mlsanityData to 14 Oct 2026

Holding the FP8 Quality Ceiling at 8-Bit Weights and Activations: INT8 and GGUF Post-Training Quantization of Ideogram 4.0 for Consumer GPUs

arXiv 2606.12280 ↗First posted 10 Jun 2026cs.LGcited by 1

On the map: Diffusion models → Diffusion foundations and sampling

Abstract (arXiv)

We study post-training quantization (PTQ) of Ideogram 4.0, a 9.3B flow-matching diffusion transformer (DiT) that realizes classifier-free guidance with two separate-weight copies of a single-stream backbone and is conditioned by a Qwen3-VL text encoder, targeting Ampere RTX~3090 GPUs, which lack FP8 tensor cores. Because Ideogram~4.0 is trained on structured JSON captions, we evaluate every variant under schema-valid JSON prompts produced by an LLM expander built to Ideogram's published caption specification, and score them with a battery spanning human-preference (HPSv2), CLIP, and PickScore for standalone quality; PP-OCR exact-match and edit distance for text; and PSNR/SSIM/LPIPS for fidelity to the FP8 reference (the highest-precision public checkpoint) output. On a 300-prompt benchmark with paired bootstrap confidence intervals, an INT8 W8A8 recipe (per-channel weights, per-token dynamic activations, SmoothQuant, and bf16 protection of a small high-fragility layer set) is statistically indistinguishable from FP8 on CLIP and PickScore (paired CIs include zero) and within ~0.004 HPSv2, and, at its 8-bit size, is the most faithful reproduction of the FP8 output (LPIPS 0.243 vs 0.277/0.306 for the half-size 4-bit baselines; the INT8-Q4_K gap excludes zero). A GGUF Q4_K quantization reaches the same standalone quality as the published NF4 baseline at the same on-disk size, making it the Pareto choice on the quality-memory frontier. We further show that under JSON prompts all four variants reach parity on standalone quality, the variants separate on fidelity and text rendering, not on aggregate image-quality scores, and that text legibility, near-zero when the model is prompted with raw strings, reaches 55% OCR exact-match under the JSON captions it expects. We release the INT8 W8A8 and GGUF Q4_K quantized weights on Hugging Face under a gated, non-commercial license.

Key facts. Cited by 1 paper in the map (data through 14 Oct 2026). Placed in Diffusion foundations and sampling, one of 14 directions in Diffusion models. Builds on 24 of its 27 references that are in the map. First posted on arXiv 10 Jun 2026.

Builds on

24 of 27 references.

Closest ancestors · 14

Same direction, newest first

  1. 2026Boundary-Protection W8A8 HiFloat8 Quantization for Large-Scale Text-to-Video Diffusion Transformers31 May
  2. 2026Timestep-Aware SVDQuant-GPTQ for W4A4 Quantization of Wan2.2-I2V26 May
  3. 2026Accelerating Rectified Flow Models via Trajectory-Aware Caching16 May
  4. 2026DiRotQ: Rotation-Aware Quantization for 4-bit Diffusion Transformers16 May
  5. 2025Q-Sched: Pushing the Boundaries of Few-Step Diffusion Models with Quantization-Aware Scheduling1 Sep
  6. 2025FP4DiT: Towards Effective Floating Point Quantization for Diffusion Transformers19 Mar
  7. 2024PQD: Post-training Quantization for Efficient Diffusion Models30 Dec
  8. 2024Efficiency Meets Fidelity: A Novel Quantization Framework for Stable Diffusion9 Dec
  9. 2024SVDQuant: Absorbing Outliers by Low-Rank Components for 4-Bit Diffusion Models7 Nov
  10. 2024Accurate Compression of Text-to-Image Diffusion Models via Vector Quantization31 Aug
  11. 2024ViDiT-Q: Efficient and Accurate Quantization of Diffusion Transformers for Image and Video Generation4 Jun
  12. 2023DeepCache: Accelerating Diffusion Models for Free1 Dec
2 more
  1. 2023PTQD: Accurate Post-Training Quantization for Diffusion Models18 May
  2. 2022Scalable Diffusion Models with Transformers19 Dec

Other references · 10

  1. 2022Flow Matching for Generative Modelingcited by 5,702
  2. 2023QLoRA: Efficient Finetuning of Quantized LLMscited by 3,418
  3. 2022GPTQ: Accurate Post-Training Quantization for Generative Pre-trained Transformerscited by 1,919
  4. 2022SmoothQuant: Accurate and Efficient Post-Training Quantization for Large Language Modelscited by 1,386
  5. 2022LLM.int8(): 8-bit Matrix Multiplication for Transformers at Scalecited by 906
  6. 2024Timestep Embedding Tells: It's Time to Cache for Video Diffusion Modelcited by 269
  7. 2024FonTS: Text Rendering with Typography and Style Controlscited by 40
  8. 2024CharGen: High Accurate Character-Level Visual Text Generation Model with MultiModal Encodercited by 14
  9. 2026Q-DiT4SR: Exploration of Detail-Preserving Diffusion Transformer Quantization for Real-World Image Super-Resolutioncited by 5
  10. 2025HDGlyph: A Hierarchical Disentangled Glyph-Based Framework for Long-Tail Text Rendering in Diffusion Modelscited by 1

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2026: 1.

1 paper.

  1. 2026Realizing Native INT8 Compute for Diffusion Transformers on Consumer GPUs: A Fused INT8 GEMM Kernel for Ideogram 4.0cited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2025INT v.s. FP: A Comprehensive Study of Fine-Grained Low-bit Quantization Formats
  2. 2023Efficient Post-training Quantization with FP8 Formats
  3. 2023ZeroQuant-FP: A Leap Forward in LLMs Post-Training W4A8 Quantization Using Floating-Point Formats
  4. 2025FP8-Flow-MoE: A Casting-Free FP8 Recipe without Double Quantization Error
  5. 2023FP8-BERT: Post-Training Quantization for Transformer
  6. 2023LLM-FP4: 4-Bit Floating-Point Quantized Transformers
  7. 2023INT-FP-QSim: Mixed Precision and Formats For Large Language Models and Vision Transformers
  8. 2024Scaling FP8 training to trillion-token LLMs