bench: TP sweep harness (xserv --tp, llama row-split, concurrent groups)

runner/servers gain --tp (xserv --tp N; llama.cpp --split-mode row) and
--llama-devices so llama can run on a disjoint GPU group. run_tp_parallel.sh
runs xserv (GPU 0..N-1) and llama.cpp (GPU 4..4+N-1) concurrently per TP,
matching the box's 0-3 / 4-7 PHB groups. summarize_tp.py tabulates the sweep.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

tools/bench/servers.py

@@ -113,14 +113,18 @@ def xserv_launch_cmd(
     *,
     max_batch: int,
     max_seq_len: int,
+    tp: int = 1,
 ) -> list[str]:
-    return [
+    cmd = [
         bin_path,
         model_dir,
         "--port", str(port),
         "--max-batch", str(max_batch),
         "--max-seq-len", str(max_seq_len),
     ]
+    if tp > 1:
+        cmd += ["--tp", str(tp)]  # xserv binds rank r -> GPU r internally
+    return cmd
 
 
 def llama_cpp_launch_cmd(
@@ -131,13 +135,14 @@ def llama_cpp_launch_cmd(
     n_parallel: int,
     ctx_per_slot: int,
     n_gpu_layers: int = 99,
+    tp: int = 1,
 ) -> list[str]:
     # llama.cpp DIVIDES total -c across --parallel slots: per-slot context is
     # n_ctx / n_parallel. xserv gives each sequence the full max_seq_len, so to
     # match we must set total -c = ctx_per_slot * n_parallel. Getting this wrong
     # silently truncates long generations (e.g. AIME) on llama.cpp's side.
     total_ctx = ctx_per_slot * n_parallel
-    return [
+    cmd = [
         bin_path,
         "-m", gguf_path,
         "--port", str(port),
@@ -148,3 +153,9 @@ def llama_cpp_launch_cmd(
         # NOTE: do NOT pass --log-disable; its startup log reports per-slot
         # n_ctx, which is exactly the diagnostic that catches ctx misconfig.
     ]
+    if tp > 1:
+        # Tensor-parallel split across the visible GPUs (caller restricts the
+        # set via CUDA_VISIBLE_DEVICES in launch_env). Row-split is llama.cpp's
+        # tensor-parallel mode (vs the default layer/pipeline split).
+        cmd += ["--split-mode", "row"]
+    return cmd