TIPSv2 (fastforks)

TIPSv2 (Google DeepMind, CVPR 2026) image encoders: ViT-B / L at patch 14 from text-image pretraining with patch-text alignment; dense patch tokens and two global tokens at any size (native 448 px).

This repo is the model card only: it holds no weights. fastforks downloads the authors' weights from the links below and checks them against the SHA-256 in the model card. Code: https://github.com/kleinicke/fastforks. Kernels for the kernels backend: https://huggingface.co/fastforks/kernels. Original code: https://github.com/google-deepmind/tips at bf10a73.

Author weight repos: https://huggingface.co/google/tipsv2-b14, https://huggingface.co/google/tipsv2-l14.

Task: features. Tier: experimental. Outputs: last_hidden_state, pooler_output, hidden_states. Backends: portable, kernels.

Input: one RGB image. Output: dense patch features and a pooled image feature.

Variants

Weights are the authors' files (not hosted here). SHA-256 shortened to 12 hex digits; the full hashes are in card.toml.

variant weights (author link) SHA-256 weights licence commercial use
tipsv2-b14@1 tipsv2-b14/model.safetensors c19aec1a777c Apache-2.0 yes
tipsv2-l14@1 tipsv2-l14/model.safetensors c75707eabca6 Apache-2.0 yes

Licence

Original-code strings, recorded as they are (research-use package; check them before any other use). Code: Apache-2.0. Licence files: LICENSE.

Training data: image-text pairs; the dataset is not named in the repository README or on the model cards at the pinned revisions.

Licence evidence: https://huggingface.co/google/tipsv2-b14/blob/ed1e4dc6b74bf3935ae099e9d5eb30fa96528454/README.md (license: apache-2.0); code: https://github.com/google-deepmind/tips/blob/bf10a73765b048edfbfcdf1f7e6ca8292542baae/LICENSE

Input size

any H x W >= 1, resized or padded inside the model Tested up to (portable, fp16, RTX 5060 Ti 16 GB): tipsv2-b14 3000x4000 (1.4 GiB peak allocated, 2.2 reserved); tipsv2-l14 3000x4000 (2.2 GiB peak allocated, 2.2 reserved)

Speed and memory

Features, synthetic, 224x224

model time (ms): original fp32 original portable kernels kernels speed-up: vs original vs portable peak memory (MiB, allocated / reserved): original portable kernels
tipsv2-b14 7.03 7.21 2.93 1.49 4.7x 2.0x 529 / 548 204 / 224 206 / 226
tipsv2-l14 21.9 14.2 6.15 4.58 3.1x 1.3x 1776 / 1786 621 / 624 624 / 648

Startup and first call

model load (s): original portable kernels first call (s): original portable kernels
tipsv2-b14 0.90 0.20 1.42 0.07 0.18 0.09
tipsv2-l14 3.10 2.91 1.44 0.02 0.08 0.11

torch.compile and TensorRT

model input time: portable compile TensorRT fp16 kernels preparation: compile (cold first call) ONNX + engine build peak MiB (alloc / reserved): portable compile kernels TensorRT (engine + torch)
tipsv2-b14 224x224 2.93 ms 2.18 ms 1.33 ms 1.49 ms 9.18 s 41.5 s 204 / 224 239 / 266 206 / 226 398 / 422
tipsv2-l14 224x224 6.15 ms 6.04 ms 4.28 ms 4.58 ms 12.4 s 47.9 s 621 / 624 659 / 666 624 / 648 1236 / 1264

Fixed-size cells: time (ms) / peak allocated / peak reserved memory (MiB).

Hardware, protocol and caveats: benchmarks (RTX 5060 Ti 16 GB, sm_120). Original = the original code named above at the precision of portable (fp16, or (fp32) where fp16 is unsupported or not finite); original fp32 = its default fp32 run. Peak memory is the torch allocator's.

Accuracy

On public datasets

Two disjoint seeded subsets of the val folders; each image shorter side 256 (bicubic), centre crop 224, the CLS / pooled feature; cosine kNN, vote weighted by similarity. Original = the original code named under Details, same checkpoint and protocol.

variant dataset original fp32 original fp16 portable kernels note
tipsv2-b14 ImageNet-1k val, kNN top-1 % (k = 20, 10000 labelled / 5000 test images) 59.34 59.30 59.36 59.16
tipsv2-l14 ImageNet-1k val, kNN top-1 % (k = 20, 10000 labelled / 5000 test images) 63.48 63.44 63.42 63.42

Measured on NVIDIA GeForce RTX 5060 Ti, torch 2.14.1+cu130, fastforks 8b2f2df5, 2026-11-10.

Parity with the original code

Every variant is checked against reference outputs of the original code at its pinned commit. CPU, float32: max |diff| <= 6.9e-05 vs the original. GPU, float16 (the default), portable and kernels: mean |diff| <= 1.9 % and p99 <= 6.9 % of the reference's median |value| (absolute error of the output last_hidden_state).

Limitations

  • Image encoder only: the text encoder in the checkpoint is not loaded (the features task has no text input), although the file is downloaded whole.
  • last_hidden_state holds 2 prefix tokens, then the patches: upstream's first cls token (pooler_output, the global image embedding aligned with text) and its second cls token (index 1, the spatially aware global embedding).
  • kernels replays a CUDA graph per small input shape without hidden states (up to about 520k activations: ViT-B ~360 px, ViT-L ~310 px), captured on the shape's second call, at most 4 per model; model.release_graphs() frees them.

Usage

from fastforks import features

model = features.load("tipsv2-b14")  # mode="auto": the preferred backend from recorded measurements
result = model("photo.jpg", size=518)
result.last_hidden_state  # tokens [N, C], as Hugging Face's output
result.backend  # what ran, e.g. Backend(mode='portable', device='cuda:0', dtype='float16')

Install with pip install fastforks (see https://github.com/kleinicke/fastforks). Other variants: tipsv2-b14, tipsv2-l14.

Kernels backend

features.load("tipsv2-b14", mode="kernels") runs our CUDA kernels (fastbb, dpt_fused). It needs a prebuilt kernel for your GPU architecture (downloaded from https://huggingface.co/fastforks/kernels on first use) and a recent NVIDIA driver; if there is none it raises BackendUnavailableError with the reason and never falls back silently. mode="auto" picks kernels when it can run and logs the choice.

Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Model tree for fastforks/tipsv2

Finetuned
(6)
this model

Collection including fastforks/tipsv2