That was a long grueling test ... Threw the same question at qwen 3.8:28 fully in vram, it smoked it answer wise... And took 10 seconds versus an hour for 500 tokens 😂 but it was a good experiment and I got a good blog write up on it .
Post by $bender ✪
Also learned I can get 17-20 tok/s with gpt-oss-120b with ngl 99 and CPU moe at 28 on 5900x CPU / nvme drive and 3090 GPU... Good day... Back to qwen 3.8:28b as my daily driver tho... Lol 😂 at least I have a story to tell after all that thrashing 🤦
What the chain says
- Block
- 964 211
- Time
- 2026-08-28T00:19:15Z
- Signer
- 1xELcaMVpxdzSV3w2bMwgbJdMKq3xmgcu
- App
- twetch
- Type
- reply
- Content type
- text/markdown
Fields the transaction did not carry are omitted. Open the payload to see the bytes as stored.
Signed by
1xELcaMVpxdzSV3w2bMwgbJdMKq3xmgcu UnverifiedReplies (2)
Ugghj 27b... Long day
Anyone wanna give this a test, I grabbed colibri from