Bro absolutely cooked with this one, but... it needed liberated.
used his convert/offload recipe on OrcaRouter’s BF16
https://t.co/vmB99kLOTE
https://t.co/0kN7VPmg8i
https://t.co/p1BvwDSp4k
@mr_r0b0t@OrcaRouter
Want to run Qwen3.8-Flash-Next on your RTX 3090?
r0b0tlab/Qwen3.8-Flash-Next-EXL3-2.50bpw
-Self-quantized, MTP + vision retained 👀
-38.6 tok/s in the speed sweep, 262K ctx
-Perfect 20/20 score on the multi-turn BFCL subset
Weights, runtime & evals:
https://t.co/f9csrOgi7p
@Murican_Pitbull@ItsmeAjayKV@OpenBMB might be getting a laptop for this! its got a 3080ti 16gb with 32gb ddr5 amd a i9-12900H guy only wants 1100. this guy might fit nicely!
No GPU acceleration, which is a bit surprising considering the source.
Going to try and run the big 1M model dataset through it and get an idea of how good this kernel actually is.
The NFA ruling has now been applied nationwide per the new ATF guidance! In this video, we explain what that means for our customers. This is a huge win for freedom and the Second Amendment. Follow us for more updates as this develops.
DISCLAIMER: This video is not legal advice. Review your local and state laws and consult with an attorney before making decisions as it pertains to this information.
@AdamTheBanned "the government of the United States of America is not in any sense founded on the Christian religion"
Article 11 of the 1797 Treaty of Tripoli, which was signed by President John Adams and unanimously ratified by the U.S. Senate.
XeStrata xe0.1.43
Local AI on smaller GPUs and less RAM: two new 3B-active MoE models (beta)
- Ornith-1.5-35B-A3B
- llm-jp-4.1-32B-A3B
From 4 GB VRAM and about 22-30 GB RAM.
Multi-GPU layer split reworked toward everyday use (experimental).