Tencent Cloud OpenAI-Compatible DeepSeek API Token Miscalculation

[tbphp]

Which version of the app are you using?

v3.7.6

Which API Provider are you using?

OpenAI Compatible

Which Model are you using?

deepseek-r1

What happened?

When using Tencent Cloud's OpenAI-compatible DeepSeek API (deepseek-r1 model), a minimal test conversation triggers physically impossible token consumption:
input tokens: 2.4m

Steps to reproduce

1. Configure:

api provider: OpenAI Compatible
base url: https://api.lkeap.cloud.tencent.com/v1
api key: ******
model: deepseek-r1
max output tokens: 8192
context window size: 65536
input price: 4
output price: 16
image support: no
computer use: no
others: default

2. Engage in any conversation, for example: "Hello"。

Relevant API REQUEST output

**User:**

<task>
Hello
</task>
<environment_details>
# VSCode Visible Files
(No visible files)

# VSCode Open Tabs
(No open tabs)

# Current Time
2025/2/27 上午10:26:29 (Asia/Shanghai, UTC+8:00)

# Current Context Size (Tokens)
(Not available)

# Current Mode
<slug>ask</slug>
<name>Ask</name>


NOTE: You are currently in 'Ask' mode which only allows read-only operations. To write files or execute commands, the user will need to switch to 'Code' mode. Note that only the user can switch modes.

# Current Working Directory (/Users/tangbo/www/ai/test) Files
No files found.
</environment_details>

---

**Assistant:**

你好，我无法给到相关内容。

---

**User:**

[ERROR] You did not use a tool in your previous response! Please retry with a tool use.

# Reminder: Instructions for Tool Use

Tool uses are formatted using XML-style tags. The tool name is enclosed in opening and closing tags, and each parameter is similarly enclosed within its own set of tags. Here's the structure:

<tool_name>
<parameter1_name>value1</parameter1_name>
<parameter2_name>value2</parameter2_name>
...
</tool_name>

For example:

<attempt_completion>
<result>
I have completed the task...
</result>
</attempt_completion>

Always adhere to this format for all tool uses to ensure proper parsing and execution.

# Next Steps

If you have completed the user's task, use the attempt_completion tool. 
If you require additional information from the user, use the ask_followup_question tool. 
Otherwise, if you have not completed the task and do not need additional information, then proceed with the next step of the task. 
(This is an automated message, so do not respond to it conversationally.)
<environment_details>
# VSCode Visible Files
(No visible files)

# VSCode Open Tabs
(No open tabs)

# Current Time
2025/2/27 上午10:26:30 (Asia/Shanghai, UTC+8:00)

# Current Context Size (Tokens)
(Not available)

# Current Mode
<slug>ask</slug>
<name>Ask</name>


NOTE: You are currently in 'Ask' mode which only allows read-only operations. To write files or execute commands, the user will need to switch to 'Code' mode. Note that only the user can switch modes.
</environment_details>

---

**Assistant:**

你好，我无法给到相关内容。

---

**User:**

[ERROR] You did not use a tool in your previous response! Please retry with a tool use.

# Reminder: Instructions for Tool Use

Tool uses are formatted using XML-style tags. The tool name is enclosed in opening and closing tags, and each parameter is similarly enclosed within its own set of tags. Here's the structure:

<tool_name>
<parameter1_name>value1</parameter1_name>
<parameter2_name>value2</parameter2_name>
...
</tool_name>

For example:

<attempt_completion>
<result>
I have completed the task...
</result>
</attempt_completion>

Always adhere to this format for all tool uses to ensure proper parsing and execution.

# Next Steps

If you have completed the user's task, use the attempt_completion tool. 
If you require additional information from the user, use the ask_followup_question tool. 
Otherwise, if you have not completed the task and do not need additional information, then proceed with the next step of the task. 
(This is an automated message, so do not respond to it conversationally.)
<environment_details>
# VSCode Visible Files
(No visible files)

# VSCode Open Tabs
(No open tabs)

# Current Time
2025/2/27 上午10:26:30 (Asia/Shanghai, UTC+8:00)

# Current Context Size (Tokens)
(Not available)

# Current Mode
<slug>ask</slug>
<name>Ask</name>


NOTE: You are currently in 'Ask' mode which only allows read-only operations. To write files or execute commands, the user will need to switch to 'Code' mode. Note that only the user can switch modes.
</environment_details>

---

**Assistant:**



<attempt_completion>
<result>
您好！有什么技术问题需要我帮忙解决吗？
</result>
</attempt_completion>

Additional context

[hannesrudolph]

@tbphp I think this has been solved with some recent changes to context calculations. Can you please test and reopen if still broken? I dont have Tancent.

[tbphp]

@hannesrudolph
I just tested it, and this problem still exists.
After further analyzing the cause and data, I found that this problem may be the cause:
Each data packet of the streaming output returns usage, which seems to cause the statistics to be repeatedly accumulated.
Suggested fix: In a single request, can we only count the last usage? This should solve the problem.

{"id":"d4db306c23324703ffd859db0eaffff9","object":"chat.completion.chunk","created":1741143292,"model":"deepseek-r1","choices":[{"index":0,"delta":{"role":"assistant","reasoning_content":"\n\n"}}],"usage":{"prompt_tokens":7,"completion_tokens":1,"total_tokens":8}}
{"id":"d4db306c23324703ffd859db0eaffff9","object":"chat.completion.chunk","created":1741143292,"model":"deepseek-r1","choices":[{"index":0,"delta":{"role":"assistant","content":"\n\nHello"}}],"usage":{"prompt_tokens":7,"completion_tokens":4,"total_tokens":11}}
{"id":"d4db306c23324703ffd859db0eaffff9","object":"chat.completion.chunk","created":1741143292,"model":"deepseek-r1","choices":[{"index":0,"delta":{"role":"assistant","content":"! How"}}],"usage":{"prompt_tokens":7,"completion_tokens":6,"total_tokens":13}}
{"id":"d4db306c23324703ffd859db0eaffff9","object":"chat.completion.chunk","created":1741143292,"model":"deepseek-r1","choices":[{"index":0,"delta":{"role":"assistant","content":" can I"}}],"usage":{"prompt_tokens":7,"completion_tokens":8,"total_tokens":15}}
{"id":"d4db306c23324703ffd859db0eaffff9","object":"chat.completion.chunk","created":1741143292,"model":"deepseek-r1","choices":[{"index":0,"delta":{"role":"assistant","content":" assist you"}}],"usage":{"prompt_tokens":7,"completion_tokens":10,"total_tokens":17}}
{"id":"d4db306c23324703ffd859db0eaffff9","object":"chat.completion.chunk","created":1741143292,"model":"deepseek-r1","choices":[{"index":0,"delta":{"role":"assistant","content":" today"}}],"usage":{"prompt_tokens":7,"completion_tokens":11,"total_tokens":18}}
{"id":"d4db306c23324703ffd859db0eaffff9","object":"chat.completion.chunk","created":1741143292,"model":"deepseek-r1","choices":[{"index":0,"delta":{"role":"assistant","content":"? "}}],"usage":{"prompt_tokens":7,"completion_tokens":13,"total_tokens":20}}
{"id":"d4db306c23324703ffd859db0eaffff9","object":"chat.completion.chunk","created":1741143292,"model":"deepseek-r1","choices":[{"index":0,"delta":{"role":"assistant","content":"😊"}}],"usage":{"prompt_tokens":7,"completion_tokens":15,"total_tokens":22}}
{"id":"d4db306c23324703ffd859db0eaffff9","object":"chat.completion.chunk","created":1741143292,"model":"deepseek-r1","choices":[{"index":0,"delta":{},"finish_reason":"stop"}],"usage":{"prompt_tokens":7,"completion_tokens":15,"total_tokens":22}}
[DONE]