Tag: #multimodal

한국어English日本語中文

← All posts

• AI for Everyone, Part 4 — Captioning Images With 1.37M Parameters, and Why Part 3 Bug Was Absent Here

  We joined a CNN encoder to a transformer decoder and captioned Fashion-MNIST images. With 1.37 million parameters and ten minutes of training, the label hit rate reached 91%. This part looks at what cross-attention does

  2026-08-22 · 7 min read #ai#captioning#multimodal#transformer#pytorch

• AI for Everyone, Part 3 — Loss of 0.0017, Accuracy of 7.5%: The Culprit Was One Padding Slot

  We built a VQA model — one that answers questions about an image — with 1.48 million parameters. Training loss fell to 0.0017 while accuracy sat at 7.5%, worse than guessing. The cause was not the model but a single line

  2026-08-21 · 8 min read #ai#vqa#multimodal#debugging#pytorch

• Vision and Multimodal: Image Understanding, OCR, and VLMs

  Vision models are not one category. Conversational VLMs, dedicated OCR, document structuring, and image retrieval embeddings are different objects and do not substitute for one another. This post organizes the card value

  2026-08-12 · 7 min read #ai#huggingface#open-source-llm#vision-language-model#ocr

• Image Generation and Understanding Technical Reports: What to Read, and Why Design Beats Sample Images

  Eleven image generation and understanding technical reports, each verified by opening the arXiv abstract page directly. Rectified flow transformers, VAR, Emu3, SANA, Janus-Pro, FLUX.1 Kontext, the Qwen-Image line, Seedre

  2026-08-12 · 8 min read #ai-papers#paper-review#technical-report#image-generation#diffusion-transformer

• How Text, Images, and Agents Are Each Measured — Why the Three Domains Measure Fundamentally Different Things

  Text, images, and agents all use the word "performance," but their measurement structures are entirely different. Text splits into multiple-choice that pretends to have a correct answer and open-ended generation that has

  2026-08-02 · 21 min read #llm-evaluation#multimodal#llm-as-judge#agent-benchmark#metrics

• Training Vision LLMs — How to Teach Input and Output

  A vision-language model is trained in stages, from alignment pretraining to instruction fine-tuning. We organize what gets taught and how, from the angle of the training pipeline: vision encoder freezing strategy, data c

  2026-06-26 · 17 min read #mlops#vision-language-model#multimodal#training#instruction-tuning

• Serving Multimodal LLMs — The New Challenges Image Input Creates

  From how multimodal LLM serving differs from text-only serving, to the added vision-encoder stage, variable visual token counts, prefill cost spikes, the difficulty of multimodal KV cache and batching, latency decomposit

  2026-06-26 · 14 min read #mlops#multimodal#llm-serving#vllm#kv-cache

• Vision LLM Architecture — How an Image Becomes Language

  A vision-language model processes an image with a vision encoder, then passes it through a projector to produce tokens an LLM can read. From patch embedding to arbitrary-resolution handling, we trace the full path by whi

  2026-06-26 · 20 min read #llm#vision-language-model#multimodal#vit#qwen2-vl

• Multimodal Tokenization and Fusion — Turning Images and Audio Into Tokens

  A deep look at how images, audio, and video become tokens and get woven into one sequence with text. We cover patch and VQ image tokenization, discrete-codec audio tokenization, frame sampling, interleaving and separator

  2026-06-26 · 14 min read #llm#multimodal#tokenization#vision-language#q-former

• Multimodal AI Training Methods — Many Senses in One Model

  A walkthrough of how multimodal AI learns to handle images, text, audio, and video in a single model. We cover modality alignment and contrastive learning, fusion strategies, shared embedding spaces, pretraining and fine

  2026-06-26 · 15 min read #ai-papers#multimodal#clip#vision-language#contrastive-learning

• Beyond OCR — OCR-free Document Understanding and Unified Models

  A traditional OCR pipeline splits into detection, recognition, and layout stages, but errors accumulate. We organize the shift in document AI: Donut-style and VLM-based OCR-free document understanding, high-resolution an

  2026-06-26 · 20 min read #ai-papers#ocr-free#document-understanding#multimodal#donut

• Vision-Language Models (VLMs) 2026 Deep Dive — CLIP, LLaVA, InternVL3, Qwen2.5-VL, GPT-4o, Gemini 2.5, Claude 4.7, DINOv2, SAM 2, and Florence-2

  Everything you need to know about Vision-Language Models in May 2026 in one place. CLIP family (SigLIP, EVA-CLIP), open VLMs (LLaVA-NeXT, InternVL3, Qwen2.5-VL, Pixtral, Molmo, Idefics3, MiniCPM-V), closed frontier (GPT-

  2026-05-16 · 19 min read #vision-language-models#vlm#clip#llava#internvl

• The Complete Guide to Multimodal LLMs: Vision, Document Understanding, OCR, Video, Audio, and the Specifics of Korean (2025)

  The text-only era is over. In 2025, LLMs handle images, documents, video, and audio naturally. GPT-4o/Claude 3.5/Gemini/Qwen2-VL/Pixtral compared, Document AI and layout understanding, the modernization of OCR, video and

  2026-04-15 · 14 min read #multimodal#vision-llm#document-ai#ocr#whisper

• Gemini 2.5 for Developers: A Practical Guide to Pro, Flash, and Flash-Lite

  A practical guide to Gemini 2.5 as of April 12, 2026, covering how to choose Pro, Flash, or Flash-Lite, how reasoning changes workflow design, and what teams should actually ship in production.

  2026-04-12 · 6 min read #google#gemini#gemini-2-5#coding#agentic-coding

• 2025 AI Research Trends: Top HuggingFace Papers and 10 Defining Research Directions

  A developer-focused review of HuggingFace trending papers and the 10 defining AI research trends of 2025. DeepSeek-R1 pure RL reasoning, Nemotron-Cascade 30B/3B MoE, GRPO, PagedAttention, million-token context limitation

  2026-03-21 · 15 min read #ai-research#papers#huggingface#reasoning#moe

• Multimodal AI Complete Guide: Master CLIP, LLaVA, GPT-4V, and Gemini Vision

  A complete guide to mastering multimodal AI from fundamentals to the latest vision-language models. Learn CLIP, BLIP-2, LLaVA, InstructBLIP, GPT-4V, Gemini Vision, and Claude Vision with practical code, plus Multimodal R

  2026-03-17 · 25 min read #multimodal#vision-language#clip#llava#gpt-4v

• LLM Multimodal Vision-Language Model Serving and Optimization Practical Guide

  A practical guide to serving and optimizing multimodal vision-language models in production.

  2026-03-05 · 24 min read #llm#multimodal#vlm#vllm#2026-03