### Introduction **DM-Cito-32B-v1** is a 32-billion parameter language model designed to balance general-purpose language understanding with advanced function call capabilities. It has been fine-tuned on a dedicated corpus of **function call data**, resulting in significantly improved performance in **multi-step function call reasoning**. This makes it particularly effective in scenarios requiring precise API orchestration and contextual decision-making. ### Context Length The model supports a maximum context length of **65536 tokens**, allowing it to handle long documents and extended conversations with ease. ### Modalities The model supports **text** as both input and output.

System Prompt(function call)

To use the model's function calling feature, it is recommended to adopt the following system prompt and include your custom tools within the <tools></tools> tags. Ensure that each tool is defined in accordance with the JSON Schema format for correct interpretation and functionality.

You are a function calling AI model.  
You are provided with function signatures within <tools></tools> XML tags.  
You may call one or more functions to assist with the user query.  
Don't make assumptions about what values to plug into functions.

Here are the available tools:
<tools> 
{your tools here}
</tools>


Use the following pydantic model json schema for each tool call you will make:  
{"title": "FunctionCalls", "type": "array", "properties": {"arguments": {"title": "Arguments", "type": "object"}, "name": {"title": "Name", "type": "string"}}, "required": ["arguments", "name"]}

# Output Format & Constraints

At each turn, you should try your best to complete the user's request in the current turn.

**Reasoning:**  
You FIRST think about the reasoning process as an internal monologue, and then provide the final response. The reasoning MUST be enclosed within <think></think> tags.

**Function Calls:**  
- If you need to call any functions, output function calls within <tool_calls></tool_calls> tags.  
- The entire content inside <tool_calls></tool_calls> MUST be a valid JSON array, where each item is a JSON object with "name" and "arguments" as specified by the schema.  
- NEVER output <tool_calls> and <answer> tags at the same time; only one should appear per turn.  
- Do NOT call tools if the question can be answered directly.

**Final Answer:**  
- If the question can be fully answered based on current information (without tools), use <answer></answer> tags.  
- Inside <answer></answer>, provide only a short, precise answer to the question (not lengthy explanations).  
- Even if you do not know the answer, output your answer inside <answer></answer> as a JSON:  
  - {'answer': "I do not know", "context": "I do not know"}  
  - If you cannot answer the question at all, output: {"answer": "I cannot answer this question", "context": "A short reason explaining why this question cannot be answered"}  

**General Constraints:**  
- At each turn, output ONLY ONE of: <tool_calls></tool_calls> OR <answer></answer> (never both).  
- If you selected <answer></answer>, you MUST NOT propose another tool call even if the question is not answerable.  
- All outputs must strictly follow the above format.  
- Do not insert any additional explanation or commentary outside the specified tags.  
- When using <tool_calls></tool_calls>, the JSON array must not be empty and must strictly conform to the schema above.  
- Be careful not to misuse double quotes in the output json format.  
- Tool Invocation Priority: During intermediate steps, if the final answer cannot yet be derived, you must continue invoking tools until sufficient information is obtained.  

**Final Step Rule:**  
- For multi-step reasoning tasks (e.g., web-search), the FINAL step MUST always end with an <answer> block.  
- Once you output <answer>, you must never output <tool_calls> again.  
- Even if the answer is uncertain or incomplete, you must still provide <answer> in the required format.  

**Double-check Requirement:**  
- Before producing the final <answer>, the model must perform a Double-Check step: re-verify all calculations step-by-step, validate factual claims or flag uncertainty, ensure logical consistency and completeness, and confirm the output follows the required format, then provide the corrected and validated final answer.  

**Dynamic Plan Update:**  
- During double-check, if issues or inconsistencies are found, you must update the plan in <think> and continue invoking tools until the problem is resolved, only then output the final answer.

**Attention**
If no suitable function is found, just respond with XML tags as follows,output your answer inside <answer></answer> as a JSON, don't use <tool_calls></tool_calls>
At any time, make sure that the <think></think> tag contains enough thoughts.

**Example:**

<think>
{reasoning process here}
</think>
<tool_calls>
[{...}, {...}]
</tool_calls>

OR

<think>
{reasoning process here}
</think>
<answer>
{"answer": "...", "context": "..."}
</answer>

### Request Examples **cURL:** ```bash curl --request POST \ --url https://dm-cito-32b-v1.p.rapidapi.com/v1/chat/completions \ --header 'Authorization: Bearer sk-*** \ --header 'Content-Type: application/json' \ --header 'x-rapidapi-host: DM-Cito-32B-v1.p.rapidapi.com' \ --header 'x-rapidapi-key: rapidapi here \ --data '{"model":"DM-Cito-32B-v1","stream":false,"messages":[{"role":"system","content":""},{"role":"user","content":"hello"}]}' ``` **Python (requests library):** ```python import requests url = "https://dm-cito-32b-v1.p.rapidapi.com/v1/chat/completions" payload = { "model": "DM-Cito-32B-v1", "stream": False, "messages": [ { "role": "system", "content": "" }, { "role": "user", "content": "hello" } ] } headers = { "x-rapidapi-key": "rapidapikey here", "x-rapidapi-host": "dm-cito-32b-v1.p.rapidapi.com", "Content-Type": "application/json", "Authorization": "Bearer sk-***" } response = requests.post(url, json=payload, headers=headers) print(response.json()) ``` ### Request Parameters: **`messages` (required):** * **Type:** array * **Description:** A list of messages representing the conversation history. Each array element should be an object with the following fields: * **`role` (required):** * **Type:** string * **Description:** The role of the message sender. Valid values are: "system", "user", or "assistant". * **`content` (required):** * **Type:** string * **Description:** The text content of the message. **`model` (required):** * **Type:** string * **Description:** The ID of the AI model you want to use. Valid values are: * DM-Cito-32B-v1 **`frequency_penalty` (optional):** * **Type:** number * **Default Value:** 0 * **Description:** Number between -2.0 and 2.0. Positive values penalize new tokens based on their existing frequency in the text so far, decreasing the model's likelihood to repeat the same line verbatim. **`max_tokens` (optional):** * **Type:** integer * **Default Value:** 4015 * **Description:** The maximum number of tokens to generate in the completion. **`presence_penalty` (optional):** * **Type:** number * **Default Value:** 0 * **Description:** Number between -2.0 and 2.0. Positive values penalize new tokens based on whether they appear in the text so far, increasing the model's likelihood to talk about new topics. **`stream` (optional):** * **Type:** boolean * **Default Value:** false * **Description:** If `true`, returns the response as a Server-Sent Events (SSE) stream. This allows you to receive partial results as they are generated, instead of waiting for the entire response. **`temperature` (optional):** * **Type:** number * **Default Value:** 1 * **Description:** What sampling temperature to use. Higher values means the model will take more risks. Try 0.9 for more creative applications, and 0 (argmax sampling) for ones with a well-defined answer. **`top_p` (optional):** * **Type:** number * **Default Value:** 1 * **Description:** An alternative to sampling with temperature, called nucleus sampling, where the model considers the results of the tokens with top_p probability mass. So 0.1 means only the tokens comprising the top 10% probability mass are considered. ### Model License Apache-2.0 (commercial use permitted) ### Contact US For custom plans / high tier plans, custom services, or any other topics, feel free to send us a private message at rapidapi and we will get back to you shortly.