CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-10-30 08:42:00 +00:00
Code Issues Packages Projects Releases Wiki Activity

swift : fix token_to_piece implementation (#4278)

Browse Source 
* Fix token_to_piece implementation in Swift

* Fix errors
This commit is contained in:
Miwa / Ensan
2023-12-02 03:19:45 +09:00
committed by GitHub
parent 511f52c334
commit b220222a64
2 changed files with 18 additions and 14 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
examples/batched.swift/Sources/main.swift
View File

@@ -230,18 +230,15 @@ private func token_to_piece(token: llama_token, buffer: inout [CChar]) -> String
var result = [CChar](repeating: 0, count: 8) var result = [CChar](repeating: 0, count: 8)
let nTokens = llama_token_to_piece(model, token, &result, Int32(result.count)) let nTokens = llama_token_to_piece(model, token, &result, Int32(result.count))
if nTokens < 0 { if nTokens < 0 {
if result.count >= -Int(nTokens) { let actualTokensCount = -Int(nTokens)
result.removeLast(-Int(nTokens)) result = .init(repeating: 0, count: actualTokensCount)
} else {
result.removeAll()
}
let check = llama_token_to_piece( let check = llama_token_to_piece(
model, model,
token, token,
&result, &result,
Int32(result.count) Int32(result.count)
) )
assert(check == nTokens) assert(check == actualTokensCount)
} else { } else {
result.removeLast(result.count - Int(nTokens)) result.removeLast(result.count - Int(nTokens))
} }
@@ -259,5 +256,4 @@ private func token_to_piece(token: llama_token, buffer: inout [CChar]) -> String
buffer = [] buffer = []
return bufferString return bufferString
} }
return nil
} }

22
examples/llama.swiftui/llama.cpp.swift/LibLlama.swift
View File

@@ -164,13 +164,21 @@ actor LlamaContext {
private func token_to_piece(token: llama_token) -> String { private func token_to_piece(token: llama_token) -> String {
let result = UnsafeMutablePointer<Int8>.allocate(capacity: 8) let result = UnsafeMutablePointer<Int8>.allocate(capacity: 8)
result.initialize(repeating: Int8(0), count: 8) result.initialize(repeating: Int8(0), count: 8)
defer {
result.deallocate()
}
let nTokens = llama_token_to_piece(model, token, result, 8)
let _ = llama_token_to_piece(model, token, result, 8) if nTokens < 0 {
let newResult = UnsafeMutablePointer<Int8>.allocate(capacity: Int(-nTokens))
let resultStr = String(cString: result) newResult.initialize(repeating: Int8(0), count: Int(-nTokens))
defer {
result.deallocate() newResult.deallocate()
}
return resultStr _ = llama_token_to_piece(model, token, newResult, -nTokens)
return String(cString: newResult)
} else {
return String(cString: result)
}
} }
} }