Up to date
4.15B parameters, text in, text out, 128K tokens of context.
Verified 2026-10-25 Rank #413 in LLMs · score 79.1 method v1 Sources 3 Downloads 30d 42.7k · #394 of 619 tracked
Picked by rule from 18 tracked pieces: the lab's own post, the most shared, a field report, the latest.
Phi-3 models achieve high performance rivaling Mixtral 8x7B by training on a scaled-up dataset of heavily filtered publicly available web data and synthetic data.
Microsoft released the Phi family of open-source small language models, which are the most capable and cost-effective SLMs available.
The 2 picks above, plus these 16. 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: 3 quantized builds and 52 Spaces.
Top Spaces running it: ysharma/Microsoft_Phi-3-Vision-128k ♥ 219 · MaziyarPanahi/Phi-3-vision-128k ♥ 47 · dwb2023/hf_extractor ♥ 6 · dwb2023/omniscience ♥ 6
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 (2024-05-19), amber triangles are 10 dated articles (each links to it), grey ticks are repo commits.
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.
Phi-3-vision-128k-instruct is a 4.15B parameter multimodal model from the Phi-3 family designed for text and vision tasks.[1][2] The model supports a 128K context length and is released under an MIT license.[3][4]
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.
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 | text-generation |
|---|---|
| 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 |
| 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-vision-128k-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.
adapter: 7finetune: 5quantized: 3
| Variant | Relation | Quant | Params | Band | License | Author | Likes | Downloads |
|---|---|---|---|---|---|---|---|---|
| NTT-hil-insight/VDocRetriever-Phi3-vision | adapter | — | — | — | — | NTT-hil-insight | 2 | 3.2k |
| NTT-hil-insight/VDocGenerator-Phi3-vision | adapter | — | — | — | — | NTT-hil-insight | 1 | 1.5k |
| MCG-NJU/VideoChatOnline-4B | finetune | — | 4.15B | M | mit | MCG-NJU | 1 | 1.1k |
| RedHatAI/Phi-3-vision-128k-instruct-W4A16-G128 | quantized | -W4A16 | 4.27B | M | apache-2.0 | RedHatAI | 1 | 855 |
| puar-playground/Phi-3-MusiX | adapter | — | — | — | mit | puar-playground | 1 | 465 |
| puar-playground/Col-Phi-3-V | adapter | — | — | — | — | puar-playground | 0 | 480 |
| onnx-community/Phi-3-vision-128k-instruct | quantized | — | — | — | — | onnx-community | 1 | 426 |
| NTT-hil-insight/VDocRetriever-Phi3-vision-pretrained | adapter | — | — | — | — | NTT-hil-insight | 0 | 305 |
| MuyeHuang/EvoChart | finetune | — | — | — | gpl-3.0 | MuyeHuang | 1 | 225 |
| SkyNetdev/Phi3v-Finetuned | finetune | — | 4.25B | M | apache-2.0 | SkyNetdev | 1 | 198 |
| yanghaoir/ReAlign-Phi3v | adapter | — | — | — | mit | yanghaoir | 1 | 132 |
| llmware/phi-3-vision-onnx | quantized | — | — | — | apache-2.0 | llmware | 0 | 127 |
| farshadafx/results | adapter | — | — | — | mit | farshadafx | 0 | 33 |
| aschirru34/results | finetune | — | — | — | — | aschirru34 | 0 | 0 |
| laicsiifes/phi3-vision-flickr30k_pt_human_generated | finetune | — | — | — | mit | laicsiifes | 0 | 0 |
| Date | Commit | SHA |
|---|---|---|
| 2025-12-10 | Update README.md (#73) | ed2772f |
| 2025-12-10 | Upload data_summary_card.md (#72) | 7c760c5 |
| 2024-08-20 | Update README.md | c45209e |
| 2024-08-20 | Update README.md | 5bc05f2 |
| 2024-08-05 | Fix generation when `repetition_penalty` is activated (#57) | 6065b7a |
| 2024-07-30 | Fix number of image tokens (#51) | b161793 |
| 2024-07-17 | Rewrite image embedding to remove the in-place op (#53) | 866d169 |
| 2024-07-17 | Use Flash Attention for CLIP image encoder (#52) | 1a87b10 |
| 2024-07-09 | Update README.md | f998a18 |
| 2024-06-11 | clean_sample_code (#36) | 7b92b8c |
| 2024-05-31 | Make flash attention configurable in user code (#26) | fea3f11 |
| 2024-05-31 | Add link to Phi-3 vision ONNX models | 11cb03c |
| 2024-05-29 | correct the paper reference (#25) | 5dcfdfb |
| 2024-05-24 | Update README.md (#14) | dbcdaaa |
| 2024-05-21 | Update README | 2b07137 |
| 2024-05-21 | Update README | 94f6e2f |
| 2024-05-21 | Update README | 034a3df |
| 2024-05-21 | Update README | d33b637 |
| 2024-05-21 | Update README | 4ed3dc0 |
| 2024-05-20 | chore(root): Initial files upload. | 44c7c8b |
| 2024-05-19 | initial commit | 7ba0150 |
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.
Hub snapshot: 2026-10-25 · HF API: 2026-10-24 21:39 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
Email alerts are not open yet; the feeds above are the same data. See feeds for the item schema.
