Tiny Diffusion
A character-level text diffusion model from scratch
Details
- External ID
- 45876742
- Source
- HN
- Company
- —
- Product
- Tiny Diffusion
- Website domain
- github.com
- Launched
- Nov. 10, 2025
- Cohort
- —
- Upvotes
- 172
- Upvotes percentile
- 0.9454148471615721
- Tags
- —
- Fetched at
- Sept. 7, 2026, 9:25 p.m.
- Updated at
- Sept. 7, 2026, 9:25 p.m.
Description
This is a character-level language diffusion model for text generation.The model is a modified version of Nanochat's GPT implementation and is trained on Tiny Shakespeare!It is only 10.7 million parameters, so you can try it out locally.
Enrichment
- Theme
- AI text humanizers and detectors
- Vertical
- Horizontal
- Function
- Model & infra
- Audience
- Developer
- AI stance
- AI-native
- Project type
- Hobby / open-source project
- Normalized one-liner
- character-level text diffusion model
- Manually corrected
- False
Could you build this?
No Implementing and training a custom diffusion model from scratch on discrete text tokens involves non-trivial mathematical formulations of forward/reverse continuous-time or categorical diffusion processes and low-level PyTorch modeling.
What it would actually take: Building this requires deep knowledge of generative diffusion mathematics adapted for discrete token sequences (e.g., score-based models or categorical diffusion like D3PM/Plaid). The stack involves PyTorch/JAX, custom loss functions parameterized over diffusion timesteps, and training routines tuned to prevent mode collapse on discrete text. Specialized machine learning research and deep learning systems skills are required.
Discussion
20 comments analyzed.
Competitors mentioned: vllm (inference engine for diffusion LLMs), NVIDIA Fast-dLLM, char-mdlm
Concerns raised: No high-performance web server for inference included, Fixed token length limits for generation regions, Handling variable-length outputs (truncation vs. abandonment), Mobile compatibility, Missing project license
Feature requests: Integration with vllm for inference optimization, Inference engine/web server implementation, Variable-length generation without fixed token limits, Embedding vector generation via diffusion instead of discrete tokens, Terminal UI visualization (curses-based)
Competitors
Other products that read as similar to this one — 36 launches clear the similarity bar, closest 8 shown.
Attention rank: #5 of 37 (itself plus its competitors, highest first — normalized so YC and Product Hunt are compared fairly).
Looks like the first mover among its competitors.
- Mercury 2.5 · ph · 2026-09-09 · 3 upvotes · similarity 0.44
- Ideogram 4.0 · hn · 2026-06-03 · 46 upvotes · similarity 0.42
- GPT‑5.4 mini and nano · ph · 2026-03-18 · 267 upvotes · similarity 0.42
- NanoEuler · hn · 2026-06-28 · 55 upvotes · similarity 0.41
- NanoEuler · hn · 2026-06-19 · 6 upvotes · similarity 0.41
- "Be horse." · hn · 2026-04-30 · 10 upvotes · similarity 0.41
- Manual Diffusion · ph · 2026-09-15 · 1 upvotes · similarity 0.39
- Distilled 0.6B text-to-SQL model · hn · 2026-01-21 · 5 upvotes · similarity 0.39
Other launches for this product
- No other launches for this product.
Same idea, different domain
Nobody's really built a model & infra tool for Fintech yet.