[https://nvbugs/5405041][fix] Update wide-ep doc (NVIDIA#6933)

Signed-off-by: Xianjie <5410381+qiaoxj07@users.noreply.github.com>

Author: qiaoxj07

examples/wide_ep/slurm_scripts/submit.sh

@@ -12,7 +12,7 @@ model_dir=<model_dir>  # Path to the model checkpoint
 repo_dir=<repo_dir>  # Path to the repo to install TensorRT-LLM, if this is empty, the pre-installed version will be used
 
 mtp_size=0
-ntasks_per_node=4 # 4 GPUs per GB200 node
+ntasks_per_node=4 # 4 GPUs per GB200 node, 8 GPUs per B200 node
 
 isl=1024
 osl=1024
@@ -23,8 +23,9 @@ streaming=true
 for b in 1 64 1024; do
     for eplb_num_slots in 0 256 288; do
         concurrency=$((b * 16))
-        ctx_num=$(((concurrency + 5499)/5500))
-        total_node_num=$((ctx_num + 4))
+        ctx_node_num=$(((concurrency + 5499)/5500)) # $(((concurrency + 10999)/11000)) for B200
+        ctx_num=${ctx_node_num} # $((ctx_node_num * 2)) for B200
+        total_node_num=$((ctx_node_num + 4)) # $((ctx_node_num + 2)) for B200
         ntasks=$((total_node_num * ntasks_per_node))
 
         args=(
@@ -58,8 +59,9 @@ done
 # dep32 eplb288
 for b in 512; do
     concurrency=$((b * 32))
-    ctx_num=$(((concurrency + 5499)/5500))
-    total_node_num=$((ctx_num + 8))
+    ctx_node_num=$(((concurrency + 5499)/5500)) # $(((concurrency + 10999)/11000)) for B200
+    ctx_num=${ctx_node_num} # $((ctx_node_num * 2)) for B200
+    total_node_num=$((ctx_node_num + 8)) # $((ctx_node_num + 4)) for B200
     ntasks=$((total_node_num * ntasks_per_node))
     eplb_num_slots=288