LLM example --> Phi-2 model implementation: _uiState function error

[mhyeonsoo]

Hi, thanks for providing a great examples. It is amazing indeed.

While I am deploying the test app for LLM chat, I tried two models which are Gemma and phi-2. For Gemma, it worked perfectly. But for phi-2, I changed the code depending on your guide in comment

// `GemmaUiState()` is optimized for the Gemma model.
// Replace `GemmaUiState` with `ChatUiState()` if you're using a different model
private val _uiState: MutableStateFlow<ChatUiState> = MutableStateFlow(ChatUiState())

When I changed the code like above, error happens in the below part.

try {
                val fullPrompt = _uiState.value.fullPrompt
                inferenceModel.generateResponseAsync(fullPrompt)
                inferenceModel.partialResults
                    .collectIndexed { index, (partialResult, done) ->
                        currentMessageId?.let {
                            if (index == 0) {
                                _uiState.value.appendFirstMessage(it, partialResult) // <-- this is the error part
                            } else {
                                _uiState.value.appendMessage(it, partialResult, done)
                            }
                            if (done) {
                                currentMessageId = null
                                // Re-enable text input
                                setInputEnabled(true)
                            }
                        }
                    }
            } catch (e: Exception) {
                _uiState.value.addMessage(e.localizedMessage ?: "Unknown Error", MODEL_PREFIX)
                setInputEnabled(true)
            }

The error message is

Unresolved reference: appendFirstMessage

can you give me a feedback for this issue?

Thanks!

[talumbau]

Hi,

We had a new release of MP and pushed a fix to the sample app. I actually think you would be OK using GemmaUIState with different models (so the naming is not actually accurate). However, for Phi2, please see https://github.com/google-ai-edge/mediapipe-samples/issues/380 to make sure the Phi2 task model is set up correctly. That is more likely to be the cause of any underlying issues.