whisper.swiftui : add model download list & bench methods (#2546)

* swift : fix resources & exclude build

* whisper : impl whisper_timings struct & api

* whisper.swiftui : model list & bench methods

* whisper : return ptr for whisper_get_timings

* revert unnecessary change

* whisper : avoid designated initializer

* whisper.swiftui: code style changes

* whisper.swiftui : get device name / os from UIDevice

* whisper.swiftui : fix UIDevice usage

* whisper.swiftui : add memcpy and ggml_mul_mat (commented)
This commit is contained in:
Jhen-Jie Hong
2024-11-14 03:51:34 +08:00
committed by GitHub
parent a28d82e373
commit 5f8a086e22
9 changed files with 403 additions and 19 deletions

View File

@ -1,4 +1,5 @@
import Foundation
import UIKit
import whisper
enum WhisperError: Error {
@ -55,11 +56,93 @@ actor WhisperContext {
return transcription
}
static func benchMemcpy(nThreads: Int32) async -> String {
return String.init(cString: whisper_bench_memcpy_str(nThreads))
}
static func benchGgmlMulMat(nThreads: Int32) async -> String {
return String.init(cString: whisper_bench_ggml_mul_mat_str(nThreads))
}
private func systemInfo() -> String {
var info = ""
if (ggml_cpu_has_neon() != 0) { info += "NEON " }
if (ggml_cpu_has_metal() != 0) { info += "METAL " }
if (ggml_cpu_has_blas() != 0) { info += "BLAS " }
return String(info.dropLast())
}
func benchFull(modelName: String, nThreads: Int32) async -> String {
let nMels = whisper_model_n_mels(context)
if (whisper_set_mel(context, nil, 0, nMels) != 0) {
return "error: failed to set mel"
}
// heat encoder
if (whisper_encode(context, 0, nThreads) != 0) {
return "error: failed to encode"
}
var tokens = [whisper_token](repeating: 0, count: 512)
// prompt heat
if (whisper_decode(context, &tokens, 256, 0, nThreads) != 0) {
return "error: failed to decode"
}
// text-generation heat
if (whisper_decode(context, &tokens, 1, 256, nThreads) != 0) {
return "error: failed to decode"
}
whisper_reset_timings(context)
// actual run
if (whisper_encode(context, 0, nThreads) != 0) {
return "error: failed to encode"
}
// text-generation
for i in 0..<256 {
if (whisper_decode(context, &tokens, 1, Int32(i), nThreads) != 0) {
return "error: failed to decode"
}
}
// batched decoding
for _ in 0..<64 {
if (whisper_decode(context, &tokens, 5, 0, nThreads) != 0) {
return "error: failed to decode"
}
}
// prompt processing
for _ in 0..<16 {
if (whisper_decode(context, &tokens, 256, 0, nThreads) != 0) {
return "error: failed to decode"
}
}
whisper_print_timings(context)
let deviceModel = await UIDevice.current.model
let systemName = await UIDevice.current.systemName
let systemInfo = self.systemInfo()
let timings: whisper_timings = whisper_get_timings(context).pointee
let encodeMs = String(format: "%.2f", timings.encode_ms)
let decodeMs = String(format: "%.2f", timings.decode_ms)
let batchdMs = String(format: "%.2f", timings.batchd_ms)
let promptMs = String(format: "%.2f", timings.prompt_ms)
return "| \(deviceModel) | \(systemName) | \(systemInfo) | \(modelName) | \(nThreads) | 1 | \(encodeMs) | \(decodeMs) | \(batchdMs) | \(promptMs) | <todo> |"
}
static func createContext(path: String) throws -> WhisperContext {
var params = whisper_context_default_params()
#if targetEnvironment(simulator)
params.use_gpu = false
print("Running on the simulator, using CPU")
#else
params.flash_attn = true // Enabled by default for Metal
#endif
let context = whisper_init_from_file_with_params(path, params)
if let context {