From 33eba6a6c0d8e5ab0c4240095a8195e6d5cb5063 Mon Sep 17 00:00:00 2001
From: Atirna <288419661+atirna@users.noreply.github.com>
Date: Fri, 14 Aug 2026 00:32:02 +0530
Subject: [PATCH] docs: remove retired zero-gradient metric

---
 docs/training.md | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/docs/training.md b/docs/training.md
index af0b084f89..91caf061c3 100644
--- a/docs/training.md
+++ b/docs/training.md
@@ -199,7 +199,7 @@ Pulled from the console log and mirrored to W&B.
 **Stability and optimization:**
 
 - `optim/grad_norm` — spikes precede divergence.
-- `optim/lr`, `optim/zero_grad_ratio` — LR schedule and the fraction of params that received zero gradients (high → dead path or wrong loss masking).
+- `optim/lr` — LR schedule.
 - For MoE: `max_vio/mean` (load-balancing violation), `routing_confidence/mean` — both are logged when non-zero.
 
 **Performance:**