Skip to content

Qwen3.8-Flash-Next on any consumer hardware: one-click install for Windows / Linux. Strata inference engine, OpenAI/Anthropic API on localhost, optional image input.

C++

setup: FP4-Local conversion installs torch 2.14.1 on Python 3.14+ (2.9.1 has no cp315 CPU wheel), older Pythons keep 2.9.1 (#3474)

beplee4681762e3c08
diff --git a/requirements-nvfp4-local.txt b/requirements-nvfp4-local.txt
index f0af1999b..6d4539254 100644
--- a/requirements-nvfp4-local.txt
+++ b/requirements-nvfp4-local.txt
@@ -2,6 +2,9 @@
# and only then. llama.cpp's converter imports torch and transformers; safetensors reads NVIDIA's shards.
# Licenses: torch BSD-3-Clause, transformers Apache-2.0, safetensors Apache-2.0 (all permissive; nothing of them is shipped with Strata).
# Pinned to the versions of the PC the conversion is developed on (torch 2.9.1 is the CPU build on Windows; on Linux setup asks for the CPU wheel).
-torch==2.9.1
+# torch 2.9.1 has no CPU wheel for Python 3.15 on the PyTorch index, so setup stopped with "No matching distribution found for
+# torch==2.9.1" there (#3474); from Python 3.14 on, 2.14.1 (+cpu wheels cp310-cp315 on Linux and Windows). Older Pythons keep 2.9.1.
+torch==2.9.1; python_version < "3.14"
+torch==2.14.1; python_version >= "3.14"
transformers==4.57.6
safetensors==0.7.0