Add support for BPE tokenization.

This commit is contained in:
2023-04-23 18:49:52 -07:00
parent 0fe35cd976
commit f61fe6a18f
9 changed files with 562 additions and 17 deletions

View File

@ -154,4 +154,15 @@ public actor ModelState {
}
return pipelineConfig
}
public func tokenize(_ request: SdTokenizeRequest) throws -> SdTokenizeResponse {
guard let tokenizer else {
throw SdCoreError.modelNotLoaded
}
let results = tokenizer.tokenize(input: request.input)
var response = SdTokenizeResponse()
response.tokens = results.tokens
response.tokenIds = results.tokenIDs.map { UInt64($0) }
return response
}
}