fix llama-cpp-python[server] breaks

ljm625 · ljm625 · commit 559fb334a35e · 2025-01-27T23:50:37.000+08:00
diff --git a/llama_cpp/llama_chat_format.py b/llama_cpp/llama_chat_format.py
@@ -594,7 +594,8 @@ def chat_completion_handler(
             tool_choice=tool_choice,
         )
         prompt = llama.tokenize(
-            result.prompt.encode("utf-8"),
+            vocab=llama.llama_model_get_vocab(model),
+            text=result.prompt.encode("utf-8"),
             add_bos=not result.added_special,
             special=True,
         )

Original file line number	Diff line number	Diff line change
`@@ -594,7 +594,8 @@ def chat_completion_handler(`
`594`	`594`	`tool_choice=tool_choice,`
`595`	`595`	`)`
`596`	`596`	`prompt = llama.tokenize(`
`597`		`- result.prompt.encode("utf-8"),`
	`597`	`+ vocab=llama.llama_model_get_vocab(model),`
	`598`	`+ text=result.prompt.encode("utf-8"),`
`598`	`599`	`add_bos=not result.added_special,`
`599`	`600`	`special=True,`
`600`	`601`	`)`