026-10-26 |
Servability
vLLM · SGLang
Servability
vLLM: supported · Idefics3ForConditionalGeneration · 4b1c04e76fe9ac3ca8b7e541dd423e7d5f38dc60SGLang: not in registry · 60d941293e3af27a12c6eb64e2c3e2e1a17f100f
vLLM registry (4b1c04e76fe9ac3ca8b7e541dd423e7d5f38dc60) · SGLang registry (60d941293e3af27a12c6eb64e2c3e2e1a17f100f) · 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 Idefics3-8B-Llama3: 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) | 5.1 GB · Single GPU (<24 GB) |
Weights only (params × bytes/param × 1.2 overhead); KV cache and activations extra.
All variants
31 variants
Lineage
finetune: 9adapter: 17quantized: 5
Top variants (31 shown, ranked by downloads)
| Variant | Relation | Quant | Params | Band | License | Author | Likes | Downloads |
|---|---|---|---|---|---|---|---|---|
| Minthy/ToriiGate-v0.3 | finetune | — | 8.46B | M | apache-2.0 | Minthy | 24 | 7.6k |
| eltorio/IDEFICS3_ROCO | adapter | — | — | — | apache-2.0 | eltorio | 9 | 3.8k |
| nectec/Pathumma-llm-vision-1.0.0 | finetune | — | 8.46B | M | — | nectec | 11 | 1.7k |
| jesse9527/Idefics3-8B-Llama3-Q4_K_M-GGUF | quantized | GGUF | — | — | apache-2.0 | jesse9527 | 2 | 766 |
| leon-se/Idefics3-8B-Llama3-bnb_nf4 | quantized | -bnb | 8.68B | M | — | leon-se | 0 | 645 |
| Minthy/ToriiGate-v0.2 | finetune | — | 8.46B | M | apache-2.0 | Minthy | 3 | 401 |
| eltorio/IDEFICS3_ROCOv2 | adapter | — | — | — | apache-2.0 | eltorio | 4 | 310 |
| Mantis-VL/mantis-8b-idefics3_16384 | finetune | — | 8.46B | M | apache-2.0 | Mantis-VL | 0 | 354 |
| 2dameneko/Idefics3-8B-Llama3-nf4 | quantized | -nf4 | 8.69B | M | apache-2.0 | 2dameneko | 0 | 338 |
| Mantis-VL/mantis-8b-idefics3-pure_16384 | finetune | — | 8.46B | M | apache-2.0 | Mantis-VL | 0 | 256 |
| eltorio/IDEFICS3_medical_instruct | adapter | — | — | — | apache-2.0 | eltorio | 2 | 137 |
| Minthy/Torii_Gate_v0.1_alpha | finetune | — | 8.46B | M | apache-2.0 | Minthy | 0 | 207 |
| leon-se/Idefics3-8B-Llama3-FP8-Dynamic | quantized | -FP8 | 8.46B | M | — | leon-se | 0 | 203 |
| ronantakizawa/idefics3-8b-llama3-awq | quantized | awq | 2.41B | M | apache-2.0 | ronantakizawa | 1 | 145 |
| Maverick17/idefics3-llama-gui-dense-descriptions | adapter | — | — | — | apache-2.0 | Maverick17 | 0 | 133 |
| justinkarlin/idefics3-qlora-faces | adapter | — | — | — | apache-2.0 | justinkarlin | 0 | 127 |
| fsommers/idefics3-llama-vqav2_1 | adapter | — | — | — | apache-2.0 | fsommers | 0 | 106 |
| Oysiyl/Idefics3-8B-Llama3-ufo | adapter | — | — | — | apache-2.0 | Oysiyl | 0 | 34 |
| justinkarlin/idefics3-qlora-faces6 | adapter | — | — | — | apache-2.0 | justinkarlin | 0 | 31 |
| Tchalla12/idefics3-8b-qlora | adapter | — | — | — | — | Tchalla12 | 0 | 28 |
| justinkarlin/idefics3-llama-vqav2 | adapter | — | — | — | apache-2.0 | justinkarlin | 0 | 27 |
| justinkarlin/idefics3-qlora-faces4 | adapter | — | — | — | apache-2.0 | justinkarlin | 0 | 27 |
| justinkarlin/idefics3-qlora-faces2 | adapter | — | — | — | apache-2.0 | justinkarlin | 0 | 23 |
| justinkarlin/idefics3-qlora-faces3 | adapter | — | — | — | apache-2.0 | justinkarlin | 0 | 23 |
| Erland/idefics3-llama-ai701 | adapter | — | — | — | apache-2.0 | Erland | 0 | 23 |
| justinkarlin/idefics3-qlora-faces5 | adapter | — | — | — | apache-2.0 | justinkarlin | 0 | 21 |
| joris-sense/idefics3-llama-vqav2 | adapter | — | — | — | apache-2.0 | joris-sense | 0 | 20 |
| kurama270296/qlora-idefics3-test | adapter | — | — | — | — | kurama270296 | 0 | 20 |
| Clark12/POS1 | finetune | — | — | — | mit | Clark12 | 0 | 0 |
| slezki/assistant_mage | finetune | — | — | — | mit | slezki | 0 | 0 |
| CATIE-AQ/Idefics3_FT_fr | finetune | — | — | — | apache-2.0 | CATIE-AQ | 0 | 0 |
Repository changelog
30 commits
Repository changelog 30 commits
| Date | Commit | SHA |
|---|---|---|
| 2024-12-02 | Update README.md | fddb4ff |
| 2024-12-02 | Update README.md | e493ce0 |
| 2024-09-18 | Add proper library tag (#16) | ebde0d8 |
| 2024-09-18 | Add EOS token id (#18) | 5388aec |
| 2024-08-26 | Update README.md | 605edaf |
| 2024-08-26 | Update README.md | a1b83a3 |
| 2024-08-15 | Update preprocessor_config.json | 30d4b60 |
| 2024-08-12 | Update README.md | 54b2639 |
| 2024-08-12 | Update README.md | 9af05b7 |
| 2024-08-09 | remove hidden size from the main config | c9ba4e1 |
| 2024-08-08 | Update config.json | d8435d5 |
| 2024-08-06 | Update README.md | b5906c5 |
| 2024-08-06 | Update README.md | 86e519e |
| 2024-08-06 | Update README.md | 657234f |
| 2024-08-06 | Update README.md | 8ed28d9 |
| 2024-08-06 | Update README.md | 2177c2f |
| 2024-08-06 | Update README.md | 43f63a3 |
| 2024-08-06 | Update tokenizer_config.json | 26d25e9 |
| 2024-08-06 | Update preprocessor_config.json | c77a21d |
| 2024-08-06 | Delete processor_config.json | c510f0c |
| 2024-08-06 | Create chat_template.json | 92a708a |
| 2024-08-06 | Create processor_config.json | c6bf80e |
| 2024-08-06 | Update preprocessor_config.json | 9cb63ce |
| 2024-08-06 | Update tokenizer_config.json | dbb5ae3 |
| 2024-08-06 | Update tokenizer_config.json | 6f16a06 |
| 2024-08-06 | Update tokenizer_config.json | d3dbbde |
| 2024-08-06 | Update tokenizer_config.json | b08cc74 |
| 2024-08-05 | fix bug in config defaults | 7f51046 |
| 2024-08-05 | Update README.md | 15733c2 |
| 2024-08-05 | Update README.md | ca8ae61 |
Papers
3 papers
Papers 3
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: 2026-10-26 · HF API: 2026-10-26 07:22 UTC · Providers: 2026-10-26 07:26 UTC · Articles: 2026-10-26 07:27 UTC · Summary: 2026-10-07 15:12 UTC · Last verified 2026-10-26 07:27 UTC
Follow Idefics3-8B-Llama3
Watch this model
- /feeds/models/HuggingFaceM4--Idefics3-8B-Llama3.xml(RSS 2.0) · every change to Idefics3-8B-Llama3
- /feeds/families/HuggingFaceM4--Idefics3.xml(RSS 2.0) · every change across the Idefics3 family
Email alerts are not open yet; the feeds above are the same data. See feeds for the item schema.