Tiny Diffusion
AI
text diffusion model
SHOW HN2025
172points
20comments
#404of 3,837 in 2025
Top 11%of 2025 by points
How HN took it
CuriousMostly questions and neutral comments
3 positive, 9 neutral or asking, 0 mixed, 1 critical, out of 13 comments about the project. Thread score +0.18.
Execution: 1 for, 1 againstDesign: 1 for, 0 against
The thread, comment by comment
Each row is a top-level comment and its replies, in HN's order. Select a cell to open that comment on HN.
positive neutral or mixed critical about another commenter or something else the maker
What HN said
What people liked
Fun project with easy-to-understand code and nice results
From 2 comments
Fun project, easy to understand and nice looking results, everything one could ask for! I played around with it locally, did some optimizations of low hanging fruits without…
embedding-shape on HNthis is so so awesome.
aa_y_ush on HN
Neat idea, curses terminal UI variant
From 1 comment
This is really neat. I noticed the diffusion-process.py demo was using matplotlib in a window, but I figured it would be cute if it used a terminal UI instead - so I had Claude…
simonw on HN
Objections
No license was attached to the project initially
From 2 comments
Fun project, easy to understand and nice looking results, everything one could ask for! I played around with it locally, did some optimizations of low hanging fruits without…
embedding-shape on HNHey, I’ll add the MIT licenses later today!
nathan-barry on HN
Requests
Add a license
From 1 comment
Fun project, easy to understand and nice looking results, everything one could ask for! I played around with it locally, did some optimizations of low hanging fruits without…
embedding-shape on HN
Build an inference engine for diffusion LLMs
From 2 comments
One year later and there is still no inference engine for diffusion LLMs Students looking for a project to break into AI - please!
gdiamos on HNtraining inspired on nanochat for diffusion models: https://github.com/ZHZisZZ/dllm now someone needs to make it work with vllm or something
tough on HN
Questions
How masked token length is handled when output length differs
From 3 comments
Why do these text diffusion demos always look like the number of allowed tokens is fixed for a specific unfilled region? Is this the case? Ie. if the region only has four…
yugretcx on HNTokens start as a special [MASK] token. Then as the diffusion process runs they are "unmasked" i.e. sampled. So yes, you define a sequence of [MASK] tokens with some length ahead…
rand0mwalk on HNYes, this is the case. During training, the model will get a sequence of text (ex, 512 tokens long) with a percentage of them masked out (with a special <MASK> token). It learns…
nathan-barry on HN
Why ReLU^2 activation was chosen
From 2 comments
The basic MLP block in this model uses a ReLU^2 activation function (x <- ReLU(x)^2). That seems to be copied from the nanochat project, and it's not present in nanoGPT. Is there…
Majromax on HNWhen it comes to compute cost the choice of activation function makes little difference nowadays (and it can often be fused with whatever operation comes before it, which makes…
kouteiheika on HN
Why game of life sampling is used
From 1 comment
Looks fun, thanks for sharing. I see you're implementing game of life sampling, what's the reasoning for using this logic?
tell_me_whai on HN
Can it run on mobile
From 1 comment
This is impressive. Can it run on mobile?
doppelgunner on HN
Alternatives named
NVIDIA Fast-dLLM
From 1 comment
Actually NVIDIA made one earlier this year, check out their Fast-dLLM paper
nathan-barry on HN
char-mdlm
From 1 comment
There is also this one that was released in October: https://github.com/kuleshov/char-mdlm
volodia on HN
dllm
From 1 comment
training inspired on nanochat for diffusion models: https://github.com/ZHZisZZ/dllm now someone needs to make it work with vllm or something
tough on HN