Parse Python response to EmbeddingResponse
(
&self,
py: Python<'_>,
result: PyObject,
)
| 42 | |
| 43 | /// Parse Python response to EmbeddingResponse |
| 44 | fn parse_python_response( |
| 45 | &self, |
| 46 | py: Python<'_>, |
| 47 | result: PyObject, |
| 48 | ) -> GraphBitResult<EmbeddingResponse> { |
| 49 | // The Python embedding method should return a list of floats (single embedding) |
| 50 | // or a list of lists of floats (multiple embeddings) |
| 51 | |
| 52 | let result_bound = result.bind(py); |
| 53 | let embeddings: Vec<Vec<f32>> = if let Ok(list) = result_bound.downcast::<PyList>() { |
| 54 | if list.is_empty() { |
| 55 | return Err(GraphBitError::llm("Empty embedding response from Python")); |
| 56 | } |
| 57 | |
| 58 | // Check if it's a single embedding (list of floats) or multiple (list of lists) |
| 59 | let first_item = list.get_item(0).map_err(|e| { |
| 60 | GraphBitError::llm(format!( |
| 61 | "Failed to get first item from Python response: {e}" |
| 62 | )) |
| 63 | })?; |
| 64 | |
| 65 | if first_item.downcast::<PyList>().is_ok() { |
| 66 | // Multiple embeddings (list of lists) |
| 67 | list.iter() |
| 68 | .map(|item| { |
| 69 | item.downcast::<PyList>() |
| 70 | .map_err(|e| { |
| 71 | GraphBitError::llm(format!("Invalid embedding format: {e}")) |
| 72 | })? |
| 73 | .iter() |
| 74 | .map(|v| { |
| 75 | v.extract::<f32>().map_err(|e| { |
| 76 | GraphBitError::llm(format!("Failed to extract float: {e}")) |
| 77 | }) |
| 78 | }) |
| 79 | .collect::<Result<Vec<f32>, _>>() |
| 80 | }) |
| 81 | .collect::<Result<Vec<Vec<f32>>, _>>()? |
| 82 | } else { |
| 83 | // Single embedding (list of floats) |
| 84 | let embedding: Vec<f32> = list |
| 85 | .iter() |
| 86 | .map(|v| { |
| 87 | v.extract::<f32>().map_err(|e| { |
| 88 | GraphBitError::llm(format!("Failed to extract float: {e}")) |
| 89 | }) |
| 90 | }) |
| 91 | .collect::<Result<Vec<f32>, _>>()?; |
| 92 | vec![embedding] |
| 93 | } |
| 94 | } else { |
| 95 | return Err(GraphBitError::llm( |
| 96 | "Python embedding response must be a list", |
| 97 | )); |
| 98 | }; |
| 99 | |
| 100 | // Estimate token usage (Python providers typically don't provide this) |
| 101 | let total_chars: usize = embeddings.iter().map(|e| e.len()).sum(); |
no test coverage detected