whisper.swiftui : add model download list & bench methods

[jhen0409]

This PR fixes the build of whisper.swiftui, and then add model download list (from ggerganov/whisper.cpp HF repo) and bench methods.

The bench method output format is following format from scripts/bench-all.sh:

whisper.cpp/scripts/bench-all.sh

Lines 61 to 62 in 31aea56

	printf "| %6s | %6s | %16s | %13s | %3s | %3s | %7s | %7s | %7s | %7s | %7s |\n" "CPU" "OS" "Config" "Model" "Th" "FA" "Enc." "Dec." "Bch5" "PP" "Commit"
	printf "| %6s | %6s | %16s | %13s | %3s | %3s | %7s | %7s | %7s | %7s | %7s |\n" "---" "---" "---" "---" "---" "---" "---" "---" "---" "---" "---"

New Buttons:

• View Models: Navigate to model list (We can download model & load downloaded model)
• Bench (bench_full for current loaded model)
• Bench All (bench_full for downloaded models)
• Copy Logs

The memcpy and ggml_mul_mat methods are added but commented.

The demo looks like in iPhone 15 Pro:

Benchmark result in iPhone 15 Pro (A17 Pro):

memcpy and ggml_mul_mat:

Running memcpy benchmark
memcpy:   14.97 GB/s (heat-up)
memcpy:   21.46 GB/s ( 1 thread)
memcpy:   20.66 GB/s ( 1 thread)
memcpy:   22.52 GB/s ( 2 thread)
memcpy:   22.57 GB/s ( 3 thread)
memcpy:   22.46 GB/s ( 4 thread)
sum:    -3072001183.000000

Running ggml_mul_mat benchmark with 4 threads
  64 x   64: Q4_0     4.1 GFLOPS (128 runs) | Q4_1     3.7 GFLOPS (128 runs)
  64 x   64: Q5_0     4.5 GFLOPS (128 runs) | Q5_1     4.6 GFLOPS (128 runs) | Q8_0     4.0 GFLOPS (128 runs)
  64 x   64: F16      4.4 GFLOPS (128 runs) | F32      4.7 GFLOPS (128 runs)
 128 x  128: Q4_0    28.2 GFLOPS (128 runs) | Q4_1    25.8 GFLOPS (128 runs)
 128 x  128: Q5_0    20.4 GFLOPS (128 runs) | Q5_1    26.4 GFLOPS (128 runs) | Q8_0    28.9 GFLOPS (128 runs)
 128 x  128: F16     16.7 GFLOPS (128 runs) | F32     31.5 GFLOPS (128 runs)
 256 x  256: Q4_0    89.1 GFLOPS (128 runs) | Q4_1    84.1 GFLOPS (128 runs)
 256 x  256: Q5_0    70.1 GFLOPS (128 runs) | Q5_1    62.5 GFLOPS (128 runs) | Q8_0   103.2 GFLOPS (128 runs)
 256 x  256: F16     66.6 GFLOPS (128 runs) | F32     73.7 GFLOPS (128 runs)
 512 x  512: Q4_0   108.0 GFLOPS (128 runs) | Q4_1   101.0 GFLOPS (128 runs)
 512 x  512: Q5_0    83.1 GFLOPS (128 runs) | Q5_1    74.1 GFLOPS (128 runs) | Q8_0   130.6 GFLOPS (128 runs)
 512 x  512: F16     73.1 GFLOPS (128 runs) | F32     83.2 GFLOPS (128 runs)
1024 x 1024: Q4_0   116.6 GFLOPS ( 55 runs) | Q4_1   107.2 GFLOPS ( 50 runs)
1024 x 1024: Q5_0    87.2 GFLOPS ( 41 runs) | Q5_1    78.7 GFLOPS ( 37 runs) | Q8_0   139.8 GFLOPS ( 66 runs)
1024 x 1024: F16     74.2 GFLOPS ( 35 runs) | F32     76.5 GFLOPS ( 36 runs)
2048 x 2048: Q4_0   120.5 GFLOPS (  8 runs) | Q4_1   110.1 GFLOPS (  7 runs)
2048 x 2048: Q5_0    89.0 GFLOPS (  6 runs) | Q5_1    80.4 GFLOPS (  5 runs) | Q8_0   143.2 GFLOPS (  9 runs)
2048 x 2048: F16     74.0 GFLOPS (  5 runs) | F32     67.5 GFLOPS (  4 runs)
4096 x 4096: Q4_0   119.0 GFLOPS (  3 runs) | Q4_1   107.6 GFLOPS (  3 runs)
4096 x 4096: Q5_0    88.2 GFLOPS (  3 runs) | Q5_1    78.7 GFLOPS (  3 runs) | Q8_0   138.9 GFLOPS (  3 runs)
4096 x 4096: F16     69.1 GFLOPS (  3 runs) | F32     56.1 GFLOPS (  3 runs)

full:

CPU	OS	Config	Model	Th	FA	Enc.	Dec.	Bch5	PP	Commit
A17 Pro	iOS	NEON METAL	tiny	4	1	50.00	4.47	1.17	0.07	31aea56
A17 Pro	iOS	NEON METAL	tiny-q5_1	4	1	52.75	3.88	1.34	0.07	31aea56
A17 Pro	iOS	NEON METAL	tiny-q8_0	4	1	50.81	3.79	1.13	0.07	31aea56
A17 Pro	iOS	NEON METAL	tiny.en	4	1	50.48	4.69	1.64	0.07	31aea56
A17 Pro	iOS	NEON METAL	tiny.en-q5_1	4	1	52.39	3.88	1.55	0.07	31aea56
A17 Pro	iOS	NEON METAL	tiny.en-q8_0	4	1	51.07	3.78	1.08	0.07	31aea56
A17 Pro	iOS	NEON METAL	base	4	1	103.23	6.06	1.32	0.11	31aea56
A17 Pro	iOS	NEON METAL	base-q5_1	4	1	108.43	4.71	1.72	0.11	31aea56
A17 Pro	iOS	NEON METAL	base-q8_0	4	1	104.34	5.46	1.36	0.11	31aea56
A17 Pro	iOS	NEON METAL	base.en	4	1	102.70	6.00	1.39	0.11	31aea56
A17 Pro	iOS	NEON METAL	base.en-q5_1	4	1	109.11	5.22	1.44	0.11	31aea56
A17 Pro	iOS	NEON METAL	base.en-q8_0	4	1	104.17	5.13	1.55	0.11	31aea56
A17 Pro	iOS	NEON METAL	small	4	1	377.67	11.65	2.78	0.41	31aea56
A17 Pro	iOS	NEON METAL	small-q5_1	4	1	366.79	8.56	2.83	0.33	31aea56
A17 Pro	iOS	NEON METAL	small-q8_0	4	1	354.77	8.90	2.61	0.32	31aea56
A17 Pro	iOS	NEON METAL	small.en	4	1	351.71	11.74	2.78	0.32	31aea56
A17 Pro	iOS	NEON METAL	small.en-q5_1	4	1	391.11	8.34	2.74	0.36	31aea56
A17 Pro	iOS	NEON METAL	small.en-q8_0	4	1	411.22	8.96	2.65	0.43	31aea56
A17 Pro	iOS	NEON METAL	medium	4	1	1236.87	29.46	8.70	1.18	31aea56
A17 Pro	iOS	NEON METAL	medium-q5_0	4	1	1400.28	19.67	8.47	1.31	31aea56
A17 Pro	iOS	NEON METAL	medium-q8_0	4	1	1347.07	22.92	9.75	1.25	31aea56
A17 Pro	iOS	NEON METAL	medium.en	4	1	1307.12	30.41	9.87	1.23	31aea56
A17 Pro	iOS	NEON METAL	medium.en-q5_0	4	1	1426.46	20.24	10.36	1.32	31aea56
A17 Pro	iOS	NEON METAL	medium.en-q8_0	4	1	1359.09	22.92	10.04	1.24	31aea56
A17 Pro	iOS	NEON METAL	large-v3-turbo	4	1	2752.62	10.63	2.60	0.54	31aea56
A17 Pro	iOS	NEON METAL	large-v3-turbo-q5_0	4	1	2617.94	6.44	2.64	0.60	31aea56
A17 Pro	iOS	NEON METAL	large-v3-turbo-q8_0	4	1	2494.79	7.20	2.63	0.69	31aea56

Skipped large-v1 ~ v3 downloads

[ggerganov]

@jhen0409 I think this change breaks the builds on MacOS:

Can you propose a fix?

[jhen0409]

@jhen0409 I think this change breaks the builds on MacOS:

Can you propose a fix?

We can change the macOS destination to Mac Catalyst or Designed for iPad.

PR: #2562