feat: mira integration #282

VaibhavSatija10 · 2025-02-07T16:08:50Z

Add Mira API Integration for LLM Support

Overview

This pull request introduces support for Mira's API, providing a unified interface for accessing various LLM models. The integration enables Rig users to leverage Mira's API for AI-powered text completions efficiently.

Features

New mira provider module with a complete API client implementation.
API key authentication for secure and straightforward access.
Chat completion support with customizable parameters.
Model listing functionality to retrieve available LLM models.
Credit management capabilities, including balance checks and transaction history.
Comprehensive test coverage to ensure reliability and stability.

Implementation Details

Introduced a Client struct with methods for interacting with Mira's API endpoints.
Implemented robust error handling to manage API failures gracefully.
Provided example usage code to demonstrate basic interactions.
Added integration tests to validate all major functionalities.

This PR enhances the flexibility and functionality of Rig by enabling seamless AI model access via Mira’s API.

Reference:
https://docs.mira.network/network/get-started/introduction

- generation request - available model list - user credits balance

0xMochan · 2025-02-07T21:08:40Z

Hey, thanks for the PR! We'll get to a review soon, but I'll note, we have a specific provider integration flow that involves leveraging our traits CompletionModel to ensure that all providers can be used cohesively throughout the rig ecosystem, please checkout some providers like openai.rs & hyperbolic.rs for examples!

VaibhavSatija10 · 2025-02-18T15:33:37Z

Thanks for the feedback and for pointing me towards the CompletionModel trait. I've reviewed some of the examples including openai.rs and hyperbolic.rs and understand how it's used to standardize provider integration within the rig ecosystem. I've refactored my PR to leverage the CompletionModel trait, ensuring consistency and compatibility with other providers.

0xMochan · 2025-02-18T22:56:02Z

Thanks for the feedback and for pointing me towards the CompletionModel trait. I've reviewed some of the examples including openai.rs and hyperbolic.rs and understand how it's used to standardize provider integration within the rig ecosystem. I've refactored my PR to leverage the CompletionModel trait, ensuring consistency and compatibility with other providers.

Feel free to join our discord if you would like pointers on how best to integrate this!

0xMochan

Additionally, the tests in the CI are failing here. Make sure to test your code before submitting for review again as we must make sure all tests pass! I also wrote a comment on good testing as the current tests make network requests which won't be good for our CI/CD!

0xMochan · 2025-02-21T18:18:56Z

rig-core/src/providers/mira.rs

+}
+
+#[derive(Debug, Serialize, Deserialize)]
+pub struct ChatMessage {


We standardize naming these Message.

0xMochan · 2025-02-21T18:19:31Z

rig-core/src/providers/mira.rs

+}
+
+#[derive(Debug, Serialize)]
+pub struct AiRequest {


Here, we standardize CompletionRequest

0xMochan · 2025-02-21T18:21:54Z

rig-core/src/providers/mira.rs

+        // Add prompt first
+        let prompt = completion_request.prompt_with_context();
+        let prompt_str = match prompt {
+            Message::User { content } => content


We use the TryFrom and From traits to bi-directionally convert between provider specific messages and our general message schema. While our general schema encompasses a lot of structure to ensure we fully cover the spec, if the provider only providers normal text completions, then the From trait can simply return empty messages if that's desired or TryFrom can specifically return errors to return information to the user.

It's important that we use the TryFrom and From traits as it allows messages to easily be converted by the user if they are maintaining their own chat history (especially in future updates).

0xMochan · 2025-02-21T18:22:35Z

rig-core/src/providers/mira.rs

+
+        // Add chat history
+        for message in completion_request.chat_history {
+            match message {


Same here as earlier, defining a impl TryFrom<Message> for completion::Message would be the best way to implement this behavior allowing you to use .try_into in the actual completion method.

0xMochan · 2025-02-21T18:26:16Z

rig-core/src/providers/mira.rs

+                        .into_iter()
+                        .filter_map(|c| match c {
+                            message::UserContent::Text(text) => Some(text.text),
+                            _ => None, // Skip other content types


I would say, it's a bit invisible to silently toss out content types. Ideally, we return an error here so that the user can react to the lack of image and other content support. Instead of filter_map, you'd just map into a collect::<Result<Vec<_>, _>>()? which easily surfaces the error out the loop and early returns!

Alternatively, you can produce a warning.

0xMochan · 2025-02-21T18:26:39Z

rig-core/src/providers/mira.rs

+                        .into_iter()
+                        .filter_map(|c| match c {
+                            message::AssistantContent::Text(text) => Some(text.text),
+                            _ => None, // Skip tool calls


Same as earlier, need to explicitly error or provide a warning via the trace module so that user understands that this model doesn't support tool calling!

0xMochan · 2025-02-21T18:27:14Z

rig-core/src/providers/mira.rs

+                            _ => None, // Skip tool calls
+                        })
+                        .collect::<Vec<_>>()
+                        .join(" ");


Could consider joining by '/n' as I think it closer matches what multiple text content in a row would represent. This is opinionated ofc, up to the implementer!

still unresolved. either write a comment here explaining thought process or make requested change here please

0xMochan · 2025-02-22T01:54:08Z

rig-core/src/providers/mira.rs

+
+impl Client {
+    /// Create a new Mira client with the given API key
+    pub fn new(api_key: impl AsRef<str>) -> Result<Self, MiraError> {


We generally just do api_key: &str, is there a reason why you used impl AsRef<str>?

rig-core/src/providers/mira.rs

0xMochan · 2025-02-22T01:55:50Z

rig-core/src/providers/mira.rs

+    }
+}
+
+#[cfg(test)]


For tests, we recommend studying the tests of other modules like openai or anthropic. We don't actually make requests to their servers, we more or less test the deserialization and serializations of our messages, especially in converting our message to the generic message and back to the provider one!

- Addressed reviewer comments - updated tests

0xMochan

I think mainly the .generate method on impl Client doesn't need to be there (couple of comments are on that method but I think if that method is removed, those comments will be resolved). Some other minor comments, but fairly good. I'd also rebase onto main so we can merge quickly

0xMochan · 2025-03-06T21:50:40Z

rig-core/src/providers/mira.rs

+{
+    let raw = RawMessage::deserialize(deserializer)?;
+    message::Message::try_from(raw).map_err(serde::de::Error::custom)
+}


It's a bit odd to define a custom deserializer here for converting between the RawMessage to message::Message. If you have a good case, I'm ok with leaving it but please leave some comments on why this approach was chosen. Otherwise, I'd ask to conform to how the other providers handle this.

0xMochan · 2025-03-06T21:51:28Z

rig-core/src/providers/mira.rs

+                })
+                .collect::<Vec<_>>()
+                .join(" "),
+            _ => return Err(MiraError::ApiError(422)),


This error message providers zero details on what happened here. Please include a string on what went wrong when matching the prompt.

It seems like this was resolved in the later method

0xMochan · 2025-03-06T21:58:34Z

rig-core/src/providers/mira.rs

+
+        if !response.status().is_success() {
+            let status = response.status();
+            return Err(MiraError::ApiError(status.as_u16()));


Additionally, here it's really important to both trace the error from the response alongside returning it in the wrapped error here. If there is an error from the mira provider or http request, there's no visibility for a dev here.

0xMochan · 2025-03-06T22:20:46Z

rig-core/src/providers/mira.rs

+        &self,
+        model: &str,
+        request: CompletionRequest,
+    ) -> Result<CompletionResponse, MiraError> {


It's unclear why this exists. The main way to generate responses from CompletionRequests is through the .completion api defined by the CompletionModel trait. This code is nearly duplicated below.

0xMochan · 2025-03-06T22:21:34Z

rig-core/src/providers/mira.rs

+                            _ => None, // Skip tool calls
+                        })
+                        .collect::<Vec<_>>()
+                        .join(" ");


still unresolved. either write a comment here explaining thought process or make requested change here please

…esolve comments

VaibhavSatija10 added 2 commits February 7, 2025 21:24

feat: Add Mira AI provider integration

3068982

example: Add Mira AI provider example

b51e518

- generation request - available model list - user credits balance

cvauclair added this to the 2025-02-24 milestone Feb 10, 2025

refactor: Improve Mira provider with agent and completion model support

c07a17c

mateobelanger changed the title ~~Feature/mira integration~~ feat:/mira integration Feb 18, 2025

mateobelanger changed the title ~~feat:/mira integration~~ feat: mira integration Feb 18, 2025

0xMochan requested changes Feb 22, 2025

View reviewed changes

joshua-mo-143 added the non-breaking label Feb 24, 2025

refactor: Mira Integration

a4a5266

- Addressed reviewer comments - updated tests

cvauclair modified the milestones: 2025-02-24, 2025-03-17 Mar 3, 2025

refactor: Remove redundant timeout setting in Mira client

ebf1553

0xMochan requested changes Mar 6, 2025

View reviewed changes

refactor: Improve Mira provider message handling, serialization and r…

8d4cafc

…esolve comments

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

feat: mira integration #282

feat: mira integration #282

VaibhavSatija10 commented Feb 7, 2025

0xMochan commented Feb 7, 2025

VaibhavSatija10 commented Feb 18, 2025

0xMochan commented Feb 18, 2025

0xMochan left a comment

0xMochan Feb 21, 2025

0xMochan Feb 21, 2025

0xMochan Feb 21, 2025

0xMochan Feb 21, 2025

0xMochan Feb 21, 2025

0xMochan Feb 21, 2025

0xMochan Feb 21, 2025

0xMochan Mar 6, 2025

0xMochan Feb 22, 2025

0xMochan Feb 22, 2025

0xMochan left a comment •

edited

Loading

0xMochan Mar 6, 2025

0xMochan Mar 6, 2025

0xMochan Mar 6, 2025

0xMochan Mar 6, 2025

0xMochan Mar 6, 2025

0xMochan Mar 6, 2025

feat: mira integration #282

Are you sure you want to change the base?

feat: mira integration #282

Conversation

VaibhavSatija10 commented Feb 7, 2025

Add Mira API Integration for LLM Support

Overview

Features

Implementation Details

0xMochan commented Feb 7, 2025

VaibhavSatija10 commented Feb 18, 2025

0xMochan commented Feb 18, 2025

0xMochan left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

0xMochan left a comment • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

0xMochan left a comment •

edited

Loading