Update final v3 quantization benchmarks

.gitattributes

@@ -39,3 +39,5 @@ assets/mossvl_quantization_benchmark_comparison_zh_4k.png filter=lfs diff=lfs me
 assets/mossvl_quantization_benchmark_comparison_en_4k.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_final_v3_en_4k.png filter=lfs diff=lfs merge=lfs -text
+assets/mossvl_quantization_benchmark_comparison_final_v3_zh_4k.png filter=lfs diff=lfs merge=lfs -text

README.md

@@ -53,14 +53,15 @@ The checkpoint carries its bitsandbytes configuration. Load it directly and
 do not add a second runtime quantization configuration. This variant does not
 enable HQQ KV8; `generation_config.json` uses the standard BF16 KV cache.
 
-## Base model benchmark
+## Quantization benchmark
 
-The following figure reports the benchmark results of the original
-MOSS-VL-Instruct-0708 release. It provides base-model context; it is not
-presented as a separate quality evaluation of this NF4 checkpoint.
+The final evaluation compares the original BF16 model with all four release
+profiles on their corresponding benchmark suites. This offline NF4 checkpoint
+scores 89.53 on DocVQA, 67.30 on VideoMME, 75.86 on MLVU_dev, 51.00/48.17/59.33
+on the three TimeLens subsets, and 61.76 on VSIBench.
 
 <p align="center">
-  <img src="assets/benchmark-offline.png" alt="MOSS-VL-Instruct-0708 benchmark" width="100%"/>
+  <img src="assets/mossvl_quantization_benchmark_comparison_final_v3_en_4k.png" alt="MOSS-VL quantization benchmark comparison" width="100%"/>
 </p>
 
 ## Hardware requirements

README_zh.md

@@ -51,13 +51,14 @@ tags:
 checkpoint 已包含 bitsandbytes 配置，加载时不要再传入第二份运行时量化配置。
 该版本没有启用 HQQ KV8，`generation_config.json` 使用标准 BF16 KV Cache。
 
-## 基础模型 Benchmark
+## 量化 Benchmark
 
-下图是原始 MOSS-VL-Instruct-0708 的 benchmark，用于说明基础模型能力；
-它不作为该 NF4 checkpoint 的独立精度测评结果。
+最终测评在各自对应的 benchmark 上对比原始 BF16 模型与四个量化发布配置。
+该离线 NF4 checkpoint 的 DocVQA、VideoMME、MLVU_dev 分别为 89.53、
+67.30、75.86；三个 TimeLens 子集为 51.00、48.17、59.33，VSIBench 为 61.76。
 
 <p align="center">
-  <img src="assets/benchmark-offline.png" alt="MOSS-VL-Instruct-0708 benchmark" width="100%"/>
+  <img src="assets/mossvl_quantization_benchmark_comparison_final_v3_zh_4k.png" alt="MOSS-VL 量化配置 benchmark 对比" width="100%"/>
 </p>
 
 ## 硬件要求