Enhance README with sampling parameter recommendations and Llama.cpp usage examples

README-cn.md

@@ -268,6 +268,13 @@ MiniCPM5-2B 的训练过程是 **[UltraData 分级数据管理体系](https://ar
 
 ## 快速上手
 
+> [!Tip]
+> 我们建议在生成时使用以下采样参数组合：`temperature=1.0, top_p=0.95, min_p=0.0`。
+>
+> 如果遇到重复输出，请尝试：`temperature=1.0, top_p=0.95, min_p=0.0, repetition_penalty=1.05`。
+>
+> 请注意，不同推理框架对采样参数的支持程度有所差异。
+
 ### vLLM
 
 ```bash
@@ -316,6 +323,26 @@ python -m sglang.launch_server \
   --port 30000
 ```
 
+### Llama.cpp
+
+```bash
+llama-server -m MiniCPM5-2B-F16.gguf -a MiniCPM5-2B --port 8080 -ngl 99 -c 8192 --jinja
+```
+
+在这个例子里 `-c 8192` 设置了上下文长度为 8192，可以根据需要修改上下文长度。
+
+```bash
+curl http://localhost:8080/v1/chat/completions \
+    -H "Content-Type: application/json" \
+    -d '{
+        "model": "MiniCPM5-2B",
+        "messages": [{"role": "user", "content": "1+1=?"}],
+        "temperature": 1.0, "top_p": 0.95, "min_p": 0.0, "max_tokens": 256
+    }'
+```
+
+llama.cpp 默认设置 `min_p=0.05`，会过滤掉概率低于最高概率 token 5% 的 token。这种过滤反而可能引发重复——它恰恰过滤掉了模型摆脱重复循环所需的那些 token。我们明确将 `min_p` 设为 `0.0` 以禁用该过滤。
+
 ### Transformers
 
 ```bash
@@ -344,8 +371,6 @@ outputs = model.generate(**inputs, max_new_tokens=128)
 print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:], skip_special_tokens=True))
 ```
 
-推荐的采样参数：`temperature=1.0, top_p=0.95`
-
 ## 工具调用
 
 工具调用**推荐使用 SGLang**。MiniCPM5-2B 以 XML 格式产出工具调用，SGLang 内置的 `minicpm5` parser 会自动将其转换为 OpenAI 兼容的 `tool_calls` 字段。

README.md

@@ -279,6 +279,13 @@ During **post-training**, we proceed in three steps: **SFT**, **RL**, and **OPD*
 
 ## Quickstart
 
+> [!Tip]
+> We recommend using the following sets of sampling parameters for generation: `temperature=1.0, top_p=0.95, min_p=0.0`.
+>
+> If you encounter repetitive outputs, try: `temperature=1.0, top_p=0.95, min_p=0.0, repetition_penalty=1.05`.
+>
+> Please note that the support for sampling parameters varies according to inference frameworks.
+
 ### vLLM
 
 ```bash
@@ -327,6 +334,26 @@ python -m sglang.launch_server \
   --port 30000
 ```
 
+### Llama.cpp
+
+```bash
+llama-server -m MiniCPM5-2B-F16.gguf -a MiniCPM5-2B --port 8080 -ngl 99 -c 8192 --jinja
+```
+
+`-c 8192` sets the context length. You can adjust this value as needed.
+
+```bash
+curl http://localhost:8080/v1/chat/completions \
+    -H "Content-Type: application/json" \
+    -d '{
+        "model": "MiniCPM5-2B",
+        "messages": [{"role": "user", "content": "1+1=?"}],
+        "temperature": 1.0, "top_p": 0.95, "min_p": 0.0, "max_tokens": 256
+    }'
+```
+
+In llama.cpp, the default `min_p=0.05` can lead to repetitive output: it filters out tokens whose probability is below 5% of the highest-probability token, potentially discarding the exact tokens needed to break out of a repetition loop. To prevent this, we set `min_p=0.0`.
+
 ### Transformers
 
 ```bash
@@ -355,8 +382,6 @@ outputs = model.generate(**inputs, max_new_tokens=128)
 print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:], skip_special_tokens=True))
 ```
 
-Recommended sampling params: `temperature=1.0, top_p=0.95`
-
 ## Tool Calling
 
 For tool / function calling, **SGLang is the recommended backend**. MiniCPM5-2B emits XML-style tool calls and SGLang's built-in `minicpm5` parser converts them to OpenAI-compatible `tool_calls` natively: