diff --git a/quanto-introduction.md b/quanto-introduction.md
index a3eedee1e6..7d3775c6d8 100644
--- a/quanto-introduction.md
+++ b/quanto-introduction.md
@@ -108,14 +108,14 @@ Note: the first bar in each group always corresponds to the non-quantized model.
 
 <div class="row"><center>
   <div class="column">
-    <img src="https://github.com/huggingface/quanto/blob/main/bench/generation/charts/mistralai-Mistral-7B-v0.1_Accuracy.png" alt="mistralai/Mistral-7B-v0.1 Lambada prediction accuracy">
+    <img src="https://github.com/huggingface/quanto/blob/main/bench/generation/charts/mistralai-Mistral-7B-v0.1_Accuracy.png?raw=true" alt="mistralai/Mistral-7B-v0.1 Lambada prediction accuracy">
   </div>
  </center>
 </div>
 
 <div class="row"><center>
   <div class="column">
-    <img src="https://github.com/huggingface/quanto/blob/main/bench/generation/charts/mistralai-Mistral-7B-v0.1_Perplexity.png" alt="mistralai/Mistral-7B-v0.1 Lambada prediction accuracy">
+    <img src="https://github.com/huggingface/quanto/blob/main/bench/generation/charts/mistralai-Mistral-7B-v0.1_Perplexity.png?raw=true" alt="mistralai/Mistral-7B-v0.1 Lambada prediction accuracy">
   </div>
  </center>
 </div>
@@ -126,7 +126,7 @@ The graph below gives the latency per-token measured on an NVIDIA A100 GPU.
 
 <div class="row"><center>
   <div class="column">
-    <img src="https://github.com/huggingface/quanto/blob/main/bench/generation/charts/mistralai-Mistral-7B-v0.1_Latency__ms_.png" alt="mistralai/Mistral-7B-v0.1 Mean Latency per token">
+    <img src="https://github.com/huggingface/quanto/blob/main/bench/generation/charts/mistralai-Mistral-7B-v0.1_Latency__ms_.png?raw=true" alt="mistralai/Mistral-7B-v0.1 Mean Latency per token">
   </div>
  </center>
 </div>