rig/providers/huggingface/

completion.rs

use super::client::Client;
use crate::completion::GetTokenUsage;
use crate::http_client::HttpClientExt;
use crate::providers::openai::StreamingCompletionResponse;
use crate::telemetry::SpanCombinator;
use crate::{
    OneOrMany,
    completion::{self, CompletionError, CompletionRequest},
    json_utils,
    message::{self},
    one_or_many::string_or_one_or_many,
};
use serde::{Deserialize, Deserializer, Serialize, Serializer};
use serde_json::Value;
use std::{convert::Infallible, str::FromStr};
use tracing::{Level, enabled, info_span};
use tracing_futures::Instrument;

#[derive(Debug, Deserialize)]
#[serde(untagged)]
pub enum ApiResponse<T> {
    Ok(T),
    Err(Value),
}

// ================================================================
// Huggingface Completion API
// ================================================================

// Conversational LLMs
/// `google/gemma-2-2b-it` completion model
pub const GEMMA_2: &str = "google/gemma-2-2b-it";
/// `meta-llama/Meta-Llama-3.1-8B-Instruct` completion model
pub const META_LLAMA_3_1: &str = "meta-llama/Meta-Llama-3.1-8B-Instruct";
/// `PowerInfer/SmallThinker-3B-Preview` completion model
pub const SMALLTHINKER_PREVIEW: &str = "PowerInfer/SmallThinker-3B-Preview";
/// `Qwen/Qwen2.5-7B-Instruct` completion model
pub const QWEN2_5: &str = "Qwen/Qwen2.5-7B-Instruct";
/// `Qwen/Qwen2.5-Coder-32B-Instruct` completion model
pub const QWEN2_5_CODER: &str = "Qwen/Qwen2.5-Coder-32B-Instruct";

// Conversational VLMs

/// `Qwen/Qwen2-VL-7B-Instruct` visual-language completion model
pub const QWEN2_VL: &str = "Qwen/Qwen2-VL-7B-Instruct";
/// `Qwen/QVQ-72B-Preview` visual-language completion model
pub const QWEN_QVQ_PREVIEW: &str = "Qwen/QVQ-72B-Preview";

#[derive(Debug, Deserialize, Serialize, PartialEq, Clone)]
pub struct Function {
    name: String,
    #[serde(
        serialize_with = "json_utils::stringified_json::serialize",
        deserialize_with = "json_utils::stringified_json::deserialize_maybe_stringified"
    )]
    pub arguments: serde_json::Value,
}

impl From<Function> for message::ToolFunction {
    fn from(value: Function) -> Self {
        message::ToolFunction {
            name: value.name,
            arguments: value.arguments,
        }
    }
}

#[derive(Default, Debug, Serialize, Deserialize, PartialEq, Clone)]
#[serde(rename_all = "lowercase")]
pub enum ToolType {
    #[default]
    Function,
}

#[derive(Debug, Deserialize, Serialize, Clone)]
pub struct ToolDefinition {
    pub r#type: String,
    pub function: completion::ToolDefinition,
}

impl From<completion::ToolDefinition> for ToolDefinition {
    fn from(tool: completion::ToolDefinition) -> Self {
        Self {
            r#type: "function".into(),
            function: tool,
        }
    }
}

#[derive(Debug, Deserialize, Serialize, PartialEq, Clone)]
pub struct ToolCall {
    pub id: String,
    pub r#type: ToolType,
    pub function: Function,
}

impl From<ToolCall> for message::ToolCall {
    fn from(value: ToolCall) -> Self {
        message::ToolCall {
            id: value.id,
            call_id: None,
            function: value.function.into(),
            signature: None,
            additional_params: None,
        }
    }
}

impl From<message::ToolCall> for ToolCall {
    fn from(value: message::ToolCall) -> Self {
        ToolCall {
            id: value.id,
            r#type: ToolType::Function,
            function: Function {
                name: value.function.name,
                arguments: value.function.arguments,
            },
        }
    }
}

#[derive(Debug, Deserialize, Serialize, PartialEq, Clone)]
pub struct ImageUrl {
    url: String,
}

#[derive(Debug, Deserialize, Serialize, PartialEq, Clone)]
#[serde(tag = "type", rename_all = "lowercase")]
pub enum UserContent {
    Text {
        text: String,
    },
    #[serde(rename = "image_url")]
    ImageUrl {
        image_url: ImageUrl,
    },
}

impl FromStr for UserContent {
    type Err = Infallible;

    fn from_str(s: &str) -> Result<Self, Self::Err> {
        Ok(UserContent::Text {
            text: s.to_string(),
        })
    }
}

#[derive(Debug, Deserialize, Serialize, PartialEq, Clone)]
#[serde(tag = "type", rename_all = "lowercase")]
pub enum AssistantContent {
    Text { text: String },
}

impl FromStr for AssistantContent {
    type Err = Infallible;

    fn from_str(s: &str) -> Result<Self, Self::Err> {
        Ok(AssistantContent::Text {
            text: s.to_string(),
        })
    }
}

#[derive(Debug, Deserialize, Serialize, PartialEq, Clone)]
#[serde(tag = "type", rename_all = "lowercase")]
pub enum SystemContent {
    Text { text: String },
}

impl FromStr for SystemContent {
    type Err = Infallible;

    fn from_str(s: &str) -> Result<Self, Self::Err> {
        Ok(SystemContent::Text {
            text: s.to_string(),
        })
    }
}

impl From<UserContent> for message::UserContent {
    fn from(value: UserContent) -> Self {
        match value {
            UserContent::Text { text } => message::UserContent::text(text),
            UserContent::ImageUrl { image_url } => {
                message::UserContent::image_url(image_url.url, None, None)
            }
        }
    }
}

impl TryFrom<message::UserContent> for UserContent {
    type Error = message::MessageError;

    fn try_from(content: message::UserContent) -> Result<Self, Self::Error> {
        match content {
            message::UserContent::Text(text) => Ok(UserContent::Text { text: text.text }),
            message::UserContent::Document(message::Document {
                data: message::DocumentSourceKind::Raw(raw),
                ..
            }) => {
                let text = String::from_utf8_lossy(raw.as_slice()).into();
                Ok(UserContent::Text { text })
            }
            message::UserContent::Document(message::Document {
                data:
                    message::DocumentSourceKind::Base64(text)
                    | message::DocumentSourceKind::String(text),
                ..
            }) => Ok(UserContent::Text { text }),
            message::UserContent::Image(message::Image { data, .. }) => match data {
                message::DocumentSourceKind::Url(url) => Ok(UserContent::ImageUrl {
                    image_url: ImageUrl { url },
                }),
                _ => Err(message::MessageError::ConversionError(
                    "Huggingface only supports images as urls".into(),
                )),
            },
            _ => Err(message::MessageError::ConversionError(
                "Huggingface only supports text and images".into(),
            )),
        }
    }
}

#[derive(Debug, Deserialize, Serialize, PartialEq, Clone)]
#[serde(tag = "role", rename_all = "lowercase")]
pub enum Message {
    System {
        #[serde(deserialize_with = "string_or_one_or_many")]
        content: OneOrMany<SystemContent>,
    },
    User {
        #[serde(deserialize_with = "string_or_one_or_many")]
        content: OneOrMany<UserContent>,
    },
    Assistant {
        #[serde(default, deserialize_with = "json_utils::string_or_vec")]
        content: Vec<AssistantContent>,
        #[serde(default, deserialize_with = "json_utils::null_or_vec")]
        tool_calls: Vec<ToolCall>,
    },
    #[serde(rename = "tool", alias = "Tool")]
    ToolResult {
        name: String,
        #[serde(skip_serializing_if = "Option::is_none")]
        arguments: Option<serde_json::Value>,
        #[serde(
            deserialize_with = "string_or_one_or_many",
            serialize_with = "serialize_tool_content"
        )]
        content: OneOrMany<String>,
    },
}

fn serialize_tool_content<S>(content: &OneOrMany<String>, serializer: S) -> Result<S::Ok, S::Error>
where
    S: Serializer,
{
    // OpenAI-compatible APIs expect tool content as a string, not an array
    let joined = content
        .iter()
        .map(String::as_str)
        .collect::<Vec<_>>()
        .join("\n");
    serializer.serialize_str(&joined)
}

impl Message {
    pub fn system(content: &str) -> Self {
        Message::System {
            content: OneOrMany::one(SystemContent::Text {
                text: content.to_string(),
            }),
        }
    }
}

impl TryFrom<message::Message> for Vec<Message> {
    type Error = message::MessageError;

    fn try_from(message: message::Message) -> Result<Vec<Message>, Self::Error> {
        match message {
            message::Message::System { content } => Ok(vec![Message::system(&content)]),
            message::Message::User { content } => {
                let (tool_results, other_content): (Vec<_>, Vec<_>) = content
                    .into_iter()
                    .partition(|content| matches!(content, message::UserContent::ToolResult(_)));

                if !tool_results.is_empty() {
                    tool_results
                        .into_iter()
                        .map(|content| match content {
                            message::UserContent::ToolResult(message::ToolResult {
                                id,
                                content,
                                ..
                            }) => Ok::<_, message::MessageError>(Message::ToolResult {
                                name: id,
                                arguments: None,
                                content: content.try_map(|content| match content {
                                    message::ToolResultContent::Text(message::Text { text }) => {
                                        Ok(text)
                                    }
                                    _ => Err(message::MessageError::ConversionError(
                                        "Tool result content does not support non-text".into(),
                                    )),
                                })?,
                            }),
                            _ => unreachable!(),
                        })
                        .collect::<Result<Vec<_>, _>>()
                } else {
                    let other_content = OneOrMany::many(other_content).expect(
                        "There must be other content here if there were no tool result content",
                    );

                    Ok(vec![Message::User {
                        content: other_content.try_map(|content| match content {
                            message::UserContent::Text(text) => {
                                Ok(UserContent::Text { text: text.text })
                            }
                            message::UserContent::Image(image) => {
                                let url = image.try_into_url()?;

                                Ok(UserContent::ImageUrl {
                                    image_url: ImageUrl { url },
                                })
                            }
                            message::UserContent::Document(message::Document {
                                data: message::DocumentSourceKind::Raw(raw), ..
                            }) => {
                                let text = String::from_utf8_lossy(raw.as_slice()).into();
                                Ok(UserContent::Text { text })
                            }
                            message::UserContent::Document(message::Document {
                                data: message::DocumentSourceKind::Base64(text) | message::DocumentSourceKind::String(text), ..
                            }) => {
                                Ok(UserContent::Text { text })
                            }
                            _ => Err(message::MessageError::ConversionError(
                                "Huggingface inputs only support text and image URLs (both base64-encoded images and regular URLs)".into(),
                            )),
                        })?,
                    }])
                }
            }
            message::Message::Assistant { content, .. } => {
                let mut text_content = Vec::new();
                let mut tool_calls = Vec::new();

                for content in content {
                    match content {
                        message::AssistantContent::Text(text) => text_content.push(text),
                        message::AssistantContent::ToolCall(tool_call) => {
                            tool_calls.push(tool_call)
                        }
                        message::AssistantContent::Reasoning(_) => {
                            // HuggingFace does not support assistant-history reasoning items.
                            // Silently skip unsupported reasoning content.
                        }
                        message::AssistantContent::Image(_) => {
                            panic!("Image content is not supported on HuggingFace via Rig");
                        }
                    }
                }

                if text_content.is_empty() && tool_calls.is_empty() {
                    return Ok(vec![]);
                }

                Ok(vec![Message::Assistant {
                    content: text_content
                        .into_iter()
                        .map(|content| AssistantContent::Text { text: content.text })
                        .collect::<Vec<_>>(),
                    tool_calls: tool_calls
                        .into_iter()
                        .map(|tool_call| tool_call.into())
                        .collect::<Vec<_>>(),
                }])
            }
        }
    }
}

impl TryFrom<Message> for message::Message {
    type Error = message::MessageError;

    fn try_from(message: Message) -> Result<Self, Self::Error> {
        Ok(match message {
            Message::User { content, .. } => message::Message::User {
                content: content.map(|content| content.into()),
            },
            Message::Assistant {
                content,
                tool_calls,
                ..
            } => {
                let mut content = content
                    .into_iter()
                    .map(|content| match content {
                        AssistantContent::Text { text } => message::AssistantContent::text(text),
                    })
                    .collect::<Vec<_>>();

                content.extend(
                    tool_calls
                        .into_iter()
                        .map(|tool_call| Ok(message::AssistantContent::ToolCall(tool_call.into())))
                        .collect::<Result<Vec<_>, _>>()?,
                );

                message::Message::Assistant {
                    id: None,
                    content: OneOrMany::many(content).map_err(|_| {
                        message::MessageError::ConversionError(
                            "Neither `content` nor `tool_calls` was provided to the Message"
                                .to_owned(),
                        )
                    })?,
                }
            }

            Message::ToolResult { name, content, .. } => message::Message::User {
                content: OneOrMany::one(message::UserContent::tool_result(
                    name,
                    content.map(message::ToolResultContent::text),
                )),
            },

            // System messages should get stripped out when converting message's, this is just a
            // stop gap to avoid obnoxious error handling or panic occurring.
            Message::System { content, .. } => message::Message::User {
                content: content.map(|c| match c {
                    SystemContent::Text { text } => message::UserContent::text(text),
                }),
            },
        })
    }
}

#[derive(Clone, Debug, Deserialize, Serialize)]
pub struct Choice {
    pub finish_reason: String,
    pub index: usize,
    #[serde(default)]
    pub logprobs: serde_json::Value,
    pub message: Message,
}

#[derive(Debug, Deserialize, Clone, Serialize)]
pub struct Usage {
    pub completion_tokens: i32,
    pub prompt_tokens: i32,
    pub total_tokens: i32,
}

impl GetTokenUsage for Usage {
    fn token_usage(&self) -> Option<crate::completion::Usage> {
        let mut usage = crate::completion::Usage::new();
        usage.input_tokens = self.prompt_tokens as u64;
        usage.output_tokens = self.completion_tokens as u64;
        usage.total_tokens = self.total_tokens as u64;

        Some(usage)
    }
}

#[derive(Clone, Debug, Deserialize, Serialize)]
pub struct CompletionResponse {
    pub created: i32,
    pub id: String,
    pub model: String,
    pub choices: Vec<Choice>,
    #[serde(default, deserialize_with = "default_string_on_null")]
    pub system_fingerprint: String,
    pub usage: Usage,
}

impl crate::telemetry::ProviderResponseExt for CompletionResponse {
    type OutputMessage = Choice;
    type Usage = Usage;

    fn get_response_id(&self) -> Option<String> {
        Some(self.id.clone())
    }

    fn get_response_model_name(&self) -> Option<String> {
        Some(self.model.clone())
    }

    fn get_output_messages(&self) -> Vec<Self::OutputMessage> {
        self.choices.clone()
    }

    fn get_text_response(&self) -> Option<String> {
        let text_response = self
            .choices
            .iter()
            .filter_map(|x| {
                let Message::User { ref content } = x.message else {
                    return None;
                };

                let text = content
                    .iter()
                    .filter_map(|x| {
                        if let UserContent::Text { text } = x {
                            Some(text.clone())
                        } else {
                            None
                        }
                    })
                    .collect::<Vec<String>>();

                if text.is_empty() {
                    None
                } else {
                    Some(text.join("\n"))
                }
            })
            .collect::<Vec<String>>()
            .join("\n");

        if text_response.is_empty() {
            None
        } else {
            Some(text_response)
        }
    }

    fn get_usage(&self) -> Option<Self::Usage> {
        Some(self.usage.clone())
    }
}

fn default_string_on_null<'de, D>(deserializer: D) -> Result<String, D::Error>
where
    D: Deserializer<'de>,
{
    match Option::<String>::deserialize(deserializer)? {
        Some(value) => Ok(value),      // Use provided value
        None => Ok(String::default()), // Use `Default` implementation
    }
}

impl TryFrom<CompletionResponse> for completion::CompletionResponse<CompletionResponse> {
    type Error = CompletionError;

    fn try_from(response: CompletionResponse) -> Result<Self, Self::Error> {
        let choice = response.choices.first().ok_or_else(|| {
            CompletionError::ResponseError("Response contained no choices".to_owned())
        })?;

        let content = match &choice.message {
            Message::Assistant {
                content,
                tool_calls,
                ..
            } => {
                let mut content = content
                    .iter()
                    .map(|c| match c {
                        AssistantContent::Text { text } => message::AssistantContent::text(text),
                    })
                    .collect::<Vec<_>>();

                content.extend(
                    tool_calls
                        .iter()
                        .map(|call| {
                            completion::AssistantContent::tool_call(
                                &call.id,
                                &call.function.name,
                                call.function.arguments.clone(),
                            )
                        })
                        .collect::<Vec<_>>(),
                );
                Ok(content)
            }
            _ => Err(CompletionError::ResponseError(
                "Response did not contain a valid message or tool call".into(),
            )),
        }?;

        let choice = OneOrMany::many(content).map_err(|_| {
            CompletionError::ResponseError(
                "Response contained no message or tool call (empty)".to_owned(),
            )
        })?;

        let usage = completion::Usage {
            input_tokens: response.usage.prompt_tokens as u64,
            output_tokens: response.usage.completion_tokens as u64,
            total_tokens: response.usage.total_tokens as u64,
            cached_input_tokens: 0,
            cache_creation_input_tokens: 0,
        };

        Ok(completion::CompletionResponse {
            choice,
            usage,
            raw_response: response,
            message_id: None,
        })
    }
}

#[derive(Debug, Serialize, Deserialize)]
pub(super) struct HuggingfaceCompletionRequest {
    model: String,
    pub messages: Vec<Message>,
    #[serde(skip_serializing_if = "Option::is_none")]
    temperature: Option<f64>,
    #[serde(skip_serializing_if = "Vec::is_empty")]
    tools: Vec<ToolDefinition>,
    #[serde(skip_serializing_if = "Option::is_none")]
    tool_choice: Option<crate::providers::openai::completion::ToolChoice>,
    #[serde(flatten, skip_serializing_if = "Option::is_none")]
    pub additional_params: Option<serde_json::Value>,
}

impl TryFrom<(&str, CompletionRequest)> for HuggingfaceCompletionRequest {
    type Error = CompletionError;

    fn try_from((model, req): (&str, CompletionRequest)) -> Result<Self, Self::Error> {
        if req.output_schema.is_some() {
            tracing::warn!("Structured outputs currently not supported for Huggingface");
        }
        let model = req.model.clone().unwrap_or_else(|| model.to_string());
        let mut full_history: Vec<Message> = match &req.preamble {
            Some(preamble) => vec![Message::system(preamble)],
            None => vec![],
        };
        if let Some(docs) = req.normalized_documents() {
            let docs: Vec<Message> = docs.try_into()?;
            full_history.extend(docs);
        }

        let chat_history: Vec<Message> = req
            .chat_history
            .clone()
            .into_iter()
            .map(|message| message.try_into())
            .collect::<Result<Vec<Vec<Message>>, _>>()?
            .into_iter()
            .flatten()
            .collect();

        full_history.extend(chat_history);

        if full_history.is_empty() {
            return Err(CompletionError::RequestError(
                std::io::Error::new(
                    std::io::ErrorKind::InvalidInput,
                    "HuggingFace request has no provider-compatible messages after conversion",
                )
                .into(),
            ));
        }

        let tool_choice = req
            .tool_choice
            .clone()
            .map(crate::providers::openai::completion::ToolChoice::try_from)
            .transpose()?;

        Ok(Self {
            model: model.to_string(),
            messages: full_history,
            temperature: req.temperature,
            tools: req
                .tools
                .clone()
                .into_iter()
                .map(ToolDefinition::from)
                .collect::<Vec<_>>(),
            tool_choice,
            additional_params: req.additional_params,
        })
    }
}

#[derive(Clone)]
pub struct CompletionModel<T = reqwest::Client> {
    pub(crate) client: Client<T>,
    /// Name of the model (e.g: google/gemma-2-2b-it)
    pub model: String,
}

impl<T> CompletionModel<T> {
    pub fn new(client: Client<T>, model: &str) -> Self {
        Self {
            client,
            model: model.to_string(),
        }
    }
}

impl<T> completion::CompletionModel for CompletionModel<T>
where
    T: HttpClientExt + Clone + 'static,
{
    type Response = CompletionResponse;
    type StreamingResponse = StreamingCompletionResponse;

    type Client = Client<T>;

    fn make(client: &Self::Client, model: impl Into<String>) -> Self {
        Self::new(client.clone(), &model.into())
    }

    async fn completion(
        &self,
        completion_request: CompletionRequest,
    ) -> Result<completion::CompletionResponse<CompletionResponse>, CompletionError> {
        let request_model = completion_request
            .model
            .clone()
            .unwrap_or_else(|| self.model.clone());
        let span = if tracing::Span::current().is_disabled() {
            info_span!(
                target: "rig::completions",
                "chat",
                gen_ai.operation.name = "chat",
                gen_ai.provider.name = "huggingface",
                gen_ai.request.model = &request_model,
                gen_ai.system_instructions = &completion_request.preamble,
                gen_ai.response.id = tracing::field::Empty,
                gen_ai.response.model = tracing::field::Empty,
                gen_ai.usage.output_tokens = tracing::field::Empty,
                gen_ai.usage.input_tokens = tracing::field::Empty,
                gen_ai.usage.cached_tokens = tracing::field::Empty,
            )
        } else {
            tracing::Span::current()
        };

        let model = self.client.subprovider().model_identifier(&request_model);
        let request = HuggingfaceCompletionRequest::try_from((model.as_ref(), completion_request))?;

        if enabled!(Level::TRACE) {
            tracing::trace!(
                target: "rig::completions",
                "Huggingface completion request: {}",
                serde_json::to_string_pretty(&request)?
            );
        }

        let request = serde_json::to_vec(&request)?;

        let path = self
            .client
            .subprovider()
            .completion_endpoint(&request_model);
        let request = self
            .client
            .post(&path)?
            .header("Content-Type", "application/json")
            .body(request)
            .map_err(|e| CompletionError::HttpError(e.into()))?;

        async move {
            let response = self.client.send(request).await?;

            if response.status().is_success() {
                let bytes: Vec<u8> = response.into_body().await?;
                let text = String::from_utf8_lossy(&bytes);

                tracing::debug!(target: "rig", "Huggingface completion error: {}", text);

                match serde_json::from_slice::<ApiResponse<CompletionResponse>>(&bytes)? {
                    ApiResponse::Ok(response) => {
                        if enabled!(Level::TRACE) {
                            tracing::trace!(
                                target: "rig::completions",
                                "Huggingface completion response: {}",
                                serde_json::to_string_pretty(&response)?
                            );
                        }

                        let span = tracing::Span::current();
                        span.record_token_usage(&response.usage);
                        span.record_response_metadata(&response);

                        response.try_into()
                    }
                    ApiResponse::Err(err) => Err(CompletionError::ProviderError(err.to_string())),
                }
            } else {
                let status = response.status();
                let text: Vec<u8> = response.into_body().await?;
                let text: String = String::from_utf8_lossy(&text).into();

                Err(CompletionError::ProviderError(format!(
                    "{}: {}",
                    status, text
                )))
            }
        }
        .instrument(span)
        .await
    }

    async fn stream(
        &self,
        request: CompletionRequest,
    ) -> Result<
        crate::streaming::StreamingCompletionResponse<Self::StreamingResponse>,
        CompletionError,
    > {
        CompletionModel::stream(self, request).await
    }
}

#[cfg(test)]
mod tests {
    use super::*;
    use serde_path_to_error::deserialize;

    #[test]
    fn test_huggingface_request_uses_request_model_override() {
        let request = CompletionRequest {
            model: Some("meta-llama/Meta-Llama-3.1-8B-Instruct".to_string()),
            preamble: None,
            chat_history: crate::OneOrMany::one("Hello".into()),
            documents: vec![],
            tools: vec![],
            temperature: None,
            max_tokens: None,
            tool_choice: None,
            additional_params: None,
            output_schema: None,
        };

        let hf_request = HuggingfaceCompletionRequest::try_from(("mistralai/Mistral-7B", request))
            .expect("request conversion should succeed");
        let serialized = serde_json::to_value(hf_request).expect("serialization should succeed");

        assert_eq!(serialized["model"], "meta-llama/Meta-Llama-3.1-8B-Instruct");
    }

    #[test]
    fn test_huggingface_request_uses_default_model_when_override_unset() {
        let request = CompletionRequest {
            model: None,
            preamble: None,
            chat_history: crate::OneOrMany::one("Hello".into()),
            documents: vec![],
            tools: vec![],
            temperature: None,
            max_tokens: None,
            tool_choice: None,
            additional_params: None,
            output_schema: None,
        };

        let hf_request = HuggingfaceCompletionRequest::try_from(("mistralai/Mistral-7B", request))
            .expect("request conversion should succeed");
        let serialized = serde_json::to_value(hf_request).expect("serialization should succeed");

        assert_eq!(serialized["model"], "mistralai/Mistral-7B");
    }

    #[test]
    fn test_deserialize_message() {
        let assistant_message_json = r#"
        {
            "role": "assistant",
            "content": "\n\nHello there, how may I assist you today?"
        }
        "#;

        let assistant_message_json2 = r#"
        {
            "role": "assistant",
            "content": [
                {
                    "type": "text",
                    "text": "\n\nHello there, how may I assist you today?"
                }
            ],
            "tool_calls": null
        }
        "#;

        let assistant_message_json3 = r#"
        {
            "role": "assistant",
            "tool_calls": [
                {
                    "id": "call_h89ipqYUjEpCPI6SxspMnoUU",
                    "type": "function",
                    "function": {
                        "name": "subtract",
                        "arguments": {"x": 2, "y": 5}
                    }
                }
            ],
            "content": null,
            "refusal": null
        }
        "#;

        let user_message_json = r#"
        {
            "role": "user",
            "content": [
                {
                    "type": "text",
                    "text": "What's in this image?"
                },
                {
                    "type": "image_url",
                    "image_url": {
                        "url": "https://upload.wikimedia.org/wikipedia/commons/thumb/d/dd/Gfp-wisconsin-madison-the-nature-boardwalk.jpg/2560px-Gfp-wisconsin-madison-the-nature-boardwalk.jpg"
                    }
                }
            ]
        }
        "#;

        let assistant_message: Message = {
            let jd = &mut serde_json::Deserializer::from_str(assistant_message_json);
            deserialize(jd).unwrap_or_else(|err| {
                panic!(
                    "Deserialization error at {} ({}:{}): {}",
                    err.path(),
                    err.inner().line(),
                    err.inner().column(),
                    err
                );
            })
        };

        let assistant_message2: Message = {
            let jd = &mut serde_json::Deserializer::from_str(assistant_message_json2);
            deserialize(jd).unwrap_or_else(|err| {
                panic!(
                    "Deserialization error at {} ({}:{}): {}",
                    err.path(),
                    err.inner().line(),
                    err.inner().column(),
                    err
                );
            })
        };

        let assistant_message3: Message = {
            let jd: &mut serde_json::Deserializer<serde_json::de::StrRead<'_>> =
                &mut serde_json::Deserializer::from_str(assistant_message_json3);
            deserialize(jd).unwrap_or_else(|err| {
                panic!(
                    "Deserialization error at {} ({}:{}): {}",
                    err.path(),
                    err.inner().line(),
                    err.inner().column(),
                    err
                );
            })
        };

        let user_message: Message = {
            let jd = &mut serde_json::Deserializer::from_str(user_message_json);
            deserialize(jd).unwrap_or_else(|err| {
                panic!(
                    "Deserialization error at {} ({}:{}): {}",
                    err.path(),
                    err.inner().line(),
                    err.inner().column(),
                    err
                );
            })
        };

        match assistant_message {
            Message::Assistant { content, .. } => {
                assert_eq!(
                    content[0],
                    AssistantContent::Text {
                        text: "\n\nHello there, how may I assist you today?".to_string()
                    }
                );
            }
            _ => panic!("Expected assistant message"),
        }

        match assistant_message2 {
            Message::Assistant {
                content,
                tool_calls,
                ..
            } => {
                assert_eq!(
                    content[0],
                    AssistantContent::Text {
                        text: "\n\nHello there, how may I assist you today?".to_string()
                    }
                );

                assert_eq!(tool_calls, vec![]);
            }
            _ => panic!("Expected assistant message"),
        }

        match assistant_message3 {
            Message::Assistant {
                content,
                tool_calls,
                ..
            } => {
                assert!(content.is_empty());
                assert_eq!(
                    tool_calls[0],
                    ToolCall {
                        id: "call_h89ipqYUjEpCPI6SxspMnoUU".to_string(),
                        r#type: ToolType::Function,
                        function: Function {
                            name: "subtract".to_string(),
                            arguments: serde_json::json!({"x": 2, "y": 5}),
                        },
                    }
                );
            }
            _ => panic!("Expected assistant message"),
        }

        match user_message {
            Message::User { content, .. } => {
                let (first, second) = {
                    let mut iter = content.into_iter();
                    (iter.next().unwrap(), iter.next().unwrap())
                };
                assert_eq!(
                    first,
                    UserContent::Text {
                        text: "What's in this image?".to_string()
                    }
                );
                assert_eq!(second, UserContent::ImageUrl { image_url: ImageUrl { url: "https://upload.wikimedia.org/wikipedia/commons/thumb/d/dd/Gfp-wisconsin-madison-the-nature-boardwalk.jpg/2560px-Gfp-wisconsin-madison-the-nature-boardwalk.jpg".to_string() } });
            }
            _ => panic!("Expected user message"),
        }
    }

    #[test]
    fn test_message_to_message_conversion() {
        let user_message = message::Message::User {
            content: OneOrMany::one(message::UserContent::text("Hello")),
        };

        let assistant_message = message::Message::Assistant {
            id: None,
            content: OneOrMany::one(message::AssistantContent::text("Hi there!")),
        };

        let converted_user_message: Vec<Message> = user_message.clone().try_into().unwrap();
        let converted_assistant_message: Vec<Message> =
            assistant_message.clone().try_into().unwrap();

        match converted_user_message[0].clone() {
            Message::User { content, .. } => {
                assert_eq!(
                    content.first(),
                    UserContent::Text {
                        text: "Hello".to_string()
                    }
                );
            }
            _ => panic!("Expected user message"),
        }

        match converted_assistant_message[0].clone() {
            Message::Assistant { content, .. } => {
                assert_eq!(
                    content[0],
                    AssistantContent::Text {
                        text: "Hi there!".to_string()
                    }
                );
            }
            _ => panic!("Expected assistant message"),
        }

        let original_user_message: message::Message =
            converted_user_message[0].clone().try_into().unwrap();
        let original_assistant_message: message::Message =
            converted_assistant_message[0].clone().try_into().unwrap();

        assert_eq!(original_user_message, user_message);
        assert_eq!(original_assistant_message, assistant_message);
    }

    #[test]
    fn test_message_from_message_conversion() {
        let user_message = Message::User {
            content: OneOrMany::one(UserContent::Text {
                text: "Hello".to_string(),
            }),
        };

        let assistant_message = Message::Assistant {
            content: vec![AssistantContent::Text {
                text: "Hi there!".to_string(),
            }],
            tool_calls: vec![],
        };

        let converted_user_message: message::Message = user_message.clone().try_into().unwrap();
        let converted_assistant_message: message::Message =
            assistant_message.clone().try_into().unwrap();

        match converted_user_message.clone() {
            message::Message::User { content } => {
                assert_eq!(content.first(), message::UserContent::text("Hello"));
            }
            _ => panic!("Expected user message"),
        }

        match converted_assistant_message.clone() {
            message::Message::Assistant { content, .. } => {
                assert_eq!(
                    content.first(),
                    message::AssistantContent::text("Hi there!")
                );
            }
            _ => panic!("Expected assistant message"),
        }

        let original_user_message: Vec<Message> = converted_user_message.try_into().unwrap();
        let original_assistant_message: Vec<Message> =
            converted_assistant_message.try_into().unwrap();

        assert_eq!(original_user_message[0], user_message);
        assert_eq!(original_assistant_message[0], assistant_message);
    }

    #[test]
    fn test_responses() {
        let fireworks_response_json = r#"
        {
            "choices": [
                {
                    "finish_reason": "tool_calls",
                    "index": 0,
                    "message": {
                        "role": "assistant",
                        "tool_calls": [
                            {
                                "function": {
                                "arguments": "{\"x\": 2, \"y\": 5}",
                                "name": "subtract"
                                },
                                "id": "call_1BspL6mQqjKgvsQbH1TIYkHf",
                                "index": 0,
                                "type": "function"
                            }
                        ]
                    }
                }
            ],
            "created": 1740704000,
            "id": "2a81f6a1-4866-42fb-9902-2655a2b5b1ff",
            "model": "accounts/fireworks/models/deepseek-v3",
            "object": "chat.completion",
            "usage": {
                "completion_tokens": 26,
                "prompt_tokens": 248,
                "total_tokens": 274
            }
        }
        "#;

        let novita_response_json = r#"
        {
            "choices": [
                {
                    "finish_reason": "tool_calls",
                    "index": 0,
                    "logprobs": null,
                    "message": {
                        "audio": null,
                        "content": null,
                        "function_call": null,
                        "reasoning_content": null,
                        "refusal": null,
                        "role": "assistant",
                        "tool_calls": [
                            {
                                "function": {
                                    "arguments": "{\"x\": \"2\", \"y\": \"5\"}",
                                    "name": "subtract"
                                },
                                "id": "chatcmpl-tool-f6d2af7c8dc041058f95e2c2eede45c5",
                                "type": "function"
                            }
                        ]
                    },
                    "stop_reason": 128008
                }
            ],
            "created": 1740704592,
            "id": "chatcmpl-a92c60ae125c47c998ecdcb53387fed4",
            "model": "meta-llama/Meta-Llama-3.1-8B-Instruct-fast",
            "object": "chat.completion",
            "prompt_logprobs": null,
            "service_tier": null,
            "system_fingerprint": null,
            "usage": {
                "completion_tokens": 28,
                "completion_tokens_details": null,
                "prompt_tokens": 335,
                "prompt_tokens_details": null,
                "total_tokens": 363
            }
        }
        "#;

        let _firework_response: CompletionResponse = {
            let jd = &mut serde_json::Deserializer::from_str(fireworks_response_json);
            deserialize(jd).unwrap_or_else(|err| {
                panic!(
                    "Deserialization error at {} ({}:{}): {}",
                    err.path(),
                    err.inner().line(),
                    err.inner().column(),
                    err
                );
            })
        };

        let _novita_response: CompletionResponse = {
            let jd = &mut serde_json::Deserializer::from_str(novita_response_json);
            deserialize(jd).unwrap_or_else(|err| {
                panic!(
                    "Deserialization error at {} ({}:{}): {}",
                    err.path(),
                    err.inner().line(),
                    err.inner().column(),
                    err
                );
            })
        };
    }

    #[test]
    fn test_assistant_reasoning_is_silently_skipped() {
        let assistant = message::Message::Assistant {
            id: None,
            content: OneOrMany::one(message::AssistantContent::reasoning("hidden")),
        };

        let converted: Vec<Message> = assistant.try_into().expect("conversion should work");
        assert!(converted.is_empty());
    }

    #[test]
    fn test_assistant_text_and_tool_call_are_preserved_when_reasoning_present() {
        let assistant = message::Message::Assistant {
            id: None,
            content: OneOrMany::many(vec![
                message::AssistantContent::reasoning("hidden"),
                message::AssistantContent::text("visible"),
                message::AssistantContent::tool_call(
                    "call_1",
                    "subtract",
                    serde_json::json!({"x": 2, "y": 1}),
                ),
            ])
            .expect("non-empty assistant content"),
        };

        let converted: Vec<Message> = assistant.try_into().expect("conversion should work");
        assert_eq!(converted.len(), 1);

        match &converted[0] {
            Message::Assistant {
                content,
                tool_calls,
                ..
            } => {
                assert_eq!(
                    content,
                    &vec![AssistantContent::Text {
                        text: "visible".to_string()
                    }]
                );
                assert_eq!(tool_calls.len(), 1);
                assert_eq!(tool_calls[0].id, "call_1");
                assert_eq!(tool_calls[0].function.name, "subtract");
                assert_eq!(
                    tool_calls[0].function.arguments,
                    serde_json::json!({"x": 2, "y": 1})
                );
            }
            _ => panic!("expected assistant message"),
        }
    }

    #[test]
    fn test_request_conversion_errors_when_all_messages_are_filtered() {
        let request = completion::CompletionRequest {
            preamble: None,
            chat_history: OneOrMany::one(message::Message::Assistant {
                id: None,
                content: OneOrMany::one(message::AssistantContent::reasoning("hidden")),
            }),
            documents: vec![],
            tools: vec![],
            temperature: None,
            max_tokens: None,
            tool_choice: None,
            additional_params: None,
            model: None,
            output_schema: None,
        };

        let result = HuggingfaceCompletionRequest::try_from(("meta/test-model", request));
        assert!(matches!(result, Err(CompletionError::RequestError(_))));
    }
}