Model API Call Guide

API Information

Chain of Thought (CoT)

CoT refers to the model's ability to generate a series of intermediate reasoning steps when solving complex problems. This capability allows the model not only to provide the final answer but also to demonstrate its reasoning process, thereby enhancing the model's explainability and transparency.

Creating a Chat Request

• Authentication description

  Inference services support API key authentication. The authentication header is in the following format:

  'Authorization': 'Bearer API key of the region where the service is deployed'

• The request and response parameters are as follows.

  Table 2 Request parameters

  Parameter	Mandatory	Default Value	Type	Description
  model	Yes	None	Str	Model to call. For details about the value, see the model parameter in Quickly Calling Preset Models.
  messages	Yes	-	Array	Input question. role shows the role, and content shows the dialog content. Example: "messages": [ {"role": "system","content": "You are a helpful AI assistant."}, {"role": "user","content": "Which number is larger, 9.11 or 9.8?"} ] For more information, see Table 3.
  messages.prefix	No	false	Boolean	Controls whether to enable continuation mode. In this mode, the user provides a message starting with assistant, and the model completes the rest based on that beginning and the input instruction. To use this feature, ensure that the last message in the messages list has the role set to assistant and the prefix parameter set to true. Example: messages = [ {"role": "user", "content": "Write a snippet of Python code"}, {"role": "assistant", "content": "```python\n", "prefix": True} ]
  stream_options	No	None	Object	Specifies whether to display the number of used tokens during streaming output. This parameter is only valid when stream is set to True. You need to set stream_options to {"include_usage": true} to print the number of tokens used. For more information, see Table 4.
  max_tokens	No	None	Int	Maximum number of tokens that can be generated for the current task, including tokens generated by the model and reasoning tokens for deep thinking.
  top_k	No	-1	Int	The candidate set size determines the sampling range during generation. For example, setting it to 50 means only the top 50 scoring tokens are sampled at each step. A larger size increases randomness; a smaller one makes the output more predictable.
  top_p	No	1.0	Float	Nucleus sampling. It keeps only the words with combined probabilities above the threshold p and removes the rest. These selected words are then normalized and sampled again. Lower settings reduce word options, making outputs focused and cautious. Higher settings expand word choices, creating varied and creative outputs. Adjust either temperature or top_p separately for best results, not both at once. Value range: 0 to 1. The value 1 indicates that all tokens are considered.
  temperature	No	1.0	Float	Model sampling temperature. The higher the value, the more random the model output; the lower the value, the more deterministic the output. Adjust either temperature or top_p separately for best results, not both at once.
  stop	No	None	None/Str/List	A list of strings used to stop generation. The output does not contain the stop strings. For example, if the value is set to ["You," "Good"], text generation will stop once either You or Good is reached.
  stream	No	False	Bool	Controls whether to enable streaming inference. The default value is False, indicating that streaming inference is disabled.
  n	No	1	Int	Number of responses generated for each input message. If beam_search is not used, the recommended value range of n is 1 ≤ n ≤10. If n is greater than 1, ensure that greedy_sample is not used for sampling, that is, top_k is greater than 1 and temperature is greater than 0. If beam_search is used, the recommended value range of n is 1 < n ≤ 10. If n is 1, the inference request will fail. NOTE: For optimal performance, keep n at 10 or below. Large values of n can significantly slow down processing. Inadequate video RAM may cause inference requests to fail.
  use_beam_search	No	False	Bool	Controls whether to use beam_search to replace sampling. When this parameter is used, the following parameters must be configured as required: n: > 1 top_p: 1.0 top_k: -1 temperature: 0.0
  presence_penalty	No	0.0	Float	Applies rewards or penalties based on the presence of new words in the generated text. The value range is [-2.0,2.0].
  frequency_penalty	No	0.0	Float	Applies rewards or penalties based on the frequency of each word in the generated text. The value range is [-2.0,2.0].
  length_penalty	No	1.0	Float	Imposes a larger penalty on longer sequences in a beam search process. When this parameter is used, the following parameters must be configured as required: top_k: -1 use_beam_search: true best_of: > 1
  chat_template_kwargs.thinking	No	false	Bool	Specifies whether to enable the CoT. For some models, you can enable the CoT by adding the template parameter "thinking": true when initiating an inference request. Example of enabling the CoT: { "model": "DeepSeek-V31", "messages": [{ "role": "system", "content": "You are a helpful assistant." }, { "role": "user", "content": "Hello" }], "chat_template_kwargs": { "thinking": true } }
  chat_template_kwargs.enable_thinking	No	true	Bool	Specifies whether to enable the CoT. For some models, you can disable the CoT by adding the template parameter "enable_thinking": false when initiating an inference request. Example of disabling the CoT: { "model": "qwen3-32b", "messages": [{ "role": "system", "content": "You are a helpful assistant." }, { "role": "user", "content": "Hello" }], "chat_template_kwargs": { "enable_thinking": false } }

  Table 3 Request parameter messages

  Parameter	Mandatory	Default Value	Type	Description
  role	Yes	None	Str	Different roles correspond to different message types. system: developer-entered instructions like response formats and roles for the model to follow. user: user-entered messages including prompts and context information. assistant: responses generated by the model. tool: information returned by the tool when the model calls it.
  content	Yes	None	Str	When role is set to system, this parameter indicates the AI model's personality. { "role": "system", "content": "You are a helpful AI assistant." } When role is set to user, this parameter indicates the question asked by the user. { "role": "user", "content": "Which one is larger, 9.11 or 9.8?" } When role is set to assistant, this parameter indicates the content output by the AI model. {"role": "assistant","content": "9.11 is larger than 9.8."} When role is set to tool, this parameter indicates the responses returned by the tool when the model calls it. {"role": "tool", "content": "The weather in Shanghai is sunny today. The temperature is 10°C."}

  Table 4 Request parameter stream_options

  Parameter	Mandatory	Default Value	Type	Description
  include_usage	No	true	Bool	Specifies whether the streaming response outputs token usage information. true: Each chunk outputs a usage field that shows the total token usage. false: The token usage is not displayed.

  Table 5 Response parameters

  Parameter	Type	Description
  id	Str	Unique ID of the request.
  object	Str	chat.completion type: Multi-turn dialogs are returned.
  created	Int	Timestamp.
  model	Str	Model to call.
  choices	Array	Model output, including the index and message parameters. In message: content is the model's final reply. reasoning content is the model's deep thinking content (for DeepSeek models only).
  usage	Object	Statistics on tokens consumed by the request: This parameter is returned by default for non-streaming requests. This parameter is returned by default for streaming requests. Each chunk outputs a usage field that shows the token usage. Parameters: prompt tokens: number of input tokens. completion tokens: number of output tokens. total tokens: total number of tokens.
  prompt_logprobs	Float	Log probability. You can use this to measure the model's confidence in its output or to explore other options the model provides.

Qwen3-235B Text Generation Request Example

This section demonstrates the basic usage of a text generation model, using the Qwen3-235B model as an example to request a text response via a Python script, cURL command, or OpenAI SDK.

• Python request example:

  import requests
  import json
  
  if __name__ == '__main__':
      url = ""https://***/v2/infer/***/v1/chat/completions""  # API address
      api_key = "API_KEY"  # Replace API_KEY with the obtained API key.
  
      # Send request.
      headers = {
          'Content-Type': 'application/json',
          'Authorization': f'Bearer {api_key}'
      }
      data = {
          "model": "qwen3-235b,"  # Model name
          "messages": [
              {"role": "system", "content": "You are a helpful assistant."},
              {"role": "user", "content": "Hello"}
          ],
          "chat_template_kwargs": {
              "thinking": True  # Specifies whether to enable deep thinking. It is disabled by default.
          }
      }
      response = requests.post(url, headers=headers, data=json.dumps(data), verify=False)
  
      # Print result.
      print(response.status_code)
      print(response.text)

• cURL request example:

  curl -X POST "https://***/v2/infer/***/v1/chat/completions" \
    -H "Content-Type: application/json" \
    -H "Authorization: Bearer $API_KEY" \
    -d '{
      "model": "qwen3-235b",
      "messages": [
        {"role": "system", "content": "You are a helpful assistant."},
        {"role": "user", "content": "Hello"}
      ],
       "chat_template_kwargs": {
         "thinking": true
       }
    }'

• OpenAI SDK request example:

  from openai import OpenAI
  
  base_url = "https://***/v2/infer/***/v1/chat/completions"  # API address
  api_key = "API_KEY"  # Replace API_KEY with the obtained API key.
  
  client = OpenAI(api_key=api_key, base_url=base_url)
  
  response = client.chat.completions.create(
      model="qwen3-235b",  # Model name
      messages=[
          {"role": "system", "content": "You are a helpful assistant"},
          {"role": "user", "content": "Hello"},
      ],
      extra_body={
          "chat_template_kwargs": {
              "thinking": True  # Specifies whether to enable deep thinking. It is disabled by default.
          }
      }
  )
  
  print(response.choices[0].message.content)

Response Example

The following is an example response of the text generation model.

{
    "id":"chat-71406e38b0d248c9b284709f8435****",
    "object":"chat.completion",
    "created":1740809549,
    "model":"qwen3-235b",
    "choices":[
        {
            "index":0,
            "message":{
                "role":"assistant",
                "content":"\n\n Compare 9.11 and 9.8.:\n\n1. **Compare the integer part**: The integer part of both is 9, which is equal.\n2. **Compare the tenths place**:\n - The tenths place of 9.11 is **1**\n - 9.8 can be considered as 9.80, and its tenths place is **8**\n - **8 > 1**, so 9.8 is larger.\n\n**Conclusion**:\n**9.8 > 9.11**\n(When comparing decimals, line up the digits and compare them directly.)",
                "reasoning_content": "Well, I now need to compare 9.11 and 9.8 which is larger. First of all, I have to recall the method of comparing decimals. When comparing decimals, start by comparing the integer parts. If the integer parts are the same, compare the tenths and hundredths of the decimal parts in sequence until the larger number is determined. \n\n The integer parts of the two numbers are both 9, so they are the same. Next, compare the tenths. The tenth digit of 9.11 is 1, while the tenth digit of 9.8 is 8. This can be problematic, as some people might directly treat 9.8 as 9.80, or focus on comparing the digits in the tenths place. \n\n Now, comparing the tenths place, 9.8 has an 8, while 9.11 has a 1. Clearly, 8 is greater than 1. So, should we conclude that 9.8 is greater than 9.11? \n\n However, it is important to note that some people might incorrectly assume that the more decimal places a number has, the larger its value. But this is not true; for instance, 0.9 is greater than 0.8999. Thus, having more decimal places does not necessarily mean a larger value. \n\n Additionally, the decimal parts of the two numbers can be aligned to have the same number of digits for comparison. For instance, 9.8 can be written as 9.80, where the tenths place is 8 and the hundredths place is 0. On the other hand, for 9.11, the tenths place is 1 and the hundredths place is 1. Since 8 in the tenths place is greater than 1, 9.80 (which is 9.8) is greater than 9.11. \n\n Therefore, the final conclusion is that 9.8 is larger than 9.11.\n",
                "tool_calls":[]
                },
                "logprobs":null,
                "finish_reason":"stop",
                "stop_reason":null        
        }
    ],
    "usage":{
        "prompt_tokens":21,
        "total_tokens":437,
        "completion_tokens":416
        },
    "prompt_logprobs":null
}