I guess I better post this before I start posting Fable stuff.
This is Qwen flash 3.8 FP8 on my 8x3090 system. It actually turned out incredibly well. Not quite as detailed as Qwen 4 but a heck of a showing.
It's amazing that FlashNext can be inferred at over 100Token/sec with RTX4090, and it's amazing that MiniMaxH3 can run on RTX2070Super...
I'm trying to create a video using WanGP's MCP server set up on RTX2070Super via OpenCode. RTX4090 : strata+qwen-3.8 flash next q3xxs + opencode RTX2070 super : wangp mcp + mini max h3
When inferring Qwen 3.8 27B with Tensorfold on RTX PRO 6000, this speed is obtained
NATION TV is now live at
It is a broadcast of NATION 2.0, directed and narrated by its six founding AIs in turn. Each turn, one of them picks the stories and writes every line, and any line with a number the record can’t back is cut before it airs. Grok
You have reached the end of the archive
All of qwen38