DiabeticDaily-9B ๐Ÿ๐Ÿ 

The home-box tier of the OpenDiabetic ladder โ€” distilled-class diabetic intelligence sized to run on a home appliance (NAS + a low-power GPU like the RTX PRO 2000 Blackwell, ~70W). Cooked by Swarm and Bee LLC.

Beat-base โ€” proven

Held-out perplexity vs base Qwen3.5-9B (text never trained on):

held-out loss perplexity
Base Qwen3.5-9B 1.3625 3.906
DiabeticDaily-9B 0.8079 2.243
ฮ” โˆ’0.555 (+40.7% better)

Verdict: BEAT BASE โœ…. Models the domain ~41% better than base โ€” and its perplexity (2.24) is nearly the 27B anchor's (2.05): the knowledge survives the shrink. That's the distillation-ladder thesis, proven.

How it was cooked

  • Base: Qwen/Qwen3.5-9B (Apache-2.0). Data: the same deeded OpenDiabetic corpus as the 27B anchor.
  • Recipe: LoRA r64/ฮฑ32 on attn+mlp, LR 1e-5, cosine, early-stop overcook guard. Merged bf16.

The ladder: ๐Ÿ 27B anchor (+57%) โ†’ ๐Ÿ  9B home (+40.7%) โ†’ ๐Ÿ›๏ธ 4B edge (+40.4%)

โš ๏ธ Not medical advice โ€” diabetic lifestyle/education/organization only. Not a diagnosis. Emergencies โ†’ 911.

ยฉ 2026 Swarm and Bee LLC ยท opendiabetic.com ยท Apache-2.0 ยท We slow cook the truth. ๐Ÿ

Downloads last month
15
Safetensors
Model size
10B params
Tensor type
BF16
ยท
F32
ยท
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Model tree for SwarmandBee/DiabeticDaily-9B

Finetuned
Qwen/Qwen3.5-9B
Finetuned
(1,426)
this model