Update README

README-cn.md

@@ -22,7 +22,7 @@
 2.5B 稠密模型。单张消费级显卡或一台笔记本就能运行，保留原生 128k 上下文。
 
 - **同尺寸下很强。** 作为最多思考 128K token 的普通单次调用，AIME 2025 92.3、AIME 2026 89.8、HMMT 2026 年 2 月赛 81.8
-  （avg@16），16 样本投票为 100 / 93.3 / 90.9。官方通用模型 MiniCPM5-2B 在同一协议下为 87.1 / 89.8 / 68.4。
+  （avg@16），16 样本投票为 100 / 93.3 / 90.9。整体 AIME，HMMT 测评上超越原版通用模型 MiniCPM5-2B。
 - **能写出可核查的证明。** 在 IMO-ProofBench Basic 上，一次最多 128K token 的普通调用平均能为 30 题中的 21 题写出完整证明，
   证明蜂群为 23 题（AI 评审盲评）。
 - **自带"证明蜂群"。** 开源的 [MiniCPM-Math harness](https://github.com/Caldalis/MiniCPM-Math-harness) 可以在任意兼容
@@ -94,8 +94,7 @@ curl http://localhost:8000/v1/chat/completions \
   }'
 ```
 
-vLLM 0.24.0 + RTX 4090 实测：单卡、短输入时单流解码约 155 tok/s（保持 CUDA Graph 开启，不要加 `--enforce-eager`，
-eager 模式只有约 48 tok/s）。在五张 4090、最多 10 路并发的长证明任务中，单请求解码速度中位数约 118 tok/s，总吞吐约 900 tok/s。
+vLLM 0.24.0 + RTX 4090 实测：单卡、短输入时单流解码约 155 tok/s。在五张 4090、最多 10 路并发的长证明任务中，单请求解码速度中位数约 118 tok/s，总吞吐约 900 tok/s。
 
 ### SGLang
 
@@ -110,9 +109,7 @@ python -m sglang.launch_server --model-path Caldalis/MiniCPM5-2B-Math --context-
 llama-server -hf Caldalis/MiniCPM5-2B-Math-GGUF:Q8_0 --jinja -ngl 99 -c 98304 -np 4 -kvu --port 8000
 ```
 
-8000 是 harness 默认连接的端口（不加 `--port` 时 llama-server 监听 8080）；配合这个服务运行 harness 时请用 `--profile quick`。
-请使用 llama.cpp b9833 或更新版本：b9354 加入了 MiniCPM5 分词器，更早的版本加载不了这个 GGUF；b9833 加入了把推理和答案分开的
-chat parser。Apple M5 笔记本（16 GB）+ Q8_0 实测：单流约 46 tok/s，4 个 slot 并发总计约 125 tok/s。llama.cpp 默认的
+8000 是 harness 默认连接的端口（不加 `--port` 时 llama-server 监听 8080）；配合这个服务运行 harness 时请用 `--profile quick`。Apple M5 笔记本（16 GB）+ Q8_0 实测：单流约 46 tok/s，4 个 slot 并发总计约 125 tok/s。llama.cpp 默认的
 `min_p=0.05` 容易引发复读，请在请求中设置 `min_p=0.0`（harness 会自动设置）。
 
 ### Transformers

README.md

@@ -41,8 +41,8 @@ English |
 writing. It runs on a single consumer GPU or a laptop and keeps the native 128K context.
 
 - **Strong for its size.** As a plain single call with up to 128K tokens of thinking, 92.3 on AIME 2025, 89.8 on
-  AIME 2026 and 81.8 on HMMT February 2026 (avg@16), and 100 / 93.3 / 90.9 with a 16-sample vote. The official
-  general-purpose MiniCPM5-2B scores 87.1 / 89.8 / 68.4 under the same protocol.
+  AIME 2026 and 81.8 on HMMT February 2026 (avg@16), and 100 / 93.3 / 90.9 with a 16-sample vote. Overall, it
+  outperforms the original general-purpose MiniCPM5-2B on AIME and HMMT.
 - **Writes checkable proofs.** On IMO-ProofBench Basic, an average single call with up to 128K tokens writes a
   complete proof for 21 of the 30 problems, and the proof swarm for 23, as graded blind by an AI grader.
 - **Comes with a proof swarm.** The open-source [MiniCPM-Math Harness](https://github.com/Caldalis/MiniCPM-Math-harness) runs
@@ -119,9 +119,9 @@ curl http://localhost:8000/v1/chat/completions \
   }'
 ```
 
-Measured with vLLM 0.24.0 on RTX 4090s: about 155 tok/s single-stream decode on one GPU with short inputs (keep
-CUDA graphs on; do not use `--enforce-eager`, which measured about 48 tok/s). On long proof runs across five 4090s
-with up to 10 concurrent requests, the median request decoded at about 118 tok/s, about 900 tok/s in total.
+Measured with vLLM 0.24.0 on RTX 4090s: about 155 tok/s single-stream decode on one GPU with short inputs. On
+long proof runs across five 4090s with up to 10 concurrent requests, the median request decoded at about 118 tok/s,
+about 900 tok/s in total.
 
 ### SGLang
 
@@ -137,10 +137,9 @@ llama-server -hf Caldalis/MiniCPM5-2B-Math-GGUF:Q8_0 --jinja -ngl 99 -c 98304 -n
 ```
 
 Port 8000 is where the harness looks by default (`llama-server` alone listens on 8080); with this server, run the
-harness with `--profile quick`. Use llama.cpp b9833 or newer: b9354 added the MiniCPM5 tokenizer, so older builds
-cannot load the GGUF, and b9833 added the chat parser that separates the reasoning from the answer. On an Apple M5
-laptop (16 GB) with Q8_0: about 46 tok/s for one stream, about 125 tok/s total with 4 slots. llama.cpp's default
-`min_p=0.05` can encourage repetition, so send `min_p=0.0` (the harness does this for you).
+harness with `--profile quick`. On an Apple M5 laptop (16 GB) with Q8_0: about 46 tok/s for one stream, about
+125 tok/s total with 4 slots. llama.cpp's default `min_p=0.05` can encourage repetition, so send `min_p=0.0` (the
+harness does this for you).
 
 ### Transformers