From dbc3eb112ba8d32ef6fdbf360cf54a9d177f59fb Mon Sep 17 00:00:00 2001 From: Alpamys Date: Tue, 24 Mar 2026 16:23:11 +0500 Subject: [PATCH] Update README: version refs to v0.10.7, add CPU compatibility note Co-Authored-By: Claude Opus 4.6 (1M context) --- README.md | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-) diff --git a/README.md b/README.md index eea7ac5..3fa98b2 100644 --- a/README.md +++ b/README.md @@ -603,7 +603,7 @@ soup version # Full system info (useful for bug reports) soup version --full -# -> soup v0.10.5 | Python 3.11.5 | CUDA 12.1 | extras: serve, data +# -> soup v0.10.7 | Python 3.11.5 | CUDA 12.1 | extras: serve, data ``` ## Web UI @@ -764,9 +764,11 @@ soup --verbose Full traceback on errors ## Requirements - Python 3.9+ -- GPU with CUDA (recommended) or Apple Silicon (MPS) or CPU (slow) +- GPU with CUDA (recommended) or Apple Silicon (MPS) or CPU (experimental) - 8 GB+ VRAM for 7B models with QLoRA +> **CPU note:** All training tasks (SFT, DPO, GRPO, PPO) work on CPU but will be very slow. Quantization (`4bit`/`8bit`) is auto-disabled on CPU. GRPO on CPU uses `min_new_tokens=1` to prevent empty generation errors. + ### Optional Extras | Extra | Install | What it adds |