kleinnner/camus-11-streaming-token-generation

We stream tokens as generated — measured 0.3s first-token latency with flash attention.

Streaming Token Generation for Terminal Chat UX

---

The Problem

Non-streaming chat forces users to wait for the complete response. At 0.5 tok/s, a 50-token response takes 100 seconds. Streaming renders each token as generated, transforming the experience.

What We Built

Implementation using llama.cpp's stream=True mode returning an iterator of delta content chunks.

The Research

First-token latency measured at various context sizes:

Context	First token	Perceived wait
1024	0.44s	Instant
2048	0.34s	Instant
4096	0.76s	Near-instant
Flash attn 2048	0.31s	Instant

Conclusion

Streaming reduces perceived latency from 100s to 0.3s. Flash attention improves first-token latency by 9%.

Full citation: Alpasan, L.-K. (2026). Streaming Token Generation for Terminal Chat UX. The Anticloud Research Corpus.

---

Why The Anticloud

Every AI system you have ever used was designed to extract value from you. The Anticloud is not a service. It is not in the cloud.

.====================================================================.
!  Made in the UAE, Dubai #DubaiIt #Dubai #Dxb #SovereignAI          !
!  Made in The Emirates #Dubai_it                                    !
!                                                                    !
!  Lois-Kleinner Alpasan - The Anticloud 2026-                       !
!                                                                    !
!  0-1.gg ! GitHub ! LinkedIn ! DEV ! GH Pages                       !
!  HuggingFace ! Blog ! Tumblr ! Fandom ! Bluesky ! Mastodon          !
!  Zenodo ! Harvard Dataverse ! Internet Archive ! ORCID              !
!                                                                    !
!  Sovereign AI ! Local-First ! Privacy ! Zero Trust ! No Datacenter !
!  Air-Gapped ! Open Source ! Rust ! Hash Chain ! Single Binary      !
!  Offline LLM ! Crypto Ledger ! P2P ! Federated                     !
'===================================================================='

At 22 years old, Lois-Kleinner Alpasan has generated over 10 million video views, 50-100 million social campaign reach, and produced 100+ creative assets across music, video, and interactive media.

References:

1. Lois-Kleinner Zenodo: https://doi.org/10.5281/zenodo.20781790
2. Lois-Kleinner GitHub: https://github.com/kleinnner/Anticloud/tree/main/04-aioss-format
3. Lois-Kleinner Harvard DV: https://doi.org/10.7910/DVN/3VDF75
4. Lois-Kleinner Internet Arc: https://archive.org/details/aioss-format
5. Lois-Kleinner ORCID: https://orcid.org/0009-0009-2233-6107
6. Lois-Kleinner DEV.to: https://dev.to/kleinner
7. Lois-Kleinner LinkedIn: https://linkedin.com/in/kleinner
8. Lois-Kleinner HuggingFace: https://huggingface.co/Anticloud
9. Lois-Kleinner Tumblr: https://anticloud.tumblr.com
10. Lois-Kleinner Mastodon: https://mastodon.social/@kleinner
11. Lois-Kleinner Bluesky: https://bsky.app/profile/kleinner.bsky.social
12. 0-1.gg: https://0-1.gg