slightly improved rtn calculate_qparams logic

Signed-off-by: Brian Dellabetta <[email protected]>
vllm-project · Mar 5, 2025 · 31884cf · 31884cf
1 parent 32b0b53
commit 31884cf
Showing 1 changed file with 1 addition and 1 deletion.
diff --git a/src/llmcompressor/observers/rtn.py b/src/llmcompressor/observers/rtn.py
@@ -42,7 +42,7 @@ def calculate_qparams(
             observed,
             symmetric=self.quantization_args.symmetric,
             bit_width=self.quantization_args.num_bits,
-            group_size=-1, #self.quantization_args.group_size,
+            group_size=self.quantization_args.group_size or -1,
         )
         return (scales, zp)