From 7de9141164117e31687f044eff9ccd6a217a4a84 Mon Sep 17 00:00:00 2001
From: Nicolas Patry <patry.nicolas@protonmail.com>
Date: Fri, 26 Jan 2024 10:30:01 +0000
Subject: [PATCH] Adding a comment.

---
 server/text_generation_server/utils/gptq/exllamav2.py | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/server/text_generation_server/utils/gptq/exllamav2.py b/server/text_generation_server/utils/gptq/exllamav2.py
index 25ff508b..2b897f25 100644
--- a/server/text_generation_server/utils/gptq/exllamav2.py
+++ b/server/text_generation_server/utils/gptq/exllamav2.py
@@ -185,6 +185,9 @@ class QuantLinear(nn.Module):
             "g_idx": self.g_idx,
         }
         temp_dq = temp_dq.get_scratch_slice(self.temp_dq_size())
+
+        # We NEED to keep a pointer on Python side, otherwise the garbage collector will mess with us,
+        # and `Memory access fault by GPU node-2` will EAT you.
         self.temp_dq = temp_dq
         self.q_handle = ext_make_q_matrix(self.q_tensors, temp_dq)