update hybrid-mtp-with-ngram (#3924)

docs/features/speculative_decoding.md

@@ -18,6 +18,13 @@ This project implements an efficient **Speculative Decoding** inference framewor
   - ⏳ Coming Soon: Support Chunk-prefill
   - ⏳ Coming Soon: Multi-layer MTP Layer
 
+- **Decoding with Hybrid MTP and Ngram Methods(Hybrid-MTP-with-Ngram)**
+
+  - Overview: A hybrid method combining MTP and Ngram. First, MTP generates N draft tokens, then Ngram matching is used to supplement additional draft tokens.
+
+  - Use Cases: Suitable when higher draft token coverage is required, leveraging both MTP’s generation capability and the efficiency of Ngram matching.
+
+
 ---
 
 ### Coming Soon
@@ -132,7 +139,13 @@ python -m fastdeploy.entrypoints.openai.api_server \
     --scheduler-password "scheduler_mtp" \
     --speculative-config '{"method": "mtp", "num_speculative_tokens": 1, "model": "${path_to_mtp_model}"}' &
 ```
+## Decoding with Hybrid MTP and Ngram Methods
 
+When starting the service, you only need to modify the --speculative-config option.
+For example, use MTP to generate two draft tokens, and then append three additional draft tokens from Ngram matching:
+```
+--speculative-config '{"method": "mtp", "num_model_steps": 2, "mtp_strategy": "with_ngram", "num_speculative_tokens": 5, "model": "'$model_path'/mtp"}'
+```
 ## 🧠 Using Ngram-Based Decoding
 This method uses an n-gram sliding window to match the prompt and generated tokens to predict draft tokens. It is particularly effective in scenarios with high input-output overlap (e.g., code completion, document search).
 

docs/zh/features/speculative_decoding.md

@@ -14,6 +14,9 @@
   - ⏳ 即将支持：兼容 Chunk Prefill
   - ⏳ 即将支持：多层 MTP layer
 
+- **混合MTP、Ngram方法解码(Hybrid-MTP-with-Ngram)**
+  - 方法概述：混合MTP与Ngram方法，先使用MTP产出N个草稿Token，再使用Ngram匹配补充草稿Token。
+  - 使用场景：适合在需要更多草稿Token时使用，兼顾MTP生成能力与Ngram匹配的高效性。
 ---
 
 ### ⏳ 规划中
@@ -110,7 +113,12 @@ python -m fastdeploy.entrypoints.openai.api_server  \
        --scheduler-password "scheduler_mtp" \
        --speculative-config '{"method": "mtp", "num_speculative_tokens": 1, "model": ""${path_to_mtp_model}"}'  &
 ```
+## 使用混合MTP、Ngram方法解码
+在启动服务时，只需改动 --speculative-config 即可。例如使用MTP产出两个DraftToken，再额外拼接三个Ngram匹配的DraftToken
+```
+--speculative-config '{"method": "mtp", "num_model_steps": 2, "mtp_strategy": "with_ngram" ,"num_speculative_tokens": 5, "model": "'$model_path'/mtp"}'
 
+```
 ## 🧠 使用 Ngram 解码
 该算法通过 n-gram 窗口从 prompt 和已生成的 Token 中进行匹配生成草稿 Token，适合输入和输出有很大 overlap 的场景，如代码续写、文档查询等。
 > 使用 4×H100；量化方式选择 WINT4