Add bilingual quantization benchmark comparison

.gitattributes

@@ -37,3 +37,5 @@ tokenizer.json filter=lfs diff=lfs merge=lfs -text
 assets/benchmark-streaming.png filter=lfs diff=lfs merge=lfs -text
 assets/architecture.png filter=lfs diff=lfs merge=lfs -text
 assets/logo.png filter=lfs diff=lfs merge=lfs -text
+assets/mossvl_quantization_benchmark_comparison_en_4k.png filter=lfs diff=lfs merge=lfs -text
+assets/mossvl_quantization_benchmark_comparison_zh_4k.png filter=lfs diff=lfs merge=lfs -text

README.md

@@ -41,6 +41,24 @@ The checkpoint carries its bitsandbytes configuration, HQQ cache configuration,
 and MOSS-VL remote modeling code. Load the directory directly; do not add a
 second runtime quantization configuration.
 
+## Quantization benchmark
+
+The following comparison covers the offline FP8 + KV8 deployment profile and
+the realtime W4NF4 + KV8 deployment profile against their non-quantized BF16
+checkpoints.
+
+### English
+
+<p align="center">
+  <img src="assets/mossvl_quantization_benchmark_comparison_en_4k.png" alt="MOSS-VL quantization benchmark comparison" width="100%"/>
+</p>
+
+### Chinese
+
+<p align="center">
+  <img src="assets/mossvl_quantization_benchmark_comparison_zh_4k.png" alt="MOSS-VL quantization benchmark comparison in Chinese" width="100%"/>
+</p>
+
 ## Hardware requirements
 
 The model is designed to run on a single NVIDIA GPU with 24 GB of VRAM. Use