Up to date
4.15B parameters, images and text in, text out, 128K tokens of context.
Verified 2026-10-25 Rank #177 in LLMs · score 90.5 method v1 Sources 3 Downloads 30d 516.8k · #163 of 619 tracked
Changed this week: 1 new quantized build on the Hub (14 in all).2026-10-24
Picked by rule from 9 tracked pieces: the lab's own post, the most shared, a field report, the latest.
RetNPhi is an experimental architecture that transforms Phi-3.5 into a byte-level language model incorporating RetNet-inspired mechanisms.
The 2 picks above, plus these 7. 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: 14 quantized builds and 100 Spaces.
Top Spaces running it: MaziyarPanahi/Phi-3.5-Vision ♥ 228 · TIGER-Lab/MEGA-Bench ♥ 51 · jiuhai/Florence-VL-8B ♥ 40 · maxiw/Phi-3.5-vision ♥ 31
Weights only, an estimate: leave room for the KV cache. Method
About 2.5 GB at INT4, 5.0 GB at FP8, 10 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 (2024-08-16), amber triangles are 3 dated articles (each links to it), grey ticks are repo commits.
43 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.
| Benchmark | Value | Setting / notes | Reported by | Date |
|---|---|---|---|---|
| author-reported Art Style | 87.2 | BLINK | model author | 2026-10-07 |
| author-reported Counting | 54.2 | BLINK | model author | 2026-10-07 |
| author-reported Forensic Detection | 92.4 | BLINK | model author | 2026-10-07 |
| author-reported Functional Correspondence | 29.2 | BLINK | model author | 2026-10-07 |
| author-reported IQ Test | 25.3 | BLINK | model author | 2026-10-07 |
| author-reported Jigsaw | 68 | BLINK | model author | 2026-10-07 |
| author-reported Multi-View Reasoning | 54.1 | BLINK | model author | 2026-10-07 |
| author-reported Object Localization | 49.2 | BLINK | model author | 2026-10-07 |
| author-reported Relative Depth | 69.4 | BLINK | model author | 2026-10-07 |
| author-reported Relative Reflectance | 37.3 | BLINK | model author | 2026-10-07 |
| author-reported Semantic Correspondence | 36.7 | BLINK | model author | 2026-10-07 |
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 | Phi3VForCausalLM |
| Model type | phi3_v |
| Dtype | bfloat16 |
| Params by dtype | BF16: 4,146,621,440 |
| Size band | M |
| MoE | no |
| Context length | 131,072 (config.json max_position_embeddings) |
| Languages | multilingual |
| Papers | arXiv:2404.14219 |
| License | MIT (mit) |
| Hub snapshot | 2026-10-25 |
vLLM: supported · Phi3VForCausalLM · 7050ef881ff0835f6890a47db1859a24c65481d6SGLang: not in registry · 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 Phi-3.5-vision-instruct: params × bytes/param × 1.2 overhead. KV cache and activations are extra. How this is computed.
| BF16 (2 bytes/param) | 10 GB · Single GPU (<24 GB) |
|---|---|
| FP8 (1 byte/param) | 5.0 GB · Single GPU (<24 GB) |
| INT4 (0.5 bytes/param) | 2.5 GB · Edge (<4 GB) |
Weights only (params × bytes/param × 1.2 overhead); KV cache and activations extra.
finetune: 23quantized: 14adapter: 6
| Variant | Relation | Quant | Params | Band | License | Author | Likes | Downloads |
|---|---|---|---|---|---|---|---|---|
| TIGER-Lab/VLM2Vec-Full | finetune | — | 4.15B | M | apache-2.0 | TIGER-Lab | 29 | 3M |
| OpenVINO/Phi-3.5-vision-instruct-int8-ov | quantized | int8 | — | — | mit | OpenVINO | 2 | 442.6k |
| OpenVINO/Phi-3.5-vision-instruct-int4-ov | quantized | int4 | — | — | mit | OpenVINO | 0 | 23.5k |
| yifeihu/TB-OCR-preview-0.1 | finetune | — | 4.25B | M | mit | yifeihu | 129 | 7.4k |
| DeepGlint-AI/UniME-Phi3.5-V-4.2B | finetune | — | — | — | mit | DeepGlint-AI | 7 | 7.5k |
| mlc-ai/Phi-3.5-vision-instruct-q4f16_1-MLC | quantized | — | — | — | — | mlc-ai | 3 | 7.3k |
| nm-testing/Phi-3.5-vision-instruct-FP8-dynamic | quantized | -FP8 | 4.25B | M | — | nm-testing | 0 | 6.7k |
| MBZUAI/BiMediX2-4B | finetune | — | — | — | cc-by-nc-sa-4.0 | MBZUAI | 1 | 5.7k |
| microsoft/Phi-Ground | finetune | — | — | — | mit | microsoft | 24 | 3.8k |
| TIGER-Lab/VLM2Vec-LoRA | finetune | — | — | — | apache-2.0 | TIGER-Lab | 12 | 3.7k |
| microsoft/Phi-Ground-Any | finetune | — | — | — | mit | microsoft | 22 | 2.6k |
| mlc-ai/Phi-3.5-vision-instruct-q4f32_1-MLC | quantized | — | — | — | — | mlc-ai | 0 | 2.2k |
| OpenVINO/Phi-3.5-vision-instruct-fp16-ov | finetune | -fp16 | — | — | mit | OpenVINO | 2 | 1.9k |
| onnx-community/Phi-3.5-vision-instruct | quantized | — | — | — | — | onnx-community | 2 | 1.2k |
| muhtasham/Phi-3.5-vision-instruct_20240915_223241 | finetune | — | 4.25B | M | mit | muhtasham | 0 | 543 |
| OPEA/Phi-3.5-vision-instruct-int4-sym-inc | quantized | int4 | — | — | mit | OPEA | 0 | 481 |
| OPEA/Phi-3.5-vision-instruct-qvision-int4-sym-inc | quantized | int4 | — | — | mit | OPEA | 1 | 344 |
| muhtasham/Phi-3.5-vision-instruct_20240916_162052 | finetune | — | 4.25B | M | mit | muhtasham | 0 | 379 |
| nm-testing/Phi-3.5-vision-instruct-W8A8-Dynamic-Per-Token | quantized | -W8A8 | 4.25B | M | — | nm-testing | 0 | 340 |
| EricB/Phi-3.5-vision-instruct-UQFF | quantized | — | — | — | — | EricB | 0 | 315 |
| mlc-ai/Phi-3.5-vision-instruct-q3f16_1-MLC | quantized | — | — | — | — | mlc-ai | 1 | 241 |
| mlc-ai/Phi-3.5-vision-instruct-q0f16-MLC | quantized | — | — | — | — | mlc-ai | 2 | 139 |
| vaibhavmeena/Phi-3.5-vision-instruct-amz-lora | finetune | — | — | — | mit | vaibhavmeena | 0 | 183 |
| helenai/Phi-3.5-vision-instruct-ov-int4 | finetune | int4 | — | — | — | helenai | 0 | 140 |
| HaiZhiYan/CheX-Phi35V | finetune | — | — | — | apache-2.0 | HaiZhiYan | 2 | 0 |
| muhtasham/Phi-3.5-vision-instruct_20240917_025100 | finetune | — | 4.25B | M | mit | muhtasham | 0 | 87 |
| muhtasham/Phi-3.5-vision-instruct_20240916_205629 | finetune | — | 4.25B | M | mit | muhtasham | 0 | 76 |
| helenai/Phi-3.5-vision-instruct-ov-fp16 | finetune | -fp16 | — | — | — | helenai | 0 | 74 |
| Abhisek987/phi35-vision-pdf-markdown | adapter | — | — | — | — | Abhisek987 | 0 | 72 |
| solvrays/mdf-form-reader-phi35-vision | adapter | — | 8.29B | M | apache-2.0 | solvrays | 0 | 66 |
| yydxlv/colphi3.5 | finetune | — | — | — | mit | yydxlv | 1 | 0 |
| Vet-Rate-org/Vet-Rate-Vision-Phi | finetune | — | — | — | mit | Vet-Rate-org | 0 | 30 |
| Erland/phi3.5-vision-ai701 | adapter | — | — | — | mit | Erland | 0 | 27 |
| dixieclick/Phi-3.5-vision-instruct-q4f16_1-MLC | quantized | — | — | — | — | dixieclick | 0 | 27 |
| directorcreator/pvisionmain | finetune | — | — | — | other | directorcreator | 0 | 23 |
| rlawlsgus/phi3.5-vision-crowd-adapter | adapter | — | — | — | — | rlawlsgus | 0 | 14 |
| OpenVINO/Phi-3.5-vision-instruct-nf4-ov | quantized | -nf4 | — | — | mit | OpenVINO | 0 | 7 |
| EVA787797/78787879zerty | adapter | — | — | — | bsl-1.0 | EVA787797 | 0 | 0 |
| tugtoy/HLPAI | finetune | — | — | — | apache-2.0 | tugtoy | 0 | 0 |
| camilo09/Avancemos | finetune | — | — | — | gpl-3.0 | camilo09 | 0 | 0 |
| Date | Commit | SHA |
|---|---|---|
| 2025-12-10 | Update README.md (#43) | 12b77fb |
| 2025-12-10 | Upload data_summary_card.md (#42) | 87713eb |
| 2024-09-26 | fix_rope_scaling (#28) | 4a0d683 |
| 2024-09-07 | Add proper library name (#23) | c69b58c |
| 2024-08-30 | Fix task tag (#16) | c881ad5 |
| 2024-08-23 | Fix spelling error in README (#10) | c68f852 |
| 2024-08-20 | Fix AutoImageProcessor mapping (#4) | ee47981 |
| 2024-08-20 | Fixing Scores for Gemini 1.5 Flash and GPT-4o (#2) | 1c4d695 |
| 2024-08-20 | Update README.md | 0eb7016 |
| 2024-08-20 | Update README.md | bea9545 |
| 2024-08-20 | fix_import (#1) | 64f88b6 |
| 2024-08-20 | upload model card | 7da0ee8 |
| 2024-08-19 | upload initial files | 249c1c2 |
| 2024-08-16 | initial commit | 40a03b9 |
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.
| Title | Source | Published | First seen | Signals |
|---|---|---|---|---|
| Discover the New Multi-Lingual, High-Quality Phi-3.5 SLMs | techcommunity.microsoft.com official | — | 2026-10-07 | exa |
| Phi-3.5-vision-instruct | Model Catalog | Microsoft Foundry | ai.azure.us | — | 2026-10-07 | exa |
| Phi-3.5 Vision Instruct | Open Laboratory | openlaboratory.com | — | 2026-10-07 | exa |
| Phi-3.5-vision-instruct Benchmarks, Pricing & Context Window | llm-stats.com | — | 2026-10-07 | exa |
| Phi 3.5 Vision Instruct – 128k context, multimodal, open source | LLM Reference | llmreference.com | — | 2026-10-07 | exa |
| LLM: Against all odds – The token paths not taken | algo-consult.com | 2025-11-21 | 2026-10-07 | hn |
| Run with Phi 3.5-mini across GPUs, CPUs and OSes | secondstate.io | 2024-09-03 | 2026-10-07 | hn |
| Show HN: RetNPhi – Phi-3.5 as a Byte-Level LM with RetNet-Inspired Efficiency | github.com | 2024-09-02 | 2026-10-07 | hn |
| Microsoft releases powerful new Phi-3.5 models, beating Google, OpenAI and more | venturebeat.com | 2024-08-21 | 2026-10-07 | hn |
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 21:36 UTC · Providers: 2026-10-24 21:44 UTC · Articles: 2026-10-24 21:45 UTC · Summary: 2026-10-07 15:03 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.
