Developer releases Strata fork for 64GB Jetson AGX Orin
A Reddit developer released a fork of Strata v0.1.45 for the 64GB Jetson AGX Orin, primarily to run Qwen3.6-35B-A3B. The developer reported about 21 tok/s decoding with MTP-3 in llama.cpp, around 30 tok/s in earlier Strata runs, and roughly 40–41 tok/s in controlled short-prompt tests with sampled MTP-2. The author said these were setup-specific results rather than matched llama.cpp-versus-Strata benchmarks. The fork also includes unified INT8 KV caching, prefix checkpoints and optional KV preallocation. The developer disclosed that it was heavily AI-assisted, said it had been tested on their Orin, and invited feedback from other owners.
Hey guys, hope all is well.
I've spent the last few days working on a fork of Strata v0.1.45 for the Jetson AGX Orin 64GB, mainly running Qwen3.6-35B-A3B.
I was getting around 21 tok/s decode with MTP-3 in llama.cpp, versus around 30 tok/s in my earlier Strata runs. With sampled MTP-2 enabled, my latest controlled short-prompt tests reached roughly 40–41 tok/s. These weren't matched llama.cpp-versus-Strata benchmarks, so treat them as results from my setup.
The fork also includes unified INT8 KV caching, prefix checkpoints and optional KV preallocation.
Not sure how many other Orin owners are here, but figured I'd share in case it's useful: https://github.com/moontato/Strata-Orin
Full disclosure: heavily AI-assisted/vibe-coded. I've tested it on my Orin, but would welcome feedback from others. Credit to the original Strata project for the foundation.
#Strata v0.1.45#Jetson AGX Orin 64GB#Qwen3.6-35B-A3B#llama.cpp