fix(cpp): address failing template deduction with some compilers

huggingface · Feb 9, 2024 · f9e7779 · f9e7779
1 parent 36124db
commit f9e7779
Showing 1 changed file with 1 addition and 1 deletion.
diff --git a/quanto/library/ext/cpp/quantize.cpp b/quanto/library/ext/cpp/quantize.cpp
@@ -12,7 +12,7 @@ torch::Tensor quantize_symmetric_per_tensor(const torch::Tensor& input, const to
     float inv_scale = float_scale == 0 ? 1.0f : 1.0f / float_scale;
     for (const auto i : c10::irange(numel)) {
         int64_t qvalue = lrintf(std::nearbyint(data[i] * inv_scale));
-        qvalue = std::max(-127LL, std::min(qvalue, 127LL));
+        qvalue = std::max<int64_t>(-127LL, std::min<int64_t>(qvalue, 127LL));
         qdata[i] = static_cast<int8_t>(qvalue);
     }
     return output;