proomt

Search

Search posts, papers, and topics

All posts

Hugging Face Daily PapersGuangyu Sun, Shlok Kumar Mishra, Wentao Bao1 min readpaperadvanced

FLAT: Resampling Image and Text into 1D Flexible-Length Aligned Transmodal Tokens for Retrieval and Generation

Summary

FLAT is a pre‑training framework that encodes images and text into a shared 1D token sequence with variable length via nested dropout, enabling the same embeddings for cross‑modal retrieval and generation. It attains state‑of‑the‑art scores on COCO and Flickr30K for captioning, retrieval, and text‑to‑image generation, and supports interpolation and zero‑shot composed retrieval.

  • FLAT maps visual and textual inputs into a unified 1D token space, using nested dropout to allow flexible prefix lengths.
  • Joint training combines contrastive alignment with bidirectional generative decoders, so embeddings serve both discriminative and generative purposes.
  • Achieves competitive results: 71.1 GenEval (pre‑train) and 83.1 (fine‑tuned) on T2I, 40.5 BLEU‑4 & 138.6 CIDEr on COCO captioning, Recall@5 86.8/75.8 (I2T/T2I) on COCO.
  • Representations support linear interpolation, latent space arithmetic, and zero‑shot composed retrieval.

Engineers building multimodal systems will care because FLAT unifies retrieval and generation in one flexible representation, simplifying pipelines and improving performance.

7/10

Related reading

  1. tokenizers v1: encode, decode and scaling, measured

    Hugging Face has released `tokenizers` v1, a major performance update that achieves 3-30x faster encoding than v0.23 while maintaining identical output and API compatibility. Key optimizations include a SIMD-accelerated splitter, a thread-local word cache, and an allocation-free BPE merge loop, ensuring tokenization doesn't bottleneck ML workflows.

    Hugging Facehuggingface.co10 min
  2. DeepSeek-V4.1-Flash: Pushing the Limits of KV Cache Compression

    DeepSeek‑V4.1‑Flash is a 552B‑parameter multimodal Mixture‑of‑Experts LLM that supports up to 1 M‑token contexts while slashing KV‑cache memory to 890 bytes/token (≈¼ of its predecessor) via cross‑layer reuse (CSA2) and FP4 quantisation, plus a SWA‑Bounded Replay scheme that cuts persistent cache to 1/8. The Causal Encoder‑Decoder design halves prefill compute (8B vs 16B active parameters) and th…

    Hugging Face Daily Papersarxiv.org3 minpaperHN12710