Most Efficient Large Language Models for AI PC#

This page is regularly updated to help you identify the best-performing LLMs on the Intel® Core™ Ultra processor family and AI PCs. The current data is as of OpenVINO 2026.3, 13 August 2026.

The tables below list the key performance indicators for inference on built-in GPUs.

Topology

Precision

Input Size

1st latency (ms)

2nd latency (ms)

2nd token per sec

minicpm4-0.5b

INT4-MIXED

32

15.5

4

250

minicpm4-0.5b

INT4-MIXED

32

16.1

4.1

243.902439

minicpm4-0.5b

INT4-MIXED

1024

22

4.1

243.902439

minicpm4-0.5b

INT4-MIXED

32

15.4

4.1

243.902439

minicpm4-0.5b

INT4-MIXED

1024

27.5

4.2

238.0952381

gemma-3-270m

INT4-MIXED

1024

19.3

4.4

227.2727273

gemma-3-270m

INT4-MIXED

32

16.8

4.4

227.2727273

gemma-3-270m

INT8-CW

1024

20.2

4.4

227.2727273

minicpm4-0.5b

INT4-MIXED

1024

22.2

4.4

227.2727273

qwen2.5-coder-0.5b-instruct

INT4-MIXED

32

15.8

4.5

222.2222222

whisper-large-v3-turbo

INT4-MIXED

1024

170.9

4.5

222.2222222

gemma-3-270m

INT8-CW

32

16.7

4.6

217.3913043

lfm2.5-350m

INT4-MIXED

32

18.1

4.6

217.3913043

qwen2.5-coder-0.5b-instruct

INT4-MIXED

32

16.6

4.6

217.3913043

qwen2.5-coder-0.5b-instruct

INT4-MIXED

1024

24.8

4.6

217.3913043

distil-large-v2

INT4-MIXED

32

110.1

4.7

212.7659574

distil-large-v2

INT4-MIXED

1024

163.1

4.7

212.7659574

nuextract-tiny

INT4-MIXED

32

17

4.7

212.7659574

qwen2.5-coder-0.5b-instruct

INT4-MIXED

1024

30.5

4.7

212.7659574

qwen2.5-coder-0.5b-instruct

INT4-MIXED

32

15.9

4.7

212.7659574

whisper-large-v3-turbo

INT4-MIXED

32

115.9

4.7

212.7659574

whisper-large-v3-turbo

INT8-CW

1024

166.8

4.7

212.7659574

distil-large-v2

INT8-CW

32

103.6

4.8

208.3333333

distil-large-v2

INT8-CW

1024

156.2

4.8

208.3333333

nuextract-tiny

INT4-MIXED

1024

26.7

4.8

208.3333333

qwen2.5-coder-0.5b-instruct

INT4-MIXED

1024

25.3

4.8

208.3333333

whisper-large-v3-turbo

INT8-CW

32

111.8

4.8

208.3333333

lfm2.5-350m

INT4-MIXED

1024

35.2

4.9

204.0816327

lfm2.5-350m

INT8-CW

32

19.5

5.6

178.5714286

minicpm4-0.5b

INT8-CW

32

17.2

5.6

178.5714286

lfm2.5-350m

INT8-CW

1024

32.8

5.8

172.4137931

nuextract-tiny

INT8-CW

32

17.8

5.8

172.4137931

minicpm4-0.5b

INT8-CW

1024

24.5

5.9

169.4915254

whisper-small

INT4-MIXED

32

66.1

5.9

169.4915254

whisper-small

INT4-MIXED

1024

115.3

6

166.6666667

qwen2.5-coder-0.5b-instruct

INT8-CW

32

18

6.1

163.9344262

whisper-small

INT4-MIXED

1024

117.3

6.1

163.9344262

nuextract-tiny

INT8-CW

1024

24.4

6.2

161.2903226

whisper-small

INT4-MIXED

1024

121.5

6.2

161.2903226

qwen2.5-coder-0.5b-instruct

INT8-CW

1024

27.3

6.3

158.7301587

whisper-small

INT8-CW

1024

124.3

6.3

158.7301587

whisper-small

INT4-MIXED

32

67

6.4

156.25

whisper-small

INT8-CW

32

69

6.4

156.25

tiny-llama-1.1b-chat

INT4-MIXED

32

14.3

6.5

153.8461538

whisper-small

INT4-MIXED

32

70.6

6.5

153.8461538

nanollava

INT4-MIXED

760

57.5

6.6

151.5151515

whisper-large-v3-turbo

FP16

32

119.3

6.6

151.5151515

whisper-large-v3-turbo

FP16

1024

175.8

6.6

151.5151515

gemma-3-270m

FP16

1024

17.4

6.7

149.2537313

gemma-3-270m

FP16

32

14.9

6.7

149.2537313

minicpm5-1b

INT4-MIXED

32

17.3

6.7

149.2537313

tiny-llama-1.1b-chat

INT4-MIXED

1024

41.2

6.7

149.2537313

distil-large-v2

FP16

1024

165.9

6.8

147.0588235

tiny-llama-1.1b-chat

INT4-MIXED

32

14.9

6.8

147.0588235

distil-large-v2

FP16

32

113.2

6.9

144.9275362

minicpm5-1b

INT4-MIXED

1024

44.1

7

142.8571429

tiny-llama-1.1b-chat

INT4-MIXED

1024

55.5

7

142.8571429

qwen3.5-0.8b

INT4-MIXED

85

45.7

7.1

140.8450704

qwen3.5-0.8b

INT4-MIXED

85

45.9

7.2

138.8888889

nanollava

INT4-MIXED

1752

100.4

7.4

135.1351351

nanollava

INT8-CW

760

56.8

7.4

135.1351351

qwen3.5-0.8b

INT4-MIXED

85

52

7.4

135.1351351

qwen3.5-0.8b

INT4-MIXED

1024

89.4

7.5

133.3333333

qwen3.5-0.8b

INT4-MIXED

1024

89.7

7.5

133.3333333

lfm2.5-1.2b-instruct

INT4-MIXED

32

18.8

8

125

llama-3.2-1b-instruct

INT4-MIXED

32

13.2

8

125

qwen3.5-0.8b

INT4-MIXED

1024

99.9

8

125

llama-3.2-1b-instruct

INT4-MIXED

32

13.5

8.1

123.4567901

whisper-small

FP16

32

64.5

8.1

123.4567901

whisper-small

FP16

1024

116.5

8.1

123.4567901

lfm2.5-1.2b-instruct

INT4-MIXED

32

18.3

8.2

121.9512195

lfm2-1.2b

INT4-MIXED

1024

65

8.3

120.4819277

lfm2-1.2b

INT4-MIXED

32

18.3

8.3

120.4819277

lfm2.5-1.2b-instruct

INT4-MIXED

32

19

8.3

120.4819277

lfm2.5-1.2b-instruct

INT4-MIXED

1024

66

8.3

120.4819277

llama-3.2-1b-instruct

INT4-MIXED

1024

41.6

8.3

120.4819277

gemma-3-1b-it

INT4-MIXED

1024

47.1

8.4

119.047619

lfm2-1.2b

INT4-MIXED

32

19.2

8.4

119.047619

llama-3.2-1b-instruct

INT4-MIXED

1024

49.3

8.4

119.047619

lfm2-1.2b

INT4-MIXED

1024

66.2

8.5

117.6470588

lfm2.5-1.2b-instruct

INT4-MIXED

1024

66.8

8.5

117.6470588

nanollava

INT8-CW

1752

92.6

8.5

117.6470588

gemma-3-1b-it

INT4-MIXED

1024

47.9

8.6

116.2790698

gemma-3-1b-it

INT4-MIXED

32

21.8

8.6

116.2790698

lfm2-1.2b

INT4-MIXED

32

18.8

8.6

116.2790698

lfm2.5-350m

FP16

32

17.3

8.6

116.2790698

lfm2-1.2b

INT4-MIXED

1024

74.9

8.7

114.9425287

minicpm4-0.5b

FP4-NORMALIZED

32

14

8.7

114.9425287

gemma-3-1b-it

INT4-MIXED

32

22.6

8.8

113.6363636

gemma-3-1b-it

INT4-MIXED

32

21.9

8.8

113.6363636

lfm2.5-350m

FP16

1024

35.7

8.8

113.6363636

minicpm4-0.5b

FP4-NORMALIZED

1024

29.5

8.8

113.6363636

lfm2.5-1.2b-instruct

INT4-MIXED

1024

75.6

8.9

112.3595506

exaone-4.0-1.2b

INT4-MIXED

32

24.1

9

111.1111111

gemma-3-1b-it

INT4-MIXED

1024

53.9

9

111.1111111

exaone-4.0-1.2b

INT4-MIXED

32

24.6

9.1

109.8901099

exaone-4.0-1.2b

INT4-MIXED

32

25.4

9.2

108.6956522

exaone-4.0-1.2b

INT4-MIXED

1024

70.9

9.2

108.6956522

qwen3.5-0.8b

INT8-CW

85

51.6

9.3

107.5268817

exaone-4.0-1.2b

INT4-MIXED

1024

72.5

9.4

106.3829787

exaone-4.0-1.2b

INT4-MIXED

1024

79.4

9.5

105.2631579

minicpm4-0.5b

FP16

32

14.7

9.5

105.2631579

minicpm5-1b

INT8-CW

32

18.4

9.5

105.2631579

qwen3.5-0.8b

INT8-CW

1024

94.1

9.6

104.1666667

minicpm4-0.5b

FP16

1024

30.3

9.7

103.0927835

minicpm-1b-sft

INT4-MIXED

31

34

9.8

102.0408163

minicpm-1b-sft

INT4-MIXED

31

34.1

9.9

101.010101

qwen2.5-coder-1.5b-instruct

INT4-MIXED

32

20

9.9

101.010101

minicpm-1b-sft

INT4-MIXED

1014

65.9

10

100

minicpm5-1b

INT8-CW

1024

40.7

10

100

nuextract-tiny

FP16

32

16.5

10

100

qwen2.5-coder-1.5b-instruct

INT4-MIXED

1024

66.9

10

100

deepseek-r1-distill-qwen-1.5b

INT4-MIXED

32

19.8

10.1

99.00990099

minicpm-1b-sft

INT4-MIXED

31

38.4

10.1

99.00990099

qwen2.5-1.5b-instruct

INT4-MIXED

32

19.8

10.1

99.00990099

qwen2.5-coder-1.5b-instruct

INT4-MIXED

32

19.8

10.1

99.00990099

qwen2.5-coder-1.5b-instruct

INT4-MIXED

32

20.9

10.2

98.03921569

deepseek-r1-distill-qwen-1.5b

INT4-MIXED

1024

78.1

10.3

97.08737864

qwen2.5-1.5b-instruct

INT4-MIXED

1024

78.3

10.3

97.08737864

minicpm-1b-sft

INT4-MIXED

1014

67

10.4

96.15384615

qwen2.5-coder-1.5b-instruct

INT4-MIXED

1024

83

10.4

96.15384615

qwen2.5-coder-1.5b-instruct

INT4-MIXED

1024

82.4

10.4

96.15384615

qwen2.5-coder-0.5b-instruct

FP16

32

16

10.5

95.23809524

qwen2.5-coder-0.5b-instruct

FP16

1024

34.3

10.7

93.45794393

minicpm-1b-sft

INT4-MIXED

1014

78.2

10.8

92.59259259

nuextract-tiny

FP16

1024

32.1

10.8

92.59259259

qwen2.5-1.5b-instruct

INT4-MIXED

32

21.8

10.9

91.74311927

nanollava

FP16

760

59

11.5

86.95652174

qwen2.5-1.5b-instruct

INT4-MIXED

1024

83.5

11.5

86.95652174

tiny-llama-1.1b-chat

INT8-CW

32

22.7

11.5

86.95652174

gemma-3-1b-it

INT8-CW

32

24.8

11.6

86.20689655

gemma-3-1b-it

INT8-CW

1024

59

11.7

85.47008547

tiny-llama-1.1b-chat

INT8-CW

1024

51.4

11.8

84.74576271

lfm2-8b-a1b

INT4-MIXED

32

77.9

12

83.33333333

llama-3.2-1b-instruct

INT8-CW

32

14.8

12.7

78.74015748

nanollava

FP16

1752

102.9

12.7

78.74015748

deepseek-r1-distill-qwen-1.5b

INT4-MIXED

32

22.3

12.9

77.51937984

deepseek-r1-distill-qwen-1.5b

INT4-MIXED

1024

87.9

13

76.92307692

lfm2-8b-a1b

INT4-MIXED

1024

184.1

13

76.92307692

llama-3.2-1b-instruct

INT8-CW

1024

54.6

13

76.92307692

lfm2.5-1.2b-instruct

INT8-CW

32

21.5

13.3

75.18796992

lfm2-1.2b

INT8-CW

32

22.1

13.5

74.07407407

qwen3.5-0.8b

FP4-NORMALIZED

85

37.9

13.5

74.07407407

lfm2.5-1.2b-instruct

INT8-CW

1024

83.9

13.6

73.52941176

lfm2-1.2b

INT8-CW

1024

84.7

13.8

72.46376812

qwen3.5-0.8b

FP4-NORMALIZED

1024

102.7

13.8

72.46376812

exaone-4.0-1.2b

INT8-CW

32

27.5

13.9

71.94244604

exaone-4.0-1.2b

INT8-CW

1024

81.4

14.5

68.96551724

phi-2

INT4-MIXED

32

29.6

15.5

64.51612903

minicpm-1b-sft

INT8-CW

31

38.9

15.6

64.1025641

qwen2.5-1.5b-instruct

INT8-CW

32

22.6

15.7

63.69426752

qwen2.5-coder-1.5b-instruct

INT8-CW

32

22.8

15.7

63.69426752

stable-zephyr-3b-dpo

INT4-MIXED

32

30.7

15.7

63.69426752

deepseek-r1-distill-qwen-1.5b

INT8-CW

32

23.1

15.8

63.29113924

qwen3.5-0.8b

FP16

85

38.1

15.9

62.89308176

qwen3.5-0.8b

FP16

1024

104

16

62.5

deepseek-r1-distill-qwen-1.5b

INT8-CW

1024

74.4

16.3

61.34969325

qwen2.5-1.5b-instruct

INT8-CW

1024

74.1

16.3

61.34969325

qwen2.5-coder-1.5b-instruct

INT8-CW

1024

73.3

16.3

61.34969325

gemma-4-e2b-it

INT4-MIXED

274

171.5

16.5

60.60606061

minicpm-1b-sft

INT8-CW

1014

78.8

16.8

59.52380952

gemma-4-e2b-it

INT4-MIXED

1024

295.5

16.9

59.17159763

gemma-4-e2b-it

INT4-MIXED

274

186.9

17

58.82352941

phi-2

INT4-MIXED

32

38.5

17

58.82352941

whisper-large-v3

INT4-MIXED

1024

261.4

17.1

58.47953216

phi-2

INT4-MIXED

1024

161.6

17.2

58.13953488

whisper-large-v3

INT4-MIXED

32

183.6

17.3

57.80346821

gemma-4-e2b-it

INT4-MIXED

1024

319.7

17.4

57.47126437

minicpm5-1b

FP16

32

19.5

17.4

57.47126437

smollm3-3b

INT4-MIXED

32

25.7

17.5

57.14285714

stable-zephyr-3b-dpo

INT4-MIXED

32

41

17.6

56.81818182

stable-zephyr-3b-dpo

INT4-MIXED

1024

153.2

17.6

56.81818182

gemma-3-1b-it

FP4-NORMALIZED

32

21.4

17.7

56.49717514

gemma-3-1b-it

FP4-NORMALIZED

1024

69.2

17.8

56.17977528

lfm2-8b-a1b

INT8-CW

32

76.7

17.8

56.17977528

minicpm5-1b

FP16

1024

52.3

17.8

56.17977528

smollm3-3b

INT4-MIXED

1024

122.8

18

55.55555556

smollm3-3b

INT4-MIXED

32

31.6

18

55.55555556

llama-3.2-3b-instruct

INT4-MIXED

32

26.4

18.2

54.94505495

smollm3-3b

INT4-MIXED

32

32.4

18.3

54.64480874

qwen2.5-coder-3b-instruct

INT4-MIXED

32

32.8

18.4

54.34782609

smollm3-3b

INT4-MIXED

1024

124.8

18.6

53.76344086

llama-3.2-3b-instruct

INT4-MIXED

32

27.1

18.7

53.47593583

lfm2-8b-a1b

INT8-CW

1024

251.6

18.8

53.19148936

qwen2.5-coder-3b-instruct

INT4-MIXED

32

32.7

18.8

53.19148936

qwen2.5-coder-3b-instruct

INT4-MIXED

1024

120.5

18.8

53.19148936

smollm3-3b

INT4-MIXED

1024

147.4

18.8

53.19148936

llama-3.2-3b-instruct

INT4-MIXED

32

27.8

18.9

52.91005291

phi-2

INT4-MIXED

1024

209.8

18.9

52.91005291

stable-zephyr-3b-dpo

INT4-MIXED

1024

175.7

18.9

52.91005291

lfm2-24b-a2b

INT4-MIXED

32

140.3

19.1

52.35602094

llama-3.2-3b-instruct

INT4-MIXED

1024

119.7

19.1

52.35602094

qwen2.5-coder-3b-instruct

INT4-MIXED

1024

118.9

19.1

52.35602094

qwen2.5-coder-3b-instruct

INT4-MIXED

32

36.9

19.4

51.54639175

llama-3.2-3b-instruct

INT4-MIXED

1024

141.1

19.6

51.02040816

lfm2-24b-a2b

INT4-MIXED

1024

420

19.7

50.76142132

tiny-llama-1.1b-chat

FP16

32

21.5

19.7

50.76142132

qwen2.5-coder-3b-instruct

INT4-MIXED

1024

212.1

19.8

50.50505051

llama-3.2-3b-instruct

INT4-MIXED

1024

142.8

19.9

50.25125628

tiny-llama-1.1b-chat

FP16

1024

63.3

20.1

49.75124378

gemma-3-1b-it

FP16

32

24.2

20.3

49.26108374

qwen3-30b-a3b

INT4-MIXED

32

154.1

20.4

49.01960784

qwen3-30b-a3b

INT4-MIXED

32

154.5

20.4

49.01960784

whisper-large-v3

INT8-CW

32

180.9

20.4

49.01960784

whisper-large-v3

INT8-CW

1024

237.9

20.6

48.54368932

gemma-3-1b-it

FP16

1024

71.9

20.7

48.30917874

gpt-oss-20b

INT4-MIXED

32

121.9

22.2

45.04504505

qwen3-30b-a3b

INT4-MIXED

1024

461.5

22.4

44.64285714

qwen3-30b-a3b

INT4-MIXED

1024

467.1

22.4

44.64285714

qwen3.6-35b-a3b

INT4-MIXED

42

160.2

22.5

44.44444444

qwen3-4b

INT4-MIXED

32

36.5

22.8

43.85964912

phi-4-mini-instruct

INT4-MIXED

32

40.4

23.1

43.29004329

phi-4-mini-reasoning

INT4-MIXED

32

40.8

23.1

43.29004329

qwen3.6-35b-a3b

INT4-MIXED

42

169.1

23.1

43.29004329

gpt-oss-20b

INT4-MIXED

1024

354.3

23.2

43.10344828

llama-3.2-1b-instruct

FP16

32

25.2

23.3

42.91845494

lfm2-1.2b

FP16

32

28.4

23.4

42.73504274

lfm2.5-1.2b-instruct

FP16

32

29

23.4

42.73504274

qwen3-4b

INT4-MIXED

32

36.9

23.4

42.73504274

qwen3.6-35b-a3b

INT4-MIXED

1039

700.2

23.4

42.73504274

llama-3.2-1b-instruct

FP16

1024

74.7

23.6

42.37288136

lfm2-1.2b

FP16

1024

105.1

23.7

42.19409283

lfm2.5-1.2b-instruct

FP16

1024

104.8

23.7

42.19409283

phi-4-mini-instruct

INT4-MIXED

32

40.9

23.7

42.19409283

phi-4-mini-reasoning

INT4-MIXED

32

41.1

23.7

42.19409283

phi-4-mini-reasoning

INT4-MIXED

32

42.4

23.9

41.84100418

qwen3.6-35b-a3b

INT4-MIXED

1039

767.7

23.9

41.84100418

phi-4-mini-instruct

INT4-MIXED

1024

185.7

24

41.66666667

qwen3.5-35b-a3b

INT4-MIXED

83

340.1

24

41.66666667

phi-4-mini-reasoning

INT4-MIXED

1024

186

24.1

41.49377593

afm-4.5b

INT4-MIXED

32

31.3

24.2

41.32231405

qwen3-4b

INT4-MIXED

1024

170

24.2

41.32231405

phi-3.5-vision-instruct

INT4-MIXED

802

285.8

24.6

40.6504065

gemma-3-4b-it

INT4-MIXED

32

46.5

24.7

40.48582996

phi-4-mini-instruct

INT4-MIXED

1024

190.7

24.7

40.48582996

phi-4-mini-reasoning

INT4-MIXED

1024

190.9

24.7

40.48582996

phi-4-mini-instruct

INT4-MIXED

32

44.7

24.8

40.32258065

qwen3-4b

INT4-MIXED

1024

175.9

24.8

40.32258065

phi-4-mini-reasoning

INT4-MIXED

1024

213.4

24.9

40.16064257

exaone-4.0-1.2b

FP16

32

30.1

25

40

gemma-4-e2b-it

INT8-CW

274

197.4

25

40

qwen3.5-35b-a3b

INT4-MIXED

1024

862.8

25

40

afm-4.5b

INT4-MIXED

1024

192.8

25.2

39.68253968

gemma-3-4b-it

INT4-MIXED

32

47.5

25.3

39.5256917

trinity-mini

INT4-MIXED

32

165.7

25.4

39.37007874

exaone-4.0-1.2b

FP16

1024

104

25.5

39.21568627

gemma-3-4b-it

INT4-MIXED

32

51.4

25.5

39.21568627

phi-3.5-vision-instruct

INT4-MIXED

1032

337.2

25.5

39.21568627

phi-4-mini-instruct

INT4-MIXED

1024

250.1

25.7

38.91050584

gemma-4-e2b-it

INT8-CW

1024

319.4

25.8

38.75968992

gemma-3-4b-it

INT4-MIXED

1024

204.5

26.1

38.31417625

gemma-3-4b-it

INT4-MIXED

1024

215.2

26.7

37.45318352

trinity-mini

INT4-MIXED

1024

432.1

26.7

37.45318352

gpt-oss-20b

INT4-MIXED

32

125.7

26.8

37.31343284

gemma-3-4b-it

INT4-MIXED

1024

231.1

27

37.03703704

phi-2

INT8-CW

32

40.1

27.1

36.900369

qwen2.5-1.5b-instruct

FP4-NORMALIZED

32

32.1

27.5

36.36363636

deepseek-r1-distill-qwen-1.5b

FP4-NORMALIZED

32

30.3

27.6

36.23188406

gpt-oss-20b

INT4-MIXED

1024

375.9

27.7

36.10108303

deepseek-r1-distill-qwen-1.5b

FP4-NORMALIZED

1024

86.5

27.9

35.84229391

gemma-4-26b-a4b-it

INT4-MIXED

278

611.1

27.9

35.84229391

qwen2.5-1.5b-instruct

FP4-NORMALIZED

1024

86.7

27.9

35.84229391

minicpm-1b-sft

FP16

31

36.2

28

35.71428571

phi-4-multimodal-instruct

INT4-MIXED

578

274.6

28.1

35.58718861

minicpm3-4b

INT4-MIXED

32

108.6

28.3

35.33568905

phi-4-multimodal-instruct

INT4-MIXED

786

314.6

28.5

35.0877193

gemma-4-26b-a4b-it

INT4-MIXED

1024

911.2

29

34.48275862

minicpm-1b-sft

FP16

1014

89.2

29

34.48275862

phi-2

INT8-CW

1024

173.7

29

34.48275862

minicpm3-4b

INT4-MIXED

32

113.1

29.1

34.36426117

phi-4-multimodal-instruct

INT4-MIXED

1362

620.3

29.4

34.01360544

minicpm3-4b

INT4-MIXED

32

112.8

29.7

33.67003367

phi-4-multimodal-instruct

INT4-MIXED

1570

675.3

29.7

33.67003367

qwen2.5-1.5b-instruct

FP16

32

31.9

29.7

33.67003367

qwen2.5-coder-1.5b-instruct

FP16

32

32

29.7

33.67003367

deepseek-r1-distill-qwen-1.5b

FP16

32

32.1

29.8

33.55704698

lfm2-24b-a2b

INT8-CW

32

168.9

29.8

33.55704698

whisper-large-v3

FP16

32

177.7

30

33.33333333

deepseek-r1-distill-qwen-1.5b

FP16

1024

89

30.1

33.22259136

qwen2.5-1.5b-instruct

FP16

1024

90

30.1

33.22259136

qwen2.5-coder-1.5b-instruct

FP16

1024

89.9

30.1

33.22259136

whisper-large-v3

FP16

1024

234.8

30.1

33.22259136

lfm2-24b-a2b

INT8-CW

1024

478.7

30.2

33.11258278

smollm3-3b

INT8-CW

32

35.1

30.4

32.89473684

qwen2.5-coder-3b-instruct

INT8-CW

32

36.9

30.6

32.67973856

smollm3-3b

INT8-CW

1024

140.4

31.1

32.15434084

minicpm3-4b

INT4-MIXED

1024

476

31.2

32.05128205

qwen2.5-coder-3b-instruct

INT8-CW

1024

138.8

31.3

31.94888179

trinity-mini

INT8-CW

32

174.7

31.8

31.44654088

minicpm3-4b

INT4-MIXED

1024

488.1

32.1

31.15264798

llama-3.2-3b-instruct

INT8-CW

32

35.8

32.3

30.95975232

codellama-7b

INT4-MIXED

32

38.4

32.5

30.76923077

minicpm3-4b

INT4-MIXED

1024

527.7

32.5

30.76923077

llama-2-7b-chat-hf

INT4-MIXED

32

38.5

32.6

30.67484663

qwen3-vl-4b-instruct

INT4-MIXED

4907

3681.6

32.8

30.48780488

trinity-mini

INT8-CW

1024

471.8

32.8

30.48780488

qwen3-vl-4b-instruct

INT4-MIXED

4937

3830.8

32.9

30.39513678

llama-3.2-3b-instruct

INT8-CW

1024

136.9

33.2

30.12048193

phi-4-multimodal-instruct

INT4-MIXED

578

315.5

33.3

30.03003003

phi-4-multimodal-instruct

INT4-MIXED

786

372.3

33.6

29.76190476

qwen3-vl-4b-instruct

INT4-MIXED

4907

3716.7

33.6

29.76190476

qwen3-vl-4b-instruct

INT4-MIXED

4937

3860.1

33.6

29.76190476

llama-2-7b-chat-hf

INT4-MIXED

32

41.7

33.7

29.6735905

qwen3-vl-4b-instruct

INT4-MIXED

4907

3860.4

33.8

29.58579882

qwen3-vl-4b-instruct

INT4-MIXED

4937

4022.4

33.9

29.49852507

codellama-7b

INT4-MIXED

32

42.4

34

29.41176471

stable-diffusion-xl-1.0-inpainting-0.1

INT8-CW

32

34.5

34

29.41176471

ltx-video

INT4-MIXED

11

34.8

34.1

29.3255132

codellama-7b

INT4-MIXED

1024

236.9

34.3

29.15451895

llama-2-7b-chat-hf

INT4-MIXED

1024

236.6

34.3

29.15451895

biomistral-7b-slerp

INT4-MIXED

7

36.4

34.6

28.9017341

mistral-7b-instruct-v0.2

INT4-MIXED

32

41

34.6

28.9017341

neural-chat-7b-v3-3

INT4-MIXED

32

41.2

34.6

28.9017341

phi-4-multimodal-instruct

INT4-MIXED

1362

719.8

34.6

28.9017341

zephyr-7b-beta

INT4-MIXED

32

41.2

34.6

28.9017341

falcon-7b-instruct

INT4-MIXED

32

40.7

34.7

28.8184438

mistral-7b-instruct-v0.3

INT4-MIXED

32

41.3

34.7

28.8184438

phi-4-multimodal-instruct

INT4-MIXED

1570

789

34.8

28.73563218

stable-diffusion-xl-1.0-inpainting-0.1

INT8-CW

32

35.3

34.9

28.65329513

llama-2-7b-chat-hf

INT4-MIXED

1024

272

35.4

28.24858757

mistral-7b-instruct-v0.2

INT4-MIXED

1024

248.2

35.8

27.93296089

mistral-7b-instruct-v0.2

INT4-MIXED

32

44.5

35.8

27.93296089

mistral-7b-instruct-v0.3

INT4-MIXED

1024

248.5

35.8

27.93296089

mistral-7b-instruct-v0.3

INT4-MIXED

32

44.5

35.8

27.93296089

codellama-7b

INT4-MIXED

1024

285.8

35.9

27.8551532

zephyr-7b-beta

INT4-MIXED

1024

248

35.9

27.8551532

falcon-7b-instruct

INT4-MIXED

1024

258.9

36

27.77777778

neural-chat-7b-v3-3

INT4-MIXED

1024

246.8

36

27.77777778

biomistral-7b-slerp

INT4-MIXED

7

38.5

36.2

27.62430939

neural-chat-7b-v3-3

INT4-MIXED

32

45.4

36.2

27.62430939

mistral-7b-instruct-v0.1

INT4-MIXED

32

45.5

36.3

27.54820937

mistral-7b-instruct-v0.2

INT4-MIXED

32

45.3

36.3

27.54820937

mistral-7b-instruct-v0.3

INT4-MIXED

32

45.9

36.3

27.54820937

qwen2.5-7b-instruct

INT4-MIXED

32

44.7

36.4

27.47252747

qwen2.5-7b-instruct-1m

INT4-MIXED

32

42.8

36.4

27.47252747

mistral-7b-instruct-v0.2

INT4-MIXED

1024

260.5

36.9

27.100271

mistral-7b-instruct-v0.3

INT4-MIXED

1024

261.5

37

27.02702703

qwen2.5-vl-7b-instruct

INT4-MIXED

32

63.8

37.3

26.80965147

minicpm-v-2_6

INT4-MIXED

228

317.2

37.4

26.73796791

qwen2.5-7b-instruct

INT4-MIXED

1024

246.8

37.4

26.73796791

qwen2.5-7b-instruct

INT4-MIXED

32

45.5

37.4

26.73796791

qwen2.5-7b-instruct-1m

INT4-MIXED

1024

245.8

37.4

26.73796791

mistral-7b-instruct-v0.2

INT4-MIXED

1024

307.5

37.5

26.66666667

mistral-7b-instruct-v0.3

INT4-MIXED

1024

307.7

37.5

26.66666667

qwen2-7b-instruct

INT4-MIXED

32

45.6

37.5

26.66666667

deepseek-r1-distill-qwen-7b

INT4-MIXED

32

45.2

37.6

26.59574468

minicpm-o-2_6

INT4-MIXED

238

312.3

37.6

26.59574468

mistral-7b-instruct-v0.1

INT4-MIXED

1025

334.4

37.6

26.59574468

neural-chat-7b-v3-3

INT4-MIXED

1024

308.9

37.6

26.59574468

ltx-video

INT8-CW

11

38.3

37.8

26.45502646

deepseek-r1-distill-qwen-7b

INT4-MIXED

32

46.1

37.9

26.38522427

qwen2-7b-instruct

INT4-MIXED

32

46.2

37.9

26.38522427

qwen2.5-7b-instruct

INT4-MIXED

32

46.1

37.9

26.38522427

qwen2.5-7b-instruct-1m

INT4-MIXED

32

46.3

37.9

26.38522427

falcon-7b-instruct

INT4-MIXED

32

47.7

38.1

26.24671916

llama-3.1-8b-instruct

INT4-MIXED

32

44.7

38.1

26.24671916

deepseek-r1-distill-llama-8b

INT4-MIXED

32

44.8

38.3

26.10966057

llama-3-8b-instruct

INT4-MIXED

32

45.4

38.4

26.04166667

qwen2.5-vl-7b-instruct

INT4-MIXED

32

58.6

38.4

26.04166667

minicpm-v-2_6

INT4-MIXED

228

320.3

38.5

25.97402597

qwen2-7b-instruct

INT4-MIXED

1024

255.6

38.5

25.97402597

qwen2.5-7b-instruct

INT4-MIXED

1024

254.2

38.5

25.97402597

qwen2.5-vl-7b-instruct

INT4-MIXED

1024

288.8

38.5

25.97402597

deepseek-r1-distill-qwen-7b

INT4-MIXED

1024

255.2

38.6

25.90673575

lcm-dreamshaper-v7

INT8-HYBRID

32

39.7

38.6

25.90673575

bloomz-7b1

INT4-MIXED

32

45.9

38.7

25.83979328

lcm-dreamshaper-v7

INT8-HYBRID

1024

39.8

38.7

25.83979328

deepseek-r1-distill-qwen-7b

INT4-MIXED

1024

310.3

38.8

25.77319588

fara-7b

INT4-MIXED

32

75.3

38.9

25.70694087

minicpm4-8b

INT4-MIXED

32

47.6

38.9

25.70694087

qwen2-7b-instruct

INT4-MIXED

1024

310.3

38.9

25.70694087

qwen2.5-7b-instruct

INT4-MIXED

1024

311.4

38.9

25.70694087

qwen2.5-7b-instruct-1m

INT4-MIXED

1024

311.5

38.9

25.70694087

minicpm-v-2_6

INT4-MIXED

228

334.5

39

25.64102564

phi-4-mini-instruct

INT8-CW

32

44.2

39.1

25.57544757

phi-4-mini-reasoning

INT8-CW

32

44.6

39.1

25.57544757

qwen3-4b

INT8-CW

32

44.6

39.1

25.57544757

llama-3.1-8b-instruct

INT4-MIXED

32

48.4

39.2

25.51020408

minicpm-o-2_6

INT4-MIXED

238

326.7

39.2

25.51020408

llama-3-8b-instruct

INT4-MIXED

32

47.9

39.3

25.44529262

llama-3.1-8b-instruct

INT4-MIXED

1024

251.1

39.3

25.44529262

qwen3-8b

INT4-MIXED

32

46.4

39.3

25.44529262

deepseek-r1-distill-llama-8b

INT4-MIXED

1024

251.3

39.4

25.38071066

deepseek-r1-distill-llama-8b

INT4-MIXED

32

48.2

39.4

25.38071066

falcon-7b-instruct

INT4-MIXED

1024

333.1

39.4

25.38071066

gemma-3-4b-it

INT8-CW

32

51.5

39.4

25.38071066

llama-3-8b-instruct

INT4-MIXED

32

48.4

39.4

25.38071066

qwen2.5-vl-7b-instruct

INT4-MIXED

32

59.9

39.4

25.38071066

llama-3-8b-instruct

INT4-MIXED

1024

256.5

39.5

25.3164557

qwen2.5-vl-7b-instruct

INT4-MIXED

1024

300.4

39.5

25.3164557

llama-3-8b-instruct

INT4-MIXED

32

49.1

39.8

25.12562814

fara-7b

INT4-MIXED

1024

459.7

39.9

25.06265664

phi-4-mini-instruct

INT8-CW

1024

206.2

40

25

qwen2.5-vl-7b-instruct

INT4-MIXED

1024

352.7

40

25

bloomz-7b1

INT4-MIXED

32

49.8

40.1

24.93765586

phi-4-mini-reasoning

INT8-CW

1024

206.2

40.1

24.93765586

minicpm4-8b

INT4-MIXED

1024

279

40.2

24.87562189

minicpm4-8b

INT4-MIXED

32

50.5

40.2

24.87562189

bloomz-7b1

INT4-MIXED

1024

329.1

40.4

24.75247525

qwen3-4b

INT8-CW

1024

189.3

40.4

24.75247525

qwen3-8b

INT4-MIXED

32

49.1

40.4

24.75247525

deepseek-r1-distill-llama-8b

INT4-MIXED

1024

265.4

40.5

24.69135802

llama-3.1-8b-instruct

INT4-MIXED

1024

297.5

40.5

24.69135802

minicpm-v-4_5

INT4-MIXED

217

327.8

40.5

24.69135802

llama-3-8b-instruct

INT4-MIXED

1024

265.5

40.6

24.63054187

llama-3-8b-instruct

INT4-MIXED

1024

265.4

40.6

24.63054187

qwen3-8b

INT4-MIXED

1024

264.3

40.6

24.63054187

minicpm4-8b

INT4-MIXED

32

51.9

40.7

24.57002457

gemma-3-4b-it

INT8-CW

1024

218.4

40.9

24.44987775

qwen3-8b

INT4-MIXED

32

50.2

40.9

24.44987775

llama-3-8b-instruct

INT4-MIXED

1024

312.3

41

24.3902439

afm-4.5b

INT8-CW

32

45.7

41.1

24.33090024

gemma-4-e2b-it

FP4-NORMALIZED

274

195.1

41.5

24.09638554

minicpm4-8b

INT4-MIXED

1024

299.1

41.6

24.03846154

phi-3.5-vision-instruct

INT8-CW

802

278

41.7

23.98081535

afm-4.5b

INT8-CW

1024

183

41.8

23.92344498

bloomz-7b1

INT4-MIXED

1024

388.6

41.8

23.92344498

qwen3-8b

INT4-MIXED

1024

277.2

41.8

23.92344498

zephyr-7b-beta

INT4-MIXED

32

51.3

41.9

23.86634845

minicpm4-8b

INT4-MIXED

1024

341.3

42.1

23.75296912

gemma-4-e2b-it

FP4-NORMALIZED

1024

342

42.2

23.69668246

minicpm-v-4_5

INT4-MIXED

217

353.2

42.2

23.69668246

qwen3-8b

INT4-MIXED

1024

318.8

42.4

23.58490566

phi-3.5-vision-instruct

INT8-CW

1032

338.5

42.6

23.4741784

zephyr-7b-beta

INT4-MIXED

1024

267.5

43.2

23.14814815

gpt-oss-20b

INT8-CW

32

187.2

43.3

23.09468822

qwen3.5-9b

INT4-MIXED

83

109.2

43.4

23.04147465

qwen3_8b_draft_eagle3

INT4-MIXED

32

53

43.4

23.04147465

videochat-flash-qwen2_5-7b_internvideo2-1b

INT4-MIXED

14531

26346.6

43.5

22.98850575

gpt-oss-20b

INT8-CW

1024

592.7

44

22.72727273

qwen3.5-9b

INT4-MIXED

1024

446.6

44

22.72727273

gemma-7b-it

INT4-MIXED

32

53.2

44.2

22.62443439

qwen3_8b_draft_eagle3

INT4-MIXED

1024

319.7

44.3

22.57336343

qwen3.5-9b

INT4-MIXED

83

114.2

45.1

22.172949

gemma-4-e2b-it

FP16

274

193.2

45.3

22.07505519

gemma-7b-it

INT4-MIXED

1024

313.8

45.7

21.88183807

qwen3.5-9b

INT4-MIXED

1024

505.4

45.7

21.88183807

gemma-4-e2b-it

FP16

1024

341.3

45.8

21.83406114

gemma-7b-it

INT4-MIXED

32

56.6

45.8

21.83406114

phi-4-multimodal-instruct

INT8-CW

578

291.1

46.7

21.41327623

minicpm3-4b

INT8-CW

32

120.4

46.9

21.32196162

phi-4-multimodal-instruct

INT8-CW

786

344.1

47

21.27659574

deepseek-r1-distill-llama-8b

INT4-MIXED

32

61.1

47.1

21.23142251

lcm-dreamshaper-v7

INT8-CW

32

47.7

47.1

21.23142251

llama-3.1-8b-instruct

INT4-MIXED

32

61.4

47.1

21.23142251

lcm-dreamshaper-v7

INT8-CW

1024

48.8

47.3

21.14164905

gemma-7b-it

INT4-MIXED

1024

380.7

47.4

21.09704641

lcm-dreamshaper-v7

INT8-CW

32

48.1

47.4

21.09704641

lcm-dreamshaper-v7

INT8-CW

1024

48.8

47.6

21.00840336

phi-4-multimodal-instruct

INT8-CW

1362

660.6

47.9

20.87682672

phi-4-multimodal-instruct

INT8-CW

1570

716.5

48.1

20.79002079

llama-3.1-8b-instruct

INT4-MIXED

1024

357

48.3

20.70393375

deepseek-r1-distill-llama-8b

INT4-MIXED

1024

360.5

48.5

20.6185567

lcm-dreamshaper-v7

FP16

1024

49.5

49

20.40816327

lcm-dreamshaper-v7

FP16

32

49.5

49

20.40816327

gemma-2-9b-it

INT4-MIXED

32

60.2

49.6

20.16129032

phi-2

FP16

32

53.8

50.7

19.72386588

gemma-2-9b-it

INT4-MIXED

32

63.8

51

19.60784314

gemma-2-9b-it

INT4-MIXED

1024

364.2

51.3

19.49317739

gemma-2-9b-it

INT4-MIXED

32

64.8

51.5

19.41747573

minicpm3-4b

INT8-CW

1024

506.2

51.6

19.37984496

gemma-2-9b-it

INT4-MIXED

1024

378.9

52.8

18.93939394

stable-diffusion-xl-1.0-inpainting-0.1

FP16

32

53.5

52.9

18.90359168

gemma-2-9b-it

INT4-MIXED

1024

447.7

53.3

18.76172608

phi-2

FP16

1024

215.5

53.6

18.65671642

ltx-video

FP16

11

54.5

53.9

18.5528757

llama-3.2-3b-instruct

FP4-NORMALIZED

32

59.2

56.1

17.82531194

llama-3.2-3b-instruct

FP4-NORMALIZED

1024

192.2

57

17.54385965

smollm3-3b

FP16

32

60.4

57

17.54385965

qwen2.5-coder-3b-instruct

FP16

32

60.8

57.3

17.45200698

smollm3-3b

FP16

1024

176.9

57.7

17.33102253

qwen2.5-coder-3b-instruct

FP16

1024

177.4

57.8

17.30103806

dolly-v2-12b

INT4-MIXED

32

72.8

59.4

16.83501684

llama-3.2-3b-instruct

FP16

32

62.4

59.7

16.75041876

llama-3.2-3b-instruct

FP16

1024

195

61

16.39344262

llama-2-13b-chat-hf

INT4-MIXED

32

70.2

61.3

16.3132137

codellama-7b

INT8-CW

32

67.6

61.7

16.20745543

llama-2-7b-chat-hf

INT8-CW

32

67.4

61.7

16.20745543

dolly-v2-12b

INT4-MIXED

1024

661.1

61.9

16.15508885

qwen3_8b_draft_eagle3

INT8-CW

1024

312.1

62

16.12903226

gemma-3-12b-it

INT4-MIXED

32

85.7

63.3

15.79778831

qwen3_8b_draft_eagle3

INT8-CW

32

80.8

63.6

15.72327044

llama-2-13b-chat-hf

INT4-MIXED

1024

458

64

15.625

mixtral-8x7b-v0.1

INT4-MIXED

32

1153.6

64.1

15.60062402

llama-2-7b-chat-hf

INT8-CW

1024

273.9

64.3

15.55209953

codellama-7b

INT8-CW

1024

272.3

64.4

15.52795031

falcon-7b-instruct

INT8-CW

32

69.9

64.5

15.50387597

mixtral-8x7b-v0.1

INT4-MIXED

1024

1099.5

65.4

15.29051988

biomistral-7b-slerp

INT8-CW

7

67.7

65.5

15.26717557

falcon-7b-instruct

INT8-CW

1024

390.4

65.5

15.26717557

mistral-7b-instruct-v0.1

INT8-CW

32

72.5

65.6

15.24390244

mistral-7b-instruct-v0.2

INT8-CW

32

72

65.6

15.24390244

mistral-7b-instruct-v0.3

INT8-CW

32

72.5

65.6

15.24390244

zephyr-7b-beta

INT8-CW

32

72.4

65.6

15.24390244

gemma-3-12b-it

INT4-MIXED

32

85.8

65.9

15.17450683

neural-chat-7b-v3-3

INT8-CW

32

72.2

65.9

15.17450683

phi-4-mini-instruct

FP4-NORMALIZED

32

70.9

66.1

15.12859304

phi-4-mini-reasoning

FP4-NORMALIZED

32

71.2

66.2

15.10574018

qwen2-7b-instruct

INT8-CW

32

72.4

66.3

15.08295626

qwen2.5-7b-instruct

INT8-CW

32

72.5

66.5

15.03759398

deepseek-r1-distill-qwen-7b

INT8-CW

32

72.4

66.6

15.01501502

qwen2.5-7b-instruct-1m

INT8-CW

32

72.6

66.7

14.99250375

mistral-7b-instruct-v0.3

INT8-CW

1024

294.7

66.8

14.97005988

mistral-7b-instruct-v0.1

INT8-CW

1025

318.6

66.9

14.94768311

mistral-7b-instruct-v0.2

INT8-CW

1024

296.1

66.9

14.94768311

qwen2-7b-instruct

INT8-CW

1024

292.1

67

14.92537313

zephyr-7b-beta

INT8-CW

1024

294.5

67

14.92537313

phi-4-mini-instruct

FP4-NORMALIZED

1024

281

67.1

14.90312966

deepseek-r1-distill-qwen-7b

INT8-CW

1024

290.3

67.2

14.88095238

gemma-3-12b-it

INT4-MIXED

1024

538.3

67.2

14.88095238

neural-chat-7b-v3-3

INT8-CW

1024

292.9

67.2

14.88095238

phi-4-mini-reasoning

FP4-NORMALIZED

1024

266.6

67.2

14.88095238

qwen2.5-7b-instruct

INT8-CW

1024

291.1

67.2

14.88095238

minicpm-v-2_6

INT8-CW

228

334.7

67.3

14.85884101

minicpm-o-2_6

INT8-CW

238

338.8

67.5

14.81481481

gemma-3-4b-it

FP4-NORMALIZED

32

73.1

67.6

14.79289941

qwen2.5-7b-instruct-1m

INT8-CW

1024

292.5

67.6

14.79289941

qwen2.5-vl-7b-instruct

INT8-CW

32

86.8

67.6

14.79289941

fara-7b

INT8-CW

32

102

67.7

14.77104874

lcm-sdxl

INT8-HYBRID

1024

68.6

68

14.70588235

lcm-sdxl

INT8-HYBRID

32

68.5

68

14.70588235

bloomz-7b1

INT8-CW

32

72.8

68.1

14.68428781

phi-4

INT4-MIXED

32

80.7

68.4

14.61988304

phi-4-reasoning

INT4-MIXED

32

79.1

68.4

14.61988304

qwen2.5-vl-7b-instruct

INT8-CW

1024

336.5

68.5

14.59854015

fara-7b

INT8-CW

1024

423.7

68.7

14.55604076

gemma-3-4b-it

FP4-NORMALIZED

1024

277.7

69

14.49275362

llama-3.1-8b-instruct

INT8-CW

32

76

69.1

14.47178003

qwen1.5-14b-chat

INT4-MIXED

32

81.2

69.2

14.45086705

deepseek-r1-distill-llama-8b

INT8-CW

32

75.8

69.3

14.43001443

lcm-sdxl

INT8-CW

32

69.9

69.3

14.43001443

llama-3-8b-instruct

INT8-CW

32

75.5

69.3

14.43001443

phi-3-mini-4k-instruct

FP16

32

72.9

69.3

14.43001443

phi-3.5-mini-instruct

FP16

32

72.8

69.3

14.43001443

lcm-sdxl

INT8-CW

1024

70.1

69.4

14.4092219

phi-3-mini-128k-instruct

FP16

32

73

69.4

14.4092219

gemma-3-12b-it

INT4-MIXED

1024

608.1

69.8

14.32664756

lcm-sdxl

INT8-CW

1024

70.7

70.1

14.26533524

phi-4-reasoning

INT4-MIXED

1024

518.6

70.1

14.26533524

stable-diffusion-v1-5

INT8-HYBRID

1024

71.7

70.1

14.26533524

phi-4

INT4-MIXED

1024

523.7

70.2

14.24501425

qwen3-8b

INT8-CW

32

77.9

70.2

14.24501425

lcm-sdxl

INT8-CW

32

71

70.4

14.20454545

llama-3-8b-instruct

INT8-CW

1024

296.1

70.4

14.20454545

llama-3.1-8b-instruct

INT8-CW

1024

297.5

70.4

14.20454545

stable-diffusion-v1-5

INT8-HYBRID

32

71.3

70.4

14.20454545

deepseek-r1-distill-llama-8b

INT8-CW

1024

297.5

70.5

14.18439716

phi-4-reasoning

INT4-MIXED

32

83.7

70.6

14.16430595

bloomz-7b1

INT8-CW

1024

401.2

70.7

14.14427157

llava-v1.6-mistral-7b-hf

INT8-CW

2944

1588.3

71.1

14.06469761

minicpm-v-4_5

INT8-CW

217

352.5

71.3

14.02524544

phi-4

INT4-MIXED

32

85.8

71.4

14.00560224

phi-4-mini-reasoning

FP16

32

76.3

71.6

13.96648045

qwen3-8b

INT8-CW

1024

306.1

71.6

13.96648045

phi-4-mini-instruct

FP16

32

76.3

71.7

13.94700139

qwen1.5-14b-chat

INT4-MIXED

1024

573.9

71.9

13.90820584

deepseek-r1-distill-qwen-14b

INT4-MIXED

32

86.4

72.4

13.8121547

phi-4-reasoning

INT4-MIXED

1024

558.5

72.5

13.79310345

phi-3.5-vision-instruct

FP16

802

356.8

72.6

13.77410468

phi-3.5-mini-instruct

FP16

1024

274.2

72.9

13.71742112

videochat-flash-qwen2_5-7b_internvideo2-1b

INT8-CW

14531

26356.7

72.9

13.71742112

phi-3-mini-4k-instruct

FP16

1024

272.6

73

13.69863014

phi-3-mini-128k-instruct

FP16

1024

274.4

73.1

13.67989056

gemma-3-4b-it

FP16

32

79

73.2

13.66120219

phi-4-mini-instruct

FP16

1024

274.1

73.2

13.66120219

phi-4-mini-reasoning

FP16

1024

276.8

73.2

13.66120219

phi-3.5-vision-instruct

FP16

1032

452.4

73.4

13.6239782

phi-4

INT4-MIXED

1024

645.9

73.4

13.6239782

sd-turbo

INT8-HYBRID

32

74.9

74

13.51351351

sd-turbo

INT8-HYBRID

1024

74.9

74.1

13.49527665

stable-diffusion-v2-1

INT8-HYBRID

32

74.8

74.1

13.49527665

stable-diffusion-v2-1

INT8-HYBRID

1024

74.9

74.3

13.4589502

phi-4-reasoning

INT4-MIXED

32

98.3

74.4

13.44086022

qwen3.5-9b

INT8-CW

83

135.3

74.5

13.42281879

deepseek-r1-distill-qwen-14b

INT4-MIXED

1024

615.5

74.6

13.40482574

gemma-3-4b-it

FP16

1024

282.4

74.6

13.40482574

minicpm4-8b

INT8-CW

32

81.3

74.7

13.38688086

qwen3-4b

FP16

32

78.8

74.7

13.38688086

qwen3.5-9b

INT8-CW

1024

482.6

75.1

13.31557923

minicpm4-8b

INT8-CW

1024

360.3

75.5

13.24503311

llama-2-13b-chat-hf

INT4-MIXED

32

94.2

75.9

13.17523057

phi-4-reasoning

INT4-MIXED

1024

722.3

76.2

13.12335958

starcoder

INT4-MIXED

32

89.3

76.4

13.08900524

qwen3-4b

FP16

1024

253.7

76.5

13.07189542

llama-2-13b-chat-hf

INT4-MIXED

1024

568.9

78.6

12.72264631

starcoder

INT4-MIXED

1024

704.6

78.6

12.72264631

afm-4.5b

FP16

32

83.4

80.2

12.46882793

minicpm3-4b

FP4-NORMALIZED

32

103.6

80.7

12.39157373

starcoder2-15b

INT4-MIXED

32

104.9

80.8

12.37623762

afm-4.5b

FP16

1024

255.2

81.1

12.33045623

gemma-7b-it

INT8-CW

32

89.3

81.1

12.33045623

phi-4-multimodal-instruct

FP16

578

344.9

82.1

12.18026797

phi-4-multimodal-instruct

FP16

786

401

82.5

12.12121212

stable-diffusion-v1-5

INT8-CW

32

83.4

82.6

12.10653753

stable-diffusion-v1-5

INT8-CW

1024

85.2

82.7

12.09189843

minicpm3-4b

FP16

32

102.5

82.8

12.07729469

stable-diffusion-v1-5

INT8-CW

1024

83.6

83.1

12.03369434

stable-diffusion-v1-5

INT8-CW

32

83.9

83.1

12.03369434

starcoder2-15b

INT4-MIXED

1024

851.5

83.2

12.01923077

phi-4-multimodal-instruct

FP16

1362

791.9

83.4

11.99040767

gemma-7b-it

INT8-CW

1024

385

83.8

11.93317422

phi-4-multimodal-instruct

FP16

1570

871.9

83.8

11.93317422

sd-turbo

INT8-CW

32

84.7

84

11.9047619

stable-diffusion-v2-1

INT8-CW

32

84.3

84

11.9047619

stable-diffusion-v2-1

INT8-CW

1024

84.7

84.1

11.89060642

qwen3-vl-4b-instruct

FP16

4937

4621.8

84.5

11.83431953

stable-diffusion-v1-5

FP16

32

85.1

84.5

11.83431953

qwen3-vl-4b-instruct

FP16

4907

4396.1

84.6

11.82033097

stable-diffusion-v2-1

INT8-CW

32

85.1

84.6

11.82033097

sd-turbo

INT8-CW

32

85.4

84.7

11.80637544

stable-diffusion-v1-5

FP16

1024

84.8

84.7

11.80637544

stable-diffusion-v2-1

INT8-CW

1024

85.1

84.8

11.79245283

sd-turbo

INT8-CW

1024

85.4

85

11.76470588

sd-turbo

INT8-CW

1024

84.4

85

11.76470588

minicpm3-4b

FP4-NORMALIZED

1024

561.9

85.5

11.69590643

stable-diffusion-v2-1

FP16

32

86

85.7

11.66861144

sd-turbo

FP16

32

85.6

86.1

11.61440186

stable-diffusion-v2-1

FP16

1024

85.7

86.3

11.58748552

sd-turbo

FP16

1024

85.8

86.6

11.54734411

gemma-2-9b-it

INT8-CW

32

96.9

88.2

11.33786848

lcm-sdxl

FP16

1024

89

88.5

11.29943503

lcm-sdxl

FP16

32

88.9

88.5

11.29943503

minicpm3-4b

FP16

1024

570.2

90

11.11111111

gemma-2-9b-it

INT8-CW

1024

428.3

90.6

11.03752759

Topology

Precision

Input Size

1st latency (ms)

2nd latency (ms)

2nd token per sec

distil-large-v2

INT4-MIXED

32

248.2

8.7

114.9425287

distil-large-v2

INT8-CW

1024

284.5

8.8

113.6363636

distil-large-v2

INT4-MIXED

1024

296.5

8.9

112.3595506

distil-large-v2

INT8-CW

32

230.5

9

111.1111111

minicpm4-0.5b

INT4-MIXED

32

35.6

9.1

109.8901099

whisper-large-v3-turbo

INT8-CW

32

242.7

9.3

107.5268817

whisper-large-v3-turbo

INT8-CW

1024

286.8

9.4

106.3829787

whisper-large-v3-turbo

INT4-MIXED

1024

309.5

9.5

105.2631579

whisper-large-v3-turbo

INT4-MIXED

32

262.9

9.6

104.1666667

minicpm4-0.5b

INT4-MIXED

1024

60.6

10

100

minicpm4-0.5b

INT4-MIXED

32

33.8

10

100

minicpm4-0.5b

INT4-MIXED

32

36.8

10.4

96.15384615

minicpm4-0.5b

INT4-MIXED

1024

70.8

11

90.90909091

minicpm4-0.5b

INT8-CW

32

38

11.1

90.09009009

minicpm4-0.5b

INT4-MIXED

1024

66

11.3

88.49557522

nuextract-tiny

INT4-MIXED

32

32.7

11.3

88.49557522

gemma-3-270m

INT4-MIXED

32

39.6

11.7

85.47008547

nuextract-tiny

INT8-CW

32

44.7

11.8

84.74576271

gemma-3-270m

INT4-MIXED

1024

53.4

12.1

82.6446281

nuextract-tiny

INT4-MIXED

1024

73.1

12.1

82.6446281

lfm2.5-350m

INT4-MIXED

32

45.2

12.4

80.64516129

minicpm4-0.5b

INT8-CW

1024

65.5

12.4

80.64516129

qwen2.5-coder-0.5b-instruct

INT4-MIXED

32

39.3

12.5

80

distil-large-v2

FP16

32

263.7

12.7

78.74015748

nuextract-tiny

INT8-CW

1024

70.3

12.7

78.74015748

qwen2.5-coder-0.5b-instruct

INT4-MIXED

1024

71.7

12.7

78.74015748

distil-large-v2

FP16

1024

320.6

12.8

78.125

qwen2.5-coder-0.5b-instruct

INT4-MIXED

32

45.6

13

76.92307692

whisper-large-v3-turbo

FP16

32

285

13.1

76.33587786

whisper-small

INT4-MIXED

32

128.8

13.2

75.75757576

whisper-small

INT4-MIXED

1024

179.5

13.2

75.75757576

gemma-3-270m

INT8-CW

32

37.8

13.3

75.18796992

whisper-large-v3-turbo

FP16

1024

346.3

13.3

75.18796992

qwen2.5-coder-0.5b-instruct

INT4-MIXED

1024

76.2

13.5

74.07407407

gemma-3-270m

INT8-CW

1024

53.2

13.7

72.99270073

lfm2.5-350m

INT8-CW

32

41.4

13.7

72.99270073

lfm2.5-350m

INT4-MIXED

1024

90.4

14.1

70.92198582

lfm2.5-350m

INT8-CW

1024

90.8

14.1

70.92198582

whisper-small

INT8-CW

32

128.7

14.1

70.92198582

qwen2.5-coder-0.5b-instruct

INT4-MIXED

32

47.7

14.2

70.42253521

qwen2.5-coder-0.5b-instruct

INT4-MIXED

1024

67.7

14.7

68.02721088

whisper-small

INT8-CW

1024

185.8

14.9

67.11409396

gemma-3-270m

FP16

1024

50.9

15

66.66666667

gemma-3-270m

FP16

32

37.5

15.1

66.22516556

whisper-small

INT4-MIXED

1024

203.3

15.2

65.78947368

minicpm5-1b

INT4-MIXED

32

46.8

15.6

64.1025641

whisper-small

INT4-MIXED

32

125

15.7

63.69426752

tiny-llama-1.1b-chat

INT4-MIXED

32

49.5

16.1

62.11180124

whisper-small

INT4-MIXED

32

133.5

16.5

60.60606061

llama-3.2-1b-instruct

INT4-MIXED

32

59.4

16.6

60.24096386

tiny-llama-1.1b-chat

INT4-MIXED

1024

141.9

16.6

60.24096386

minicpm5-1b

INT4-MIXED

1024

120

16.7

59.88023952

minicpm5-1b

INT8-CW

32

48.3

16.8

59.52380952

whisper-small

FP16

1024

183.8

17

58.82352941

whisper-small

FP16

32

128.1

17.3

57.80346821

whisper-small

INT4-MIXED

1024

193.2

17.3

57.80346821

tiny-llama-1.1b-chat

INT4-MIXED

32

36

17.5

57.14285714

qwen2.5-coder-0.5b-instruct

INT8-CW

32

62.8

18

55.55555556

tiny-llama-1.1b-chat

INT4-MIXED

1024

119.3

18.1

55.24861878

llama-3.2-1b-instruct

INT4-MIXED

1024

130.2

18.2

54.94505495

minicpm5-1b

INT8-CW

1024

106.1

18.4

54.34782609

qwen2.5-coder-0.5b-instruct

INT8-CW

1024

81.6

18.4

54.34782609

qwen3.5-0.8b

INT4-MIXED

85

110.7

18.5

54.05405405

qwen3.5-0.8b

INT4-MIXED

1024

241.3

18.7

53.47593583

tiny-llama-1.1b-chat

INT8-CW

32

42

18.8

53.19148936

qwen3.5-0.8b

INT4-MIXED

85

103.5

18.9

52.91005291

qwen3.5-0.8b

INT4-MIXED

85

131.1

19.3

51.8134715

nanollava

INT4-MIXED

760

150

19.4

51.54639175

qwen3.5-0.8b

INT4-MIXED

1024

234.1

19.6

51.02040816

tiny-llama-1.1b-chat

INT8-CW

1024

122.6

19.7

50.76142132

qwen3.5-0.8b

INT4-MIXED

1024

256.9

19.8

50.50505051

minicpm4-0.5b

FP16

32

42.2

20

50

qwen3.5-0.8b

INT8-CW

85

124.4

20

50

nanollava

INT8-CW

760

157.2

20.1

49.75124378

qwen3.5-0.8b

INT8-CW

1024

251.6

20.5

48.7804878

llama-3.2-1b-instruct

INT4-MIXED

32

38.4

20.8

48.07692308

exaone-4.0-1.2b

INT4-MIXED

32

66.3

21.1

47.39336493

llama-3.2-1b-instruct

INT8-CW

32

38

21.1

47.39336493

llama-3.2-1b-instruct

INT4-MIXED

1024

112.5

21.3

46.94835681

minicpm4-0.5b

FP16

1024

88.3

21.4

46.72897196

nanollava

INT8-CW

1752

267.3

21.7

46.08294931

qwen2.5-coder-1.5b-instruct

INT4-MIXED

32

81.5

21.7

46.08294931

lfm2.5-1.2b-instruct

INT4-MIXED

32

75.3

21.8

45.87155963

gemma-3-1b-it

INT4-MIXED

1024

118.1

22

45.45454545

gemma-3-1b-it

INT4-MIXED

32

52.2

22.3

44.84304933

lfm2.5-350m

FP16

32

54.7

22.3

44.84304933

llama-3.2-1b-instruct

INT8-CW

1024

129.8

22.4

44.64285714

qwen2.5-coder-1.5b-instruct

INT4-MIXED

1024

148.9

22.4

44.64285714

exaone-4.0-1.2b

INT4-MIXED

1024

202.4

22.5

44.44444444

lfm2-1.2b

INT4-MIXED

32

72.2

22.5

44.44444444

nanollava

INT4-MIXED

1752

279

22.6

44.24778761

exaone-4.0-1.2b

INT4-MIXED

32

83.5

22.9

43.66812227

lfm2-1.2b

INT4-MIXED

1024

165.2

22.9

43.66812227

gemma-3-1b-it

INT4-MIXED

32

74.7

23

43.47826087

gemma-3-1b-it

INT4-MIXED

32

79.1

23.1

43.29004329

lfm2.5-1.2b-instruct

INT4-MIXED

1024

166.3

23.1

43.29004329

gemma-3-1b-it

INT4-MIXED

1024

134.5

23.3

42.91845494

lfm2.5-350m

FP16

1024

98.2

23.4

42.73504274

gemma-3-1b-it

INT4-MIXED

1024

121.6

23.5

42.55319149

qwen2.5-coder-1.5b-instruct

INT4-MIXED

32

89.6

23.8

42.01680672

lfm2.5-1.2b-instruct

INT4-MIXED

32

79.6

24

41.66666667

nuextract-tiny

FP16

32

57.3

24

41.66666667

lfm2-1.2b

INT4-MIXED

32

80.5

24.2

41.32231405

qwen2.5-coder-0.5b-instruct

FP16

32

70.2

24.4

40.98360656

qwen2.5-1.5b-instruct

INT4-MIXED

32

87.6

24.6

40.6504065

qwen2.5-coder-1.5b-instruct

INT4-MIXED

32

72.9

24.6

40.6504065

exaone-4.0-1.2b

INT8-CW

32

75.6

24.7

40.48582996

lfm2.5-1.2b-instruct

INT4-MIXED

1024

204.2

24.8

40.32258065

lfm2-1.2b

INT4-MIXED

1024

196.5

24.9

40.16064257

exaone-4.0-1.2b

INT4-MIXED

1024

220.3

25

40

qwen2.5-1.5b-instruct

INT4-MIXED

1024

156.2

25.1

39.84063745

qwen2.5-1.5b-instruct

INT4-MIXED

32

68.8

25.1

39.84063745

qwen2.5-coder-1.5b-instruct

INT4-MIXED

1024

187.7

25.1

39.84063745

qwen2.5-coder-0.5b-instruct

FP16

1024

86.5

25.2

39.68253968

nuextract-tiny

FP16

1024

88.5

25.3

39.5256917

qwen2.5-coder-1.5b-instruct

INT4-MIXED

1024

156.1

25.4

39.37007874

minicpm-1b-sft

INT4-MIXED

31

102.8

25.5

39.21568627

qwen2.5-1.5b-instruct

INT4-MIXED

1024

182.9

25.5

39.21568627

deepseek-r1-distill-qwen-1.5b

INT4-MIXED

32

59.5

25.6

39.0625

exaone-4.0-1.2b

INT8-CW

1024

203

26

38.46153846

deepseek-r1-distill-qwen-1.5b

INT4-MIXED

32

84.5

26.1

38.31417625

deepseek-r1-distill-qwen-1.5b

INT4-MIXED

1024

157.7

26.1

38.31417625

lfm2.5-1.2b-instruct

INT4-MIXED

32

59

26.1

38.31417625

exaone-4.0-1.2b

INT4-MIXED

32

57.1

26.5

37.73584906

lfm2-1.2b

INT4-MIXED

32

61.4

26.7

37.45318352

deepseek-r1-distill-qwen-1.5b

INT4-MIXED

1024

199.9

26.8

37.31343284

minicpm-1b-sft

INT4-MIXED

31

113.4

26.9

37.17472119

minicpm-1b-sft

INT4-MIXED

1014

174.9

27

37.03703704

exaone-4.0-1.2b

INT4-MIXED

1024

201.2

27.1

36.900369

deepseek-r1-distill-qwen-1.5b

INT8-CW

32

66.1

27.2

36.76470588

lfm2-1.2b

INT4-MIXED

1024

175.8

27.2

36.76470588

lfm2.5-1.2b-instruct

INT4-MIXED

1024

173.8

27.2

36.76470588

qwen2.5-coder-1.5b-instruct

INT8-CW

32

56

27.3

36.63003663

qwen2.5-1.5b-instruct

INT8-CW

32

76

27.4

36.49635036

gemma-3-1b-it

INT8-CW

32

102

27.9

35.84229391

qwen2.5-1.5b-instruct

INT8-CW

1024

170.2

28

35.71428571

lfm2.5-1.2b-instruct

INT8-CW

32

76

28.2

35.46099291

lfm2-1.2b

INT8-CW

32

75.4

28.3

35.33568905

qwen2.5-coder-1.5b-instruct

INT8-CW

1024

163.8

28.4

35.21126761

minicpm-1b-sft

INT8-CW

31

89.9

28.5

35.0877193

deepseek-r1-distill-qwen-1.5b

INT8-CW

1024

162

28.6

34.96503497

minicpm-1b-sft

INT4-MIXED

31

114.9

28.6

34.96503497

gemma-3-1b-it

INT8-CW

1024

135

29

34.48275862

minicpm-1b-sft

INT4-MIXED

1014

211.8

29.4

34.01360544

lfm2.5-1.2b-instruct

INT8-CW

1024

192.4

29.5

33.89830508

lfm2-1.2b

INT8-CW

1024

187.3

29.7

33.67003367

minicpm-1b-sft

INT8-CW

1014

185.1

30.8

32.46753247

minicpm-1b-sft

INT4-MIXED

1014

180.9

31

32.25806452

lfm2-8b-a1b

INT4-MIXED

32

167.5

31.9

31.34796238

nanollava

FP16

760

191.6

32.2

31.05590062

nanollava

FP16

1752

322.5

32.4

30.86419753

lfm2-8b-a1b

INT4-MIXED

1024

1132.8

33.4

29.94011976

phi-2

INT4-MIXED

32

131.3

34.5

28.98550725

qwen3.5-0.8b

FP16

85

97.9

34.7

28.8184438

smollm3-3b

INT4-MIXED

32

104.9

34.7

28.8184438

smollm3-3b

INT4-MIXED

1024

287.7

35

28.57142857

llama-3.2-3b-instruct

INT4-MIXED

32

90.2

35.3

28.3286119

gemma-4-e2b-it

INT4-MIXED

274

461.1

35.4

28.24858757

qwen3.5-0.8b

FP16

1024

269.5

35.6

28.08988764

qwen2.5-coder-3b-instruct

INT4-MIXED

32

105

35.7

28.01120448

stable-zephyr-3b-dpo

INT4-MIXED

32

120.4

35.9

27.8551532

minicpm5-1b

FP16

32

61.9

36.5

27.39726027

phi-2

INT4-MIXED

1024

334

36.5

27.39726027

gemma-4-e2b-it

INT4-MIXED

1024

627.2

36.7

27.2479564

smollm3-3b

INT4-MIXED

32

118.1

36.7

27.2479564

qwen2.5-coder-3b-instruct

INT4-MIXED

1024

276

36.8

27.17391304

afm-4.5b

INT4-MIXED

32

84

37

27.02702703

llama-3.2-3b-instruct

INT4-MIXED

1024

287

37

27.02702703

tiny-llama-1.1b-chat

FP16

32

61.7

37.1

26.9541779

stable-zephyr-3b-dpo

INT4-MIXED

1024

328.4

37.5

26.66666667

phi-2

INT4-MIXED

32

144.9

37.6

26.59574468

llama-3.2-3b-instruct

INT4-MIXED

32

100.6

37.8

26.45502646

minicpm5-1b

FP16

1024

148.4

37.8

26.45502646

smollm3-3b

INT4-MIXED

1024

373.7

37.9

26.38522427

afm-4.5b

INT4-MIXED

1024

440.3

38

26.31578947

qwen2.5-coder-3b-instruct

INT4-MIXED

32

125.3

38.3

26.10966057

tiny-llama-1.1b-chat

FP16

1024

173.3

38.5

25.97402597

stable-zephyr-3b-dpo

INT4-MIXED

32

145

38.7

25.83979328

llama-3.2-3b-instruct

INT4-MIXED

1024

389.8

39.3

25.44529262

qwen2.5-coder-3b-instruct

INT4-MIXED

1024

465.1

39.6

25.25252525

gemma-4-e2b-it

INT4-MIXED

274

485.8

40.1

24.93765586

lfm2-8b-a1b

INT8-CW

32

180.8

40.4

24.75247525

phi-2

INT4-MIXED

1024

427.4

40.4

24.75247525

gemma-3-1b-it

FP16

32

67

40.8

24.50980392

stable-zephyr-3b-dpo

INT4-MIXED

1024

408.5

41.5

24.09638554

gemma-3-1b-it

FP16

1024

165.1

41.6

24.03846154

lfm2-8b-a1b

INT8-CW

1024

932.4

41.7

23.98081535

gemma-4-e2b-it

INT4-MIXED

1024

697.6

42

23.80952381

gemma-4-e2b-it

INT8-CW

274

448

43.7

22.88329519

phi-2

INT8-CW

32

113.7

43.9

22.77904328

llama-3.2-3b-instruct

INT4-MIXED

32

88.4

44.6

22.42152466

llama-3.2-1b-instruct

FP16

32

53.2

44.9

22.27171492

qwen3-4b

INT4-MIXED

32

139.9

45

22.22222222

phi-4-mini-reasoning

INT4-MIXED

32

141.9

45.2

22.12389381

phi-4-mini-instruct

INT4-MIXED

32

137.9

45.4

22.02643172

llama-3.2-1b-instruct

FP16

1024

197.9

45.6

21.92982456

gemma-4-e2b-it

INT8-CW

1024

657.9

45.8

21.83406114

llama-3.2-3b-instruct

INT4-MIXED

1024

309.3

46.1

21.69197397

phi-4-mini-instruct

INT4-MIXED

1024

380.7

46.4

21.55172414

phi-4-mini-reasoning

INT4-MIXED

1024

385.3

46.8

21.36752137

phi-2

INT8-CW

1024

356.6

47.1

21.23142251

qwen3-4b

INT4-MIXED

1024

410.4

47.4

21.09704641

phi-4-mini-reasoning

INT4-MIXED

32

146.8

48.1

20.79002079

qwen2.5-coder-3b-instruct

INT4-MIXED

32

64.2

48.6

20.57613169

phi-4-mini-reasoning

INT4-MIXED

1024

484

49.5

20.2020202

qwen2.5-coder-3b-instruct

INT4-MIXED

1024

283.4

49.8

20.08032129

phi-3.5-vision-instruct

INT4-MIXED

802

704

50.2

19.92031873

smollm3-3b

INT8-CW

32

97.4

50.2

19.92031873

llama-3.2-3b-instruct

INT8-CW

32

92.1

50.5

19.8019802

qwen2.5-coder-3b-instruct

INT8-CW

32

113.1

50.5

19.8019802

phi-4-mini-instruct

INT4-MIXED

32

145

50.7

19.72386588

phi-4-mini-reasoning

INT4-MIXED

32

139.4

50.9

19.64636542

phi-4-mini-instruct

INT4-MIXED

32

150.2

51.1

19.56947162

phi-3.5-vision-instruct

INT4-MIXED

1032

844.7

51.5

19.41747573

exaone-4.0-1.2b

FP16

32

104.9

51.6

19.37984496

qwen2.5-coder-3b-instruct

INT8-CW

1024

385

51.8

19.30501931

phi-4-mini-instruct

INT4-MIXED

1024

525.6

51.9

19.26782274

smollm3-3b

INT8-CW

1024

308.5

51.9

19.26782274

phi-4-mini-reasoning

INT4-MIXED

1024

405.1

52

19.23076923

llama-3.2-3b-instruct

INT8-CW

1024

331.8

52.4

19.08396947

phi-4-mini-instruct

INT4-MIXED

1024

401.4

52.6

19.01140684

exaone-4.0-1.2b

FP16

1024

273.2

52.8

18.93939394

smollm3-3b

INT4-MIXED

32

64.6

52.9

18.90359168

lfm2.5-1.2b-instruct

FP16

32

93.4

53.4

18.72659176

lfm2-1.2b

FP16

32

90.4

53.5

18.69158879

whisper-large-v3

INT8-CW

32

458.1

53.6

18.65671642

smollm3-3b

INT4-MIXED

1024

303.5

53.7

18.62197393

lfm2.5-1.2b-instruct

FP16

1024

262.1

53.9

18.5528757

lfm2-1.2b

FP16

1024

262.4

54

18.51851852

whisper-large-v3

INT4-MIXED

32

473.2

54

18.51851852

whisper-large-v3

INT8-CW

1024

525.1

54

18.51851852

whisper-large-v3

INT4-MIXED

1024

538.4

54.3

18.41620626

minicpm-1b-sft

FP16

31

145.8

54.4

18.38235294

bloomz-7b1

INT4-MIXED

32

80.3

54.6

18.31501832

minicpm-1b-sft

FP16

1014

256.9

55.6

17.98561151

qwen2.5-coder-1.5b-instruct

FP16

32

89.6

56

17.85714286

qwen3-4b

INT4-MIXED

32

131

56.4

17.73049645

deepseek-r1-distill-qwen-1.5b

FP16

32

94

56.5

17.69911504

qwen2.5-1.5b-instruct

FP16

32

100

56.5

17.69911504

bloomz-7b1

INT4-MIXED

1024

887.3

57.1

17.51313485

qwen2.5-coder-1.5b-instruct

FP16

1024

246.3

57.1

17.51313485

qwen2.5-1.5b-instruct

FP16

1024

239

57.2

17.48251748

codellama-7b

INT4-MIXED

32

124.9

58

17.24137931

deepseek-r1-distill-qwen-1.5b

FP16

1024

236.6

58.1

17.21170396

llama-2-7b-chat-hf

INT4-MIXED

32

117.3

58.1

17.21170396

qwen3-4b

INT4-MIXED

1024

437.6

59.1

16.92047377

lfm2-24b-a2b

INT4-MIXED

32

506.9

59.2

16.89189189

gemma-3-4b-it

INT4-MIXED

32

199.1

59.4

16.83501684

phi-4-mini-reasoning

INT8-CW

32

137.5

60.3

16.58374793

phi-4-mini-instruct

INT8-CW

32

125.1

60.6

16.50165017

codellama-7b

INT4-MIXED

1024

615.3

61

16.39344262

zephyr-7b-beta

INT4-MIXED

32

105.1

61.2

16.33986928

neural-chat-7b-v3-3

INT4-MIXED

32

103.4

61.3

16.3132137

mistral-7b-instruct-v0.3

INT4-MIXED

32

104.2

61.5

16.2601626

codellama-7b

INT4-MIXED

32

112.1

61.6

16.23376623

biomistral-7b-slerp

INT4-MIXED

7

88

61.7

16.20745543

mistral-7b-instruct-v0.2

INT4-MIXED

32

112.5

61.8

16.18122977

llama-2-7b-chat-hf

INT4-MIXED

1024

637.8

62

16.12903226

falcon-7b-instruct

INT4-MIXED

32

121.9

62.1

16.10305958

gemma-3-4b-it

INT4-MIXED

32

187.1

62.5

16

qwen3-vl-4b-instruct

INT4-MIXED

4907

7492.3

62.5

16

qwen3-vl-4b-instruct

INT4-MIXED

4937

7660.2

62.6

15.97444089

phi-4-mini-instruct

INT8-CW

1024

421

62.8

15.92356688

phi-4-mini-reasoning

INT8-CW

1024

426.4

63.3

15.79778831

zephyr-7b-beta

INT4-MIXED

1024

656.4

63.3

15.79778831

neural-chat-7b-v3-3

INT4-MIXED

1024

662.7

63.4

15.77287066

mistral-7b-instruct-v0.3

INT4-MIXED

1024

662.6

63.5

15.7480315

gemma-3-4b-it

INT4-MIXED

1024

513

63.7

15.69858713

falcon-7b-instruct

INT4-MIXED

1024

684.6

63.9

15.64945227

whisper-large-v3

FP16

32

490.7

63.9

15.64945227

mistral-7b-instruct-v0.2

INT4-MIXED

1024

664.5

64

15.625

codellama-7b

INT4-MIXED

1024

880.2

64.1

15.60062402

qwen3-4b

INT8-CW

32

117.9

64.1

15.60062402

whisper-large-v3

FP16

1024

565

64.1

15.60062402

qwen2.5-7b-instruct-1m

INT4-MIXED

32

107.5

64.2

15.57632399

qwen2.5-7b-instruct

INT4-MIXED

32

87.6

64.4

15.52795031

biomistral-7b-slerp

INT4-MIXED

7

98.9

64.7

15.45595054

phi-3.5-vision-instruct

INT8-CW

802

591.1

64.7

15.45595054

neural-chat-7b-v3-3

INT4-MIXED

32

122.1

64.8

15.43209877

mistral-7b-instruct-v0.1

INT4-MIXED

32

116.5

64.9

15.40832049

mistral-7b-instruct-v0.2

INT4-MIXED

32

97.1

64.9

15.40832049

qwen2.5-7b-instruct

INT4-MIXED

1024

607.4

64.9

15.40832049

gemma-3-4b-it

INT4-MIXED

1024

603.4

65.1

15.3609831

mistral-7b-instruct-v0.3

INT4-MIXED

32

119.5

65.1

15.3609831

qwen2.5-7b-instruct-1m

INT4-MIXED

1024

597

65.3

15.31393568

bloomz-7b1

INT4-MIXED

32

109.1

65.4

15.29051988

qwen3-vl-4b-instruct

INT4-MIXED

4907

7999.4

65.5

15.26717557

qwen3-vl-4b-instruct

INT4-MIXED

4937

8165.1

65.6

15.24390244

phi-3.5-vision-instruct

INT8-CW

1032

764

66.2

15.10574018

minicpm3-4b

INT4-MIXED

32

311.9

66.3

15.08295626

afm-4.5b

INT8-CW

32

94.7

66.4

15.06024096

llama-3.1-8b-instruct

INT4-MIXED

32

114.6

66.6

15.01501502

qwen3-4b

INT8-CW

1024

449.2

66.6

15.01501502

llama-3-8b-instruct

INT4-MIXED

32

118.5

66.8

14.97005988

mistral-7b-instruct-v0.3

INT4-MIXED

1024

936.4

66.8

14.97005988

deepseek-r1-distill-llama-8b

INT4-MIXED

32

124.7

66.9

14.94768311

mistral-7b-instruct-v0.2

INT4-MIXED

1024

934.2

66.9

14.94768311

neural-chat-7b-v3-3

INT4-MIXED

1024

929.8

66.9

14.94768311

qwen2.5-vl-7b-instruct

INT4-MIXED

32

171.3

67.1

14.90312966

minicpm-v-2_6

INT4-MIXED

228

718.1

67.3

14.85884101

mistral-7b-instruct-v0.1

INT4-MIXED

1025

936.8

67.4

14.83679525

gpt-oss-20b

INT4-MIXED

32

251.2

67.5

14.81481481

phi-4-multimodal-instruct

INT4-MIXED

578

586.6

67.6

14.79289941

phi-4-multimodal-instruct

INT4-MIXED

786

669.7

68

14.70588235

qwen2.5-7b-instruct

INT4-MIXED

32

104.2

68.1

14.68428781

qwen2-7b-instruct

INT4-MIXED

32

104.1

68.2

14.6627566

qwen2.5-7b-instruct-1m

INT4-MIXED

32

111.6

68.2

14.6627566

minicpm-o-2_6

INT4-MIXED

238

688

68.3

14.64128843

afm-4.5b

INT8-CW

1024

419.9

68.5

14.59854015

minicpm4-8b

INT4-MIXED

32

120.9

68.5

14.59854015

deepseek-r1-distill-qwen-7b

INT4-MIXED

32

124.3

68.6

14.57725948

gpt-oss-20b

INT4-MIXED

1024

3065.9

68.6

14.57725948

bloomz-7b1

INT4-MIXED

1024

845.1

68.8

14.53488372

llama-3.1-8b-instruct

INT4-MIXED

1024

666.3

68.8

14.53488372

phi-4-multimodal-instruct

INT4-MIXED

1362

1405.8

69

14.49275362

deepseek-r1-distill-llama-8b

INT4-MIXED

1024

664

69.1

14.47178003

qwen3-8b

INT4-MIXED

32

151.4

69.1

14.47178003

qwen2.5-vl-7b-instruct

INT4-MIXED

1024

711.9

69.3

14.43001443

llama-3-8b-instruct

INT4-MIXED

1024

676.1

69.4

14.4092219

qwen2-7b-instruct

INT4-MIXED

1024

859

69.6

14.36781609

qwen2.5-7b-instruct

INT4-MIXED

1024

858.5

69.6

14.36781609

qwen2.5-7b-instruct-1m

INT4-MIXED

1024

851.8

69.6

14.36781609

phi-4-multimodal-instruct

INT4-MIXED

1570

1630.8

69.8

14.32664756

deepseek-r1-distill-qwen-7b

INT4-MIXED

1024

847.3

69.9

14.30615165

gemma-3-4b-it

INT4-MIXED

32

171.9

70.4

14.20454545

minicpm4-8b

INT4-MIXED

1024

737.8

70.6

14.16430595

llama-3-8b-instruct

INT4-MIXED

32

123.1

70.8

14.12429379

qwen3-8b

INT4-MIXED

1024

700.4

71.2

14.04494382

falcon-7b-instruct

INT4-MIXED

32

119

71.7

13.94700139

fara-7b

INT4-MIXED

32

266.1

71.7

13.94700139

qwen2.5-vl-7b-instruct

INT4-MIXED

32

179.5

71.9

13.90820584

minicpm-v-2_6

INT4-MIXED

228

806.3

72.1

13.86962552

llama-3-8b-instruct

INT4-MIXED

1024

925.4

72.5

13.79310345

minicpm-o-2_6

INT4-MIXED

238

785.6

72.6

13.77410468

minicpm3-4b

INT4-MIXED

32

354.5

72.6

13.77410468

minicpm4-8b

INT4-MIXED

32

116.2

72.9

13.71742112

qwen3-8b

INT4-MIXED

32

143.9

73

13.69863014

minicpm3-4b

INT4-MIXED

1024

819.7

73.2

13.66120219

minicpm3-4b

INT4-MIXED

32

315.8

73.3

13.6425648

falcon-7b-instruct

INT4-MIXED

1024

924.6

73.4

13.6239782

qwen2.5-vl-7b-instruct

INT4-MIXED

1024

963.4

73.6

13.58695652

fara-7b

INT4-MIXED

1024

1167.1

73.9

13.53179973

mistral-7b-instruct-v0.3

INT4-MIXED

32

112.9

73.9

13.53179973

minicpm-v-4_5

INT4-MIXED

217

758.4

74.2

13.47708895

gemma-3-4b-it

INT8-CW

32

191

74.3

13.4589502

minicpm4-8b

INT4-MIXED

1024

1028.2

74.5

13.42281879

qwen3-8b

INT4-MIXED

1024

960.8

75.1

13.31557923

qwen3-vl-4b-instruct

INT4-MIXED

4907

7854.1

75.5

13.24503311

qwen3-vl-4b-instruct

INT4-MIXED

4937

7886.6

75.5

13.24503311

gemma-3-4b-it

INT4-MIXED

1024

511

75.6

13.22751323

mistral-7b-instruct-v0.3

INT4-MIXED

1024

674.2

76.2

13.12335958

minicpm-v-4_5

INT4-MIXED

217

833.1

78.1

12.80409731

qwen2-7b-instruct

INT4-MIXED

32

101

78.2

12.78772379

qwen3.5-9b

INT4-MIXED

83

351

78.2

12.78772379

gemma-7b-it

INT4-MIXED

32

127.3

78.3

12.77139208

qwen3.5-9b

INT4-MIXED

1024

1122.8

78.5

12.7388535

minicpm3-4b

INT8-CW

32

289.4

79

12.65822785

videochat-flash-qwen2_5-7b_internvideo2-1b

INT4-MIXED

14531

61450.5

79.1

12.64222503

minicpm3-4b

INT4-MIXED

1024

939.9

79.4

12.59445844

gemma-3-4b-it

INT8-CW

1024

522.6

79.9

12.51564456

qwen2-7b-instruct

INT4-MIXED

1024

617.2

79.9

12.51564456

zephyr-7b-beta

INT4-MIXED

32

125.8

80

12.5

qwen2.5-7b-instruct

INT4-MIXED

32

96.9

80.2

12.46882793

minicpm3-4b

INT4-MIXED

1024

859.2

80.4

12.43781095

gemma-7b-it

INT4-MIXED

1024

799.3

81.2

12.31527094

qwen3.5-9b

INT4-MIXED

83

371.9

81.6

12.25490196

llama-2-7b-chat-hf

INT4-MIXED

32

103.2

81.8

12.22493888

qwen2.5-7b-instruct

INT4-MIXED

1024

622.3

81.9

12.21001221

zephyr-7b-beta

INT4-MIXED

1024

692.9

81.9

12.21001221

bloomz-7b1

INT8-CW

32

108

82

12.19512195

phi-4-multimodal-instruct

INT8-CW

578

638

82.1

12.18026797

phi-4-multimodal-instruct

INT8-CW

786

729.5

82.1

12.18026797

qwen3.5-9b

INT4-MIXED

1024

1406.1

82.4

12.13592233

deepseek-r1-distill-qwen-7b

INT4-MIXED

32

95.7

82.6

12.10653753

phi-4-multimodal-instruct

INT8-CW

1362

1556.3

83.2

12.01923077

gemma-7b-it

INT4-MIXED

32

124.6

83.4

11.99040767

llama-3.1-8b-instruct

INT4-MIXED

32

125.6

83.4

11.99040767

deepseek-r1-distill-llama-8b

INT4-MIXED

32

135.8

83.5

11.9760479

phi-4-multimodal-instruct

INT8-CW

1570

1713.3

83.6

11.96172249

deepseek-r1-distill-qwen-7b

INT4-MIXED

1024

618.6

83.8

11.93317422

gpt-oss-20b

INT4-MIXED

32

296.4

83.8

11.93317422

mistral-7b-instruct-v0.2

INT4-MIXED

32

119.9

84

11.9047619

qwen2.5-vl-7b-instruct

INT4-MIXED

32

172

84.1

11.89060642

llama-2-7b-chat-hf

INT4-MIXED

1024

633.2

84.4

11.84834123

llama-3.1-8b-instruct

INT4-MIXED

1024

896.5

84.9

11.77856302

deepseek-r1-distill-llama-8b

INT4-MIXED

1024

900.7

85

11.76470588

phi-4-multimodal-instruct

INT4-MIXED

578

736.2

85

11.76470588

phi-4-multimodal-instruct

INT4-MIXED

1362

1812.8

85.1

11.75088132

bloomz-7b1

INT8-CW

1024

680.1

85.4

11.70960187

minicpm-v-2_6

INT4-MIXED

228

713.1

85.4

11.70960187

gemma-7b-it

INT4-MIXED

1024

1103.7

85.6

11.68224299

qwen2.5-vl-7b-instruct

INT4-MIXED

1024

733.8

85.8

11.65501166

phi-4-multimodal-instruct

INT4-MIXED

1570

2016.5

85.9

11.64144354

phi-4-multimodal-instruct

INT4-MIXED

786

903.1

85.9

11.64144354

mistral-7b-instruct-v0.2

INT4-MIXED

1024

659.4

86

11.62790698

minicpm3-4b

INT8-CW

1024

885

86.5

11.56069364

gpt-oss-20b

INT4-MIXED

1024

3349.6

86.6

11.54734411

gemma-2-9b-it

INT4-MIXED

32

167.3

87.7

11.40250855

deepseek-r1-distill-llama-8b

INT4-MIXED

32

109.3

87.8

11.38952164

phi-2

FP16

32

133.5

88

11.36363636

llama-3.1-8b-instruct

INT4-MIXED

32

104.6

89.3

11.19820829

stable-diffusion-xl-1.0-inpainting-0.1

INT8-CW

32

91.2

89.3

11.19820829

deepseek-r1-distill-llama-8b

INT4-MIXED

1024

679.4

89.4

11.18568233

llama-3-8b-instruct

INT4-MIXED

32

100.5

89.7

11.14827202

qwen3-8b

INT4-MIXED

32

123.3

90.1

11.09877913

llama-3-8b-instruct

INT4-MIXED

32

102.2

90.2

11.0864745

gemma-2-9b-it

INT4-MIXED

1024

906.4

90.5

11.04972376

llama-3.1-8b-instruct

INT4-MIXED

1024

670.5

91.6

10.91703057

stable-diffusion-xl-1.0-inpainting-0.1

INT8-CW

32

95.3

91.6

10.91703057

qwen3-8b

INT4-MIXED

1024

714

91.8

10.89324619

llama-3-8b-instruct

INT4-MIXED

1024

672

91.9

10.88139282

llama-3-8b-instruct

INT4-MIXED

1024

663.7

92.1

10.8577633

minicpm4-8b

INT4-MIXED

32

109.6

92.5

10.81081081

gemma-2-9b-it

INT4-MIXED

32

167

92.6

10.79913607

phi-2

FP16

1024

541.4

92.8

10.77586207

minicpm4-8b

INT4-MIXED

1024

762.6

94.3

10.60445387

ltx-video

INT8-CW

11

98.9

95.8

10.43841336

gemma-2-9b-it

INT4-MIXED

1024

1219.9

95.9

10.42752868

codellama-7b

INT8-CW

32

123.1

96.1

10.40582726

llama-2-7b-chat-hf

INT8-CW

32

130.2

98.2

10.18329939

ltx-video

INT4-MIXED

11

93.2

98.6

10.14198783

dolly-v2-12b

INT4-MIXED

32

173.1

98.9

10.11122346

Topology

Precision

Input Size

1st latency (ms)

2nd latency (ms)

2nd token per sec

lfm2.5-350m

INT8-CW

32

20.2

5

200

lfm2.5-350m

INT8-CW

1024

40

5.1

196.0784314

lfm2.5-350m

INT4-MIXED

1024

40.8

5.2

192.3076923

lfm2.5-350m

INT4-MIXED

32

19

5.2

192.3076923

distil-large-v2

INT4-MIXED

32

165.8

5.9

169.4915254

distil-large-v2

INT4-MIXED

1024

213.2

6

166.6666667

distil-large-v2

INT8-CW

32

164.1

6

166.6666667

distil-large-v2

INT8-CW

1024

209.4

6

166.6666667

whisper-large-v3-turbo

INT8-CW

1024

225.7

6.3

158.7301587

whisper-large-v3-turbo

INT8-CW

32

176.3

6.4

156.25

whisper-large-v3-turbo

INT4-MIXED

32

183.8

7

142.8571429

whisper-large-v3-turbo

INT4-MIXED

1024

235.6

7

142.8571429

minicpm4-0.5b

INT4-MIXED

32

28.3

7.4

135.1351351

minicpm4-0.5b

INT4-MIXED

1024

43.9

7.4

135.1351351

minicpm4-0.5b

INT4-MIXED

32

27.9

7.4

135.1351351

gemma-3-270m

INT4-MIXED

32

29

7.5

133.3333333

minicpm4-0.5b

INT4-MIXED

1024

41

7.5

133.3333333

tiny-llama-1.1b-chat

INT4-MIXED

32

25

7.5

133.3333333

gemma-3-270m

INT8-CW

1024

35.7

7.6

131.5789474

minicpm4-0.5b

INT8-CW

32

31.4

7.7

129.8701299

qwen2.5-coder-0.5b-instruct

INT4-MIXED

32

28.6

7.7

129.8701299

qwen2.5-coder-0.5b-instruct

INT4-MIXED

1024

45.4

7.7

129.8701299

qwen2.5-coder-0.5b-instruct

INT4-MIXED

32

28.6

7.7

129.8701299

minicpm4-0.5b

INT4-MIXED

32

30.2

7.8

128.2051282

qwen2.5-coder-0.5b-instruct

INT4-MIXED

1024

42.8

7.8

128.2051282

minicpm4-0.5b

INT4-MIXED

1024

50.1

7.9

126.5822785

lfm2.5-350m

FP16

32

16.7

8

125

minicpm4-0.5b

INT8-CW

1024

49.7

8

125

qwen2.5-coder-0.5b-instruct

INT4-MIXED

32

30

8

125

lfm2.5-1.2b-instruct

INT4-MIXED

32

20

8.1

123.4567901

lfm2.5-350m

FP16

1024

41

8.1

123.4567901

qwen2.5-coder-0.5b-instruct

INT4-MIXED

1024

52.5

8.1

123.4567901

gemma-3-270m

INT8-CW

32

30.8

8.2

121.9512195

nuextract-tiny

INT4-MIXED

32

30.5

8.2

121.9512195

tiny-llama-1.1b-chat

INT4-MIXED

32

26.4

8.2

121.9512195

tiny-llama-1.1b-chat

INT4-MIXED

1024

81.6

8.2

121.9512195

gemma-3-270m

FP16

1024

29.1

8.3

120.4819277

gemma-3-270m

FP16

32

25.1

8.3

120.4819277

lfm2-1.2b

INT4-MIXED

32

20.1

8.3

120.4819277

lfm2.5-1.2b-instruct

INT4-MIXED

1024

81.6

8.3

120.4819277

lfm2.5-1.2b-instruct

INT4-MIXED

32

20.4

8.3

120.4819277

nuextract-tiny

INT4-MIXED

1024

50.9

8.3

120.4819277

qwen2.5-coder-0.5b-instruct

INT8-CW

32

30.7

8.3

120.4819277

whisper-large-v3-turbo

FP16

32

177.3

8.3

120.4819277

whisper-large-v3-turbo

FP16

1024

224.5

8.3

120.4819277

lfm2-1.2b

INT4-MIXED

32

20.3

8.4

119.047619

qwen2.5-coder-0.5b-instruct

INT8-CW

1024

49.7

8.4

119.047619

lfm2-1.2b

INT4-MIXED

32

21.2

8.5

117.6470588

lfm2-1.2b

INT4-MIXED

1024

80.9

8.5

117.6470588

lfm2.5-1.2b-instruct

INT4-MIXED

32

21.4

8.5

117.6470588

lfm2.5-1.2b-instruct

INT4-MIXED

1024

86

8.5

117.6470588

distil-large-v2

FP16

32

167.8

8.6

116.2790698

lfm2-1.2b

INT4-MIXED

1024

85.8

8.6

116.2790698

distil-large-v2

FP16

1024

214.4

8.7

114.9425287

lfm2-1.2b

INT4-MIXED

1024

96.9

8.7

114.9425287

lfm2.5-1.2b-instruct

INT4-MIXED

1024

95.8

8.7

114.9425287

nuextract-tiny

INT8-CW

32

31.1

8.9

112.3595506

tiny-llama-1.1b-chat

INT4-MIXED

1024

106

8.9

112.3595506

gemma-3-270m

INT4-MIXED

1024

32.9

9.1

109.8901099

nuextract-tiny

INT8-CW

1024

47.8

9.1

109.8901099

minicpm5-1b

INT4-MIXED

32

29.7

9.5

105.2631579

minicpm5-1b

INT4-MIXED

1024

78.1

9.6

104.1666667

llama-3.2-1b-instruct

INT4-MIXED

32

21.7

9.7

103.0927835

llama-3.2-1b-instruct

INT4-MIXED

32

21.3

9.7

103.0927835

llama-3.2-1b-instruct

INT4-MIXED

1024

82.2

10.2

98.03921569

llama-3.2-1b-instruct

INT4-MIXED

1024

91.9

10.3

97.08737864

whisper-small

INT4-MIXED

1024

139.3

10.5

95.23809524

whisper-small

INT4-MIXED

32

94.6

10.6

94.33962264

whisper-small

INT4-MIXED

32

103.7

10.8

92.59259259

whisper-small

INT4-MIXED

1024

145.3

10.8

92.59259259

minicpm4-0.5b

FP16

32

24.9

10.9

91.74311927

nuextract-tiny

FP16

32

26.9

11.1

90.09009009

whisper-small

INT8-CW

32

101.5

11.1

90.09009009

lfm2-8b-a1b

INT4-MIXED

32

119

11.2

89.28571429

minicpm4-0.5b

FP16

1024

55.8

11.3

88.49557522

minicpm5-1b

INT8-CW

32

31

11.4

87.71929825

whisper-small

INT8-CW

1024

144.8

11.4

87.71929825

gemma-3-1b-it

INT4-MIXED

1024

83.1

11.6

86.20689655

gemma-3-1b-it

INT4-MIXED

1024

85.7

11.7

85.47008547

gemma-3-1b-it

INT4-MIXED

32

38.1

11.8

84.74576271

qwen2.5-coder-1.5b-instruct

INT4-MIXED

32

33.9

11.8

84.74576271

gemma-3-1b-it

INT4-MIXED

32

37.6

11.9

84.03361345

minicpm5-1b

INT8-CW

1024

77.5

11.9

84.03361345

qwen2.5-1.5b-instruct

INT4-MIXED

32

34.2

11.9

84.03361345

tiny-llama-1.1b-chat

INT8-CW

32

28.1

11.9

84.03361345

exaone-4.0-1.2b

INT4-MIXED

32

42.1

12

83.33333333

exaone-4.0-1.2b

INT4-MIXED

32

41.8

12

83.33333333

gemma-3-1b-it

INT4-MIXED

32

39.5

12

83.33333333

qwen2.5-coder-1.5b-instruct

INT4-MIXED

32

35.4

12

83.33333333

qwen2.5-coder-1.5b-instruct

INT4-MIXED

32

34.2

12

83.33333333

whisper-small

FP16

32

93

12

83.33333333

nuextract-tiny

FP16

1024

54.7

12.1

82.6446281

exaone-4.0-1.2b

INT4-MIXED

1024

131.2

12.2

81.96721311

exaone-4.0-1.2b

INT4-MIXED

1024

133.4

12.2

81.96721311

lfm2-8b-a1b

INT4-MIXED

1024

253.6

12.2

81.96721311

nanollava

INT4-MIXED

760

97

12.2

81.96721311

qwen2.5-coder-0.5b-instruct

FP16

32

27.2

12.2

81.96721311

qwen2.5-coder-1.5b-instruct

INT4-MIXED

1024

108.7

12.2

81.96721311

whisper-small

FP16

1024

138.4

12.2

81.96721311

deepseek-r1-distill-qwen-1.5b

INT4-MIXED

32

39.2

12.3

81.30081301

qwen2.5-1.5b-instruct

INT4-MIXED

1024

132.3

12.4

80.64516129

qwen2.5-1.5b-instruct

INT4-MIXED

32

34.3

12.4

80.64516129

qwen2.5-coder-0.5b-instruct

FP16

1024

51

12.4

80.64516129

exaone-4.0-1.2b

INT4-MIXED

32

44.6

12.5

80

gemma-3-1b-it

INT4-MIXED

1024

91.5

12.5

80

qwen2.5-coder-1.5b-instruct

INT4-MIXED

1024

139

12.5

80

tiny-llama-1.1b-chat

INT8-CW

1024

93.8

12.5

80

exaone-4.0-1.2b

INT4-MIXED

1024

142.1

12.6

79.36507937

qwen2.5-1.5b-instruct

INT4-MIXED

1024

133.1

12.6

79.36507937

qwen2.5-coder-1.5b-instruct

INT4-MIXED

1024

131.2

12.6

79.36507937

whisper-small

INT4-MIXED

32

101.9

12.8

78.125

deepseek-r1-distill-qwen-1.5b

INT4-MIXED

1024

132.2

13

76.92307692

qwen2.5-1.5b-instruct

INT4-MIXED

32

36.9

13

76.92307692

qwen2.5-1.5b-instruct

INT4-MIXED

32

36.9

13

76.92307692

nanollava

FP16

760

105.6

13.1

76.33587786

whisper-small

INT4-MIXED

1024

147.9

13.1

76.33587786

nanollava

INT4-MIXED

1752

191.4

13.3

75.18796992

nanollava

INT8-CW

760

100.1

13.5

74.07407407

qwen2.5-1.5b-instruct

INT4-MIXED

1024

141.7

13.5

74.07407407

lfm2.5-1.2b-instruct

INT8-CW

32

22.6

13.6

73.52941176

nanollava

INT8-CW

1752

183.8

13.6

73.52941176

gemma-3-1b-it

INT8-CW

32

40.6

13.7

72.99270073

lfm2.5-1.2b-instruct

INT8-CW

1024

106.2

13.8

72.46376812

lfm2-1.2b

INT8-CW

32

22.8

13.9

71.94244604

qwen2.5-1.5b-instruct

INT4-MIXED

1024

140.3

13.9

71.94244604

lfm2-1.2b

INT8-CW

1024

107

14.1

70.92198582

gemma-3-1b-it

INT8-CW

1024

92.7

14.3

69.93006993

llama-3.2-1b-instruct

INT8-CW

32

24.5

14.5

68.96551724

nanollava

FP16

1752

200.2

14.6

68.49315068

deepseek-r1-distill-qwen-1.5b

INT4-MIXED

32

37.4

15.5

64.51612903

llama-3.2-1b-instruct

INT8-CW

1024

97.3

15.5

64.51612903

deepseek-r1-distill-qwen-1.5b

INT4-MIXED

1024

146.5

15.9

62.89308176

minicpm-1b-sft

INT4-MIXED

31

59.3

16.3

61.34969325

minicpm-1b-sft

INT4-MIXED

31

58.3

16.4

60.97560976

minicpm-1b-sft

INT4-MIXED

1014

122.3

16.6

60.24096386

minicpm-1b-sft

INT4-MIXED

1014

127.1

16.7

59.88023952

exaone-4.0-1.2b

INT8-CW

32

46.8

17

58.82352941

qwen3.5-0.8b

INT4-MIXED

1024

236.9

17.2

58.13953488

lfm2-8b-a1b

INT8-CW

32

97.4

17.3

57.80346821

qwen3.5-0.8b

INT4-MIXED

1024

238.9

17.4

57.47126437

qwen3.5-0.8b

INT4-MIXED

85

106.7

17.5

57.14285714

lfm2-8b-a1b

INT8-CW

1024

315

18

55.55555556

qwen3.5-0.8b

INT8-CW

1024

248

18

55.55555556

exaone-4.0-1.2b

INT8-CW

1024

141.5

18.1

55.24861878

qwen3.5-0.8b

INT4-MIXED

85

118.9

18.1

55.24861878

qwen3.5-0.8b

INT8-CW

85

116.7

18.2

54.94505495

qwen2.5-1.5b-instruct

INT8-CW

32

38.1

18.3

54.64480874

qwen2.5-coder-1.5b-instruct

INT8-CW

32

37

18.3

54.64480874

deepseek-r1-distill-qwen-1.5b

INT8-CW

32

37.8

18.4

54.34782609

qwen2.5-1.5b-instruct

INT8-CW

32

38.6

18.4

54.34782609

lfm2-24b-a2b

INT4-MIXED

1024

727.6

18.7

53.47593583

lfm2-24b-a2b

INT4-MIXED

32

228.4

18.7

53.47593583

phi-2

INT4-MIXED

32

59.5

18.7

53.47593583

qwen3.5-0.8b

INT4-MIXED

1024

254.2

18.7

53.47593583

minicpm-1b-sft

INT4-MIXED

31

64.5

18.8

53.19148936

stable-zephyr-3b-dpo

INT4-MIXED

32

59.6

18.8

53.19148936

minicpm-1b-sft

INT8-CW

31

66.7

18.9

52.91005291

qwen3.5-0.8b

INT4-MIXED

85

106.5

18.9

52.91005291

minicpm-1b-sft

INT4-MIXED

1014

148.2

19

52.63157895

deepseek-r1-distill-qwen-1.5b

INT8-CW

1024

139.2

19.1

52.35602094

qwen2.5-coder-1.5b-instruct

INT8-CW

1024

138.7

19.1

52.35602094

qwen2.5-1.5b-instruct

INT8-CW

1024

138.8

19.2

52.08333333

qwen2.5-1.5b-instruct

INT8-CW

1024

138.6

19.2

52.08333333

smollm3-3b

INT4-MIXED

32

60.1

20.2

49.5049505

phi-2

INT4-MIXED

32

61.7

20.3

49.26108374

minicpm-1b-sft

INT8-CW

1014

139.6

20.4

49.01960784

minicpm5-1b

FP16

32

28.5

20.7

48.30917874

phi-2

INT4-MIXED

1024

286.8

21

47.61904762

smollm3-3b

INT4-MIXED

32

51.3

21

47.61904762

stable-zephyr-3b-dpo

INT4-MIXED

1024

281

21

47.61904762

stable-zephyr-3b-dpo

INT4-MIXED

32

66

21.1

47.39336493

qwen2.5-coder-3b-instruct

INT4-MIXED

32

62.1

21.5

46.51162791

qwen2.5-coder-3b-instruct

INT4-MIXED

32

54.6

21.5

46.51162791

smollm3-3b

INT4-MIXED

32

53.5

21.5

46.51162791

smollm3-3b

INT4-MIXED

1024

222.5

21.5

46.51162791

minicpm5-1b

FP16

1024

86.1

21.6

46.2962963

qwen2.5-1.5b-instruct

INT4-MIXED

1024

179.7

21.6

46.2962963

llama-3.2-3b-instruct

INT4-MIXED

32

42.1

21.7

46.08294931

qwen2.5-1.5b-instruct

INT4-MIXED

32

75.6

22.1

45.24886878

qwen2.5-1.5b-instruct

INT4-MIXED

1024

188.2

22.2

45.04504505

qwen2.5-1.5b-instruct

INT4-MIXED

32

73.2

22.3

44.84304933

smollm3-3b

INT4-MIXED

1024

233.9

22.4

44.64285714

smollm3-3b

INT4-MIXED

1024

265.3

22.6

44.24778761

qwen2.5-coder-3b-instruct

INT4-MIXED

32

57.3

22.7

44.05286344

qwen2.5-coder-3b-instruct

INT4-MIXED

1024

223.3

22.7

44.05286344

qwen2.5-coder-3b-instruct

INT4-MIXED

1024

228.9

22.7

44.05286344

llama-3.2-3b-instruct

INT4-MIXED

32

43.4

22.8

43.85964912

phi-2

INT4-MIXED

1024

361.2

22.8

43.85964912

llama-3.2-3b-instruct

INT4-MIXED

32

44.7

22.9

43.66812227

llama-3.2-3b-instruct

INT4-MIXED

1024

228.1

23.3

42.91845494

stable-zephyr-3b-dpo

INT4-MIXED

1024

314.1

23.4

42.73504274

gemma-3-1b-it

FP16

32

37.6

23.7

42.19409283

qwen3.5-0.8b

FP16

85

91.6

23.9

41.84100418

qwen3.5-0.8b

FP16

1024

249.6

24

41.66666667

qwen2.5-coder-3b-instruct

INT4-MIXED

1024

419

24.1

41.49377593

gemma-3-1b-it

FP16

1024

103.7

24.2

41.32231405

llama-3.2-3b-instruct

INT4-MIXED

1024

243.1

24.3

41.15226337

llama-3.2-3b-instruct

INT4-MIXED

1024

265.5

24.5

40.81632653

qwen2.5-1.5b-instruct

INT8-CW

32

79.1

25

40

qwen2.5-1.5b-instruct

INT8-CW

1024

178.8

25.7

38.91050584

tiny-llama-1.1b-chat

FP16

32

29.3

25.9

38.61003861

gemma-4-e2b-it

INT4-MIXED

274

284.3

26.6

37.59398496

qwen3-4b

INT4-MIXED

32

58.6

26.7

37.45318352

tiny-llama-1.1b-chat

FP16

1024

106.7

26.7

37.45318352

phi-4-mini-reasoning

INT4-MIXED

32

62.5

27.6

36.23188406

phi-4-mini-instruct

INT4-MIXED

32

63.4

27.7

36.10108303

qwen3-4b

INT4-MIXED

32

60.9

27.7

36.10108303

gemma-4-e2b-it

INT4-MIXED

274

313.5

28

35.71428571

gpt-oss-20b

INT4-MIXED

32

221.7

28.1

35.58718861

lfm2.5-1.2b-instruct

FP16

32

33

28.2

35.46099291

lfm2-1.2b

FP16

1024

140.1

28.3

35.33568905

lfm2-1.2b

FP16

32

33.5

28.3

35.33568905

lfm2.5-1.2b-instruct

FP16

1024

139.3

28.3

35.33568905

afm-4.5b

INT4-MIXED

32

51.9

28.4

35.21126761

phi-4-mini-instruct

INT4-MIXED

32

65.5

28.4

35.21126761

phi-4-mini-reasoning

INT4-MIXED

32

63.5

28.8

34.72222222

gemma-3-4b-it

INT4-MIXED

32

74.7

28.9

34.60207612

phi-4-mini-reasoning

INT4-MIXED

32

65

28.9

34.60207612

qwen3-4b

INT4-MIXED

1024

324.6

29.2

34.24657534

phi-4-mini-reasoning

INT4-MIXED

1024

376.8

29.3

34.12969283

phi-4-mini-instruct

INT4-MIXED

1024

355.7

29.5

33.89830508

phi-3.5-vision-instruct

INT4-MIXED

802

491.2

29.6

33.78378378

gemma-3-4b-it

INT4-MIXED

32

75.3

29.8

33.55704698

gpt-oss-20b

INT4-MIXED

1024

648.2

29.8

33.55704698

llama-3.2-1b-instruct

FP16

32

35

29.8

33.55704698

gemma-3-4b-it

INT4-MIXED

32

76.6

30.1

33.22259136

phi-4-mini-reasoning

INT4-MIXED

1024

368

30.2

33.11258278

afm-4.5b

INT4-MIXED

1024

378.4

30.3

33.00330033

phi-4-mini-instruct

INT4-MIXED

1024

378.1

30.3

33.00330033

whisper-large-v3

INT4-MIXED

32

288.2

30.3

33.00330033

qwen3-4b

INT4-MIXED

1024

345.6

30.4

32.89473684

phi-4-mini-instruct

INT4-MIXED

32

68.1

30.5

32.78688525

gemma-3-4b-it

INT4-MIXED

1024

355.4

30.6

32.67973856

phi-3.5-vision-instruct

INT4-MIXED

1032

601

30.6

32.67973856

llama-3.2-1b-instruct

FP16

1024

131.7

30.7

32.5732899

phi-4-mini-reasoning

INT4-MIXED

1024

389.4

30.8

32.46753247

phi-2

INT8-CW

32

67.3

31.5

31.74603175

gemma-3-4b-it

INT4-MIXED

1024

397.4

31.7

31.54574132

gemma-3-4b-it

INT4-MIXED

1024

374.7

31.8

31.44654088

phi-4-mini-instruct

INT4-MIXED

1024

462

32.2

31.05590062

gemma-4-e2b-it

INT4-MIXED

1024

463.3

32.7

30.58103976

minicpm-1b-sft

FP16

31

61.7

33

30.3030303

whisper-large-v3

INT4-MIXED

1024

361.1

33.3

30.03003003

whisper-large-v3

INT8-CW

32

314.9

33.3

30.03003003

deepseek-r1-distill-qwen-1.5b

FP16

32

41.1

33.5

29.85074627

qwen2.5-1.5b-instruct

FP16

32

40.8

33.6

29.76190476

qwen2.5-coder-1.5b-instruct

FP16

32

40.4

33.6

29.76190476

whisper-large-v3

INT8-CW

1024

364.7

33.8

29.58579882

gemma-4-e2b-it

INT4-MIXED

1024

496

33.9

29.49852507

phi-2

INT8-CW

1024

325

34

29.41176471

smollm3-3b

INT8-CW

32

57.1

34

29.41176471

qwen2.5-coder-3b-instruct

INT8-CW

32

58.2

34.1

29.3255132

deepseek-r1-distill-qwen-1.5b

FP16

1024

166.6

34.4

29.06976744

exaone-4.0-1.2b

FP16

32

41.9

34.4

29.06976744

minicpm-1b-sft

FP16

1014

169.4

34.5

28.98550725

qwen2.5-1.5b-instruct

FP16

1024

166.8

34.6

28.9017341

qwen2.5-coder-1.5b-instruct

FP16

1024

162.4

34.7

28.8184438

gpt-oss-20b

INT4-MIXED

32

215.6

35.2

28.40909091

phi-4-multimodal-instruct

INT4-MIXED

786

540.4

35.2

28.40909091

smollm3-3b

INT8-CW

1024

277.4

35.2

28.40909091

phi-4-multimodal-instruct

INT4-MIXED

578

456.9

35.3

28.3286119

exaone-4.0-1.2b

FP16

1024

330.3

35.4

28.24858757

qwen2.5-coder-3b-instruct

INT8-CW

1024

268.4

35.5

28.16901408

gemma-4-e2b-it

INT8-CW

274

317.9

35.8

27.93296089

phi-4-multimodal-instruct

INT4-MIXED

1362

1124.1

36.3

27.54820937

whisper-large-v3

FP16

32

269.7

36.3

27.54820937

codellama-7b

INT4-MIXED

32

52.6

36.6

27.32240437

llama-2-7b-chat-hf

INT4-MIXED

32

53.1

36.6

27.32240437

phi-4-multimodal-instruct

INT4-MIXED

1570

1240.6

36.6

27.32240437

gpt-oss-20b

INT4-MIXED

1024

682.1

36.7

27.2479564

whisper-large-v3

FP16

1024

330

37.1

26.9541779

llama-3.2-3b-instruct

INT8-CW

32

47

37.8

26.45502646

llama-2-7b-chat-hf

INT4-MIXED

32

56.2

38.7

25.83979328

biomistral-7b-slerp

INT4-MIXED

7

44.9

38.8

25.77319588

mistral-7b-instruct-v0.2

INT4-MIXED

32

54.2

38.9

25.70694087

zephyr-7b-beta

INT4-MIXED

32

53.8

38.9

25.70694087

mistral-7b-instruct-v0.3

INT4-MIXED

32

55

39

25.64102564

codellama-7b

INT4-MIXED

32

55.7

39.1

25.57544757

llama-3.2-3b-instruct

INT8-CW

1024

272

39.1

25.57544757

neural-chat-7b-v3-3

INT4-MIXED

32

53.6

39.1

25.57544757

llama-2-7b-chat-hf

INT4-MIXED

1024

493.8

39.4

25.38071066

codellama-7b

INT4-MIXED

1024

486.2

39.6

25.25252525

minicpm3-4b

INT4-MIXED

32

171.8

40.2

24.87562189

falcon-7b-instruct

INT4-MIXED

32

56.9

40.5

24.69135802

minicpm3-4b

INT4-MIXED

32

176

40.5

24.69135802

phi-4-multimodal-instruct

INT4-MIXED

578

526.6

40.6

24.63054187

minicpm3-4b

INT4-MIXED

32

176.6

40.7

24.57002457

qwen3-vl-4b-instruct

INT4-MIXED

4937

6463.6

40.7

24.57002457

phi-4-multimodal-instruct

INT4-MIXED

786

637.4

40.9

24.44987775

qwen3-vl-4b-instruct

INT4-MIXED

4907

6197.7

40.9

24.44987775

mistral-7b-instruct-v0.3

INT4-MIXED

32

60.8

41.1

24.33090024

mistral-7b-instruct-v0.2

INT4-MIXED

32

59.4

41.2

24.27184466

zephyr-7b-beta

INT4-MIXED

1024

506.8

41.2

24.27184466

mistral-7b-instruct-v0.2

INT4-MIXED

1024

510.8

41.3

24.21307506

biomistral-7b-slerp

INT4-MIXED

7

48.3

41.4

24.15458937

llama-2-7b-chat-hf

INT4-MIXED

1024

533.9

41.4

24.15458937

mistral-7b-instruct-v0.3

INT4-MIXED

1024

512

41.4

24.15458937

neural-chat-7b-v3-3

INT4-MIXED

32

58.4

41.4

24.15458937

mistral-7b-instruct-v0.1

INT4-MIXED

32

60.9

41.5

24.09638554

minicpm3-4b

INT4-MIXED

1024

727.4

41.6

24.03846154

qwen3-vl-4b-instruct

INT4-MIXED

4907

6260.5

41.6

24.03846154

mistral-7b-instruct-v0.3

INT4-MIXED

32

59.1

41.7

23.98081535

qwen3-vl-4b-instruct

INT4-MIXED

4937

6553

41.7

23.98081535

mistral-7b-instruct-v0.2

INT4-MIXED

32

59.6

41.8

23.92344498

neural-chat-7b-v3-3

INT4-MIXED

1024

507.2

41.8

23.92344498

qwen3-vl-4b-instruct

INT4-MIXED

4907

6485.5

41.8

23.92344498

codellama-7b

INT4-MIXED

1024

587.4

41.9

23.86634845

qwen2.5-7b-instruct-1m

INT4-MIXED

32

54.7

42

23.80952381

qwen3-vl-4b-instruct

INT4-MIXED

4937

6781.2

42

23.80952381

qwen2.5-7b-instruct

INT4-MIXED

32

56.5

42.2

23.69668246

qwen2.5-7b-instruct

INT4-MIXED

32

56.8

42.3

23.64066194

phi-4-multimodal-instruct

INT4-MIXED

1570

1448.8

42.4

23.58490566

stable-diffusion-xl-1.0-inpainting-0.1

INT8-CW

32

43.7

42.4

23.58490566

minicpm3-4b

INT4-MIXED

1024

825.3

42.6

23.4741784

phi-4-multimodal-instruct

INT4-MIXED

1362

1303.4

42.6

23.4741784

stable-diffusion-xl-1.0-inpainting-0.1

INT8-CW

32

44.9

42.6

23.4741784

falcon-7b-instruct

INT4-MIXED

1024

527.7

42.8

23.36448598

llama-3-8b-instruct

INT4-MIXED

32

58.3

42.9

23.31002331

deepseek-r1-distill-llama-8b

INT4-MIXED

35

65.7

43

23.25581395

minicpm3-4b

INT4-MIXED

1024

780.7

43

23.25581395

qwen2.5-vl-7b-instruct

INT4-MIXED

32

79.8

43

23.25581395

gemma-4-e2b-it

INT8-CW

1024

508.7

43.1

23.20185615

minicpm-v-2_6

INT4-MIXED

228

546.7

43.2

23.14814815

llama-3.1-8b-instruct

INT4-MIXED

32

59.6

43.4

23.04147465

minicpm-o-2_6

INT4-MIXED

238

538.8

43.4

23.04147465

mistral-7b-instruct-v0.2

INT4-MIXED

1024

555.8

43.4

23.04147465

mistral-7b-instruct-v0.3

INT4-MIXED

1024

554.9

43.4

23.04147465

deepseek-r1-distill-qwen-7b

INT4-MIXED

32

58.7

43.5

22.98850575

qwen2-7b-instruct

INT4-MIXED

32

60.1

43.6

22.93577982

neural-chat-7b-v3-3

INT4-MIXED

1024

625.4

43.7

22.88329519

qwen2.5-7b-instruct

INT4-MIXED

32

59.8

43.7

22.88329519

qwen2.5-7b-instruct-1m

INT4-MIXED

1024

458.7

43.7

22.88329519

qwen2.5-7b-instruct

INT4-MIXED

1024

460.2

43.8

22.83105023

mistral-7b-instruct-v0.1

INT4-MIXED

1025

670.7

44

22.72727273

mistral-7b-instruct-v0.3

INT4-MIXED

1024

630.2

44

22.72727273

qwen2-7b-instruct

INT4-MIXED

32

61.4

44

22.72727273

qwen2.5-7b-instruct

INT4-MIXED

32

60.3

44

22.72727273

qwen2.5-7b-instruct

INT4-MIXED

32

59.2

44

22.72727273

deepseek-r1-distill-qwen-7b

INT4-MIXED

32

59.8

44.1

22.67573696

qwen2.5-7b-instruct

INT4-MIXED

32

60.2

44.1

22.67573696

qwen2.5-7b-instruct

INT4-MIXED

1024

464.5

44.1

22.67573696

qwen2.5-7b-instruct-1m

INT4-MIXED

32

60.7

44.1

22.67573696

mistral-7b-instruct-v0.2

INT4-MIXED

1024

641.6

44.3

22.57336343

bloomz-7b1

INT4-MIXED

32

60.9

44.5

22.47191011

qwen3-8b

INT4-MIXED

32

68.5

44.5

22.47191011

qwen2.5-vl-7b-instruct

INT4-MIXED

1024

543.5

44.6

22.42152466

qwen2.5-vl-7b-instruct

INT4-MIXED

32

80.9

44.8

22.32142857

llama-3-8b-instruct

INT4-MIXED

32

63

44.9

22.27171492

llama-3-8b-instruct

INT4-MIXED

32

63.3

44.9

22.27171492

minicpm-v-2_6

INT4-MIXED

228

569.7

44.9

22.27171492

falcon-7b-instruct

INT4-MIXED

32

60.8

45

22.22222222

gemma-3-4b-it

INT8-CW

32

81.4

45

22.22222222

llama-3.1-8b-instruct

INT4-MIXED

32

63.5

45

22.22222222

phi-4-mini-reasoning

INT8-CW

32

68.5

45

22.22222222

fara-7b

INT4-MIXED

32

111.6

45.1

22.172949

minicpm-v-2_6

INT4-MIXED

228

578.7

45.1

22.172949

minicpm4-8b

INT4-MIXED

32

65.4

45.1

22.172949

deepseek-r1-distill-llama-8b

INT4-MIXED

35

67.2

45.3

22.07505519

deepseek-r1-distill-qwen-7b

INT4-MIXED

1024

505.1

45.3

22.07505519

llama-3-8b-instruct

INT4-MIXED

1024

515.2

45.3

22.07505519

qwen2-7b-instruct

INT4-MIXED

1024

501.3

45.4

22.02643172

qwen2.5-7b-instruct

INT4-MIXED

1024

500.9

45.4

22.02643172

deepseek-r1-distill-llama-8b

INT4-MIXED

1135

581.7

45.5

21.97802198

minicpm-o-2_6

INT4-MIXED

238

566.6

45.5

21.97802198

llama-3-8b-instruct

INT4-MIXED

32

63.2

45.6

21.92982456

qwen2.5-7b-instruct

INT4-MIXED

1024

501.2

45.6

21.92982456

minicpm-v-4_5

INT4-MIXED

217

569.5

45.7

21.88183807

deepseek-r1-distill-qwen-7b

INT4-MIXED

1024

572.8

45.8

21.83406114

phi-4-mini-instruct

INT8-CW

32

67.2

45.8

21.83406114

qwen2-7b-instruct

INT4-MIXED

1024

572.1

45.8

21.83406114

qwen2.5-7b-instruct

INT4-MIXED

1024

574.7

45.8

21.83406114

qwen2.5-7b-instruct

INT4-MIXED

1024

574.1

45.9

21.78649237

llama-3.1-8b-instruct

INT4-MIXED

1024

518.4

46

21.73913043

qwen2.5-7b-instruct-1m

INT4-MIXED

1024

578.3

46.1

21.69197397

qwen2.5-vl-7b-instruct

INT4-MIXED

1024

599.7

46.2

21.64502165

qwen3-4b

INT8-CW

32

65.4

46.2

21.64502165

qwen2.5-vl-7b-instruct

INT4-MIXED

32

82.8

46.3

21.59827214

qwen3-8b

INT4-MIXED

32

67.9

46.5

21.50537634

minicpm4-8b

INT4-MIXED

1024

553.3

46.7

21.41327623

qwen2.5-vl-7b-instruct

INT4-MIXED

1024

677.6

46.7

21.41327623

bloomz-7b1

INT4-MIXED

1024

677.4

46.8

21.36752137

fara-7b

INT4-MIXED

1024

786.1

46.8

21.36752137

phi-4-mini-reasoning

INT8-CW

1024

402.4

46.8

21.36752137

bloomz-7b1

INT4-MIXED

32

66.9

47.1

21.23142251

gemma-3-4b-it

INT8-CW

1024

407.2

47.1

21.23142251

minicpm4-8b

INT4-MIXED

32

71.9

47.1

21.23142251

qwen3-8b

INT4-MIXED

1024

545.4

47.1

21.23142251

afm-4.5b

INT8-CW

32

60.4

47.2

21.18644068

qwen3-8b

INT4-MIXED

32

70.1

47.2

21.18644068

falcon-7b-instruct

INT4-MIXED

1024

637

47.3

21.14164905

llama-3-8b-instruct

INT4-MIXED

1024

557.9

47.3

21.14164905

llama-3-8b-instruct

INT4-MIXED

1024

562

47.4

21.09704641

phi-3.5-vision-instruct

INT8-CW

802

503.6

47.5

21.05263158

deepseek-r1-distill-llama-8b

INT4-MIXED

1135

636.1

47.6

21.00840336

phi-4-mini-instruct

INT8-CW

1024

410.3

47.6

21.00840336

qwen3_8b_draft_eagle3

INT4-MIXED

32

65.4

47.6

21.00840336

minicpm4-8b

INT4-MIXED

32

71.4

47.7

20.96436059

llama-3.1-8b-instruct

INT4-MIXED

1024

565

47.9

20.87682672

llama-3-8b-instruct

INT4-MIXED

1024

640

48

20.83333333

minicpm-v-4_5

INT4-MIXED

217

587.4

48.5

20.6185567

phi-3.5-vision-instruct

INT8-CW

1032

620.2

48.5

20.6185567

qwen3-4b

INT8-CW

1024

375.9

48.9

20.44989775

minicpm4-8b

INT4-MIXED

1024

602.7

49.1

20.36659878

afm-4.5b

INT8-CW

1024

372.1

49.2

20.32520325

bloomz-7b1

INT4-MIXED

1024

786.4

49.2

20.32520325

qwen3-8b

INT4-MIXED

1024

575.8

49.2

20.32520325

qwen3-8b

INT4-MIXED

1024

654.9

49.8

20.08032129

zephyr-7b-beta

INT4-MIXED

32

65.8

49.9

20.04008016

minicpm4-8b

INT4-MIXED

1024

688

50.2

19.92031873

gemma-7b-it

INT4-MIXED

32

74.8

51

19.60784314

zephyr-7b-beta

INT4-MIXED

1024

547

52.3

19.12045889

qwen3_8b_draft_eagle3

INT4-MIXED

1024

617.5

53.3

18.76172608

gemma-7b-it

INT4-MIXED

32

79.8

53.7

18.62197393

gemma-7b-it

INT4-MIXED

1024

623.1

53.8

18.58736059

phi-4-multimodal-instruct

INT8-CW

578

501.3

54.4

18.38235294

phi-4-multimodal-instruct

INT8-CW

786

603.3

54.8

18.24817518

qwen3.5-9b

INT4-MIXED

83

203.5

55.1

18.14882033

ltx-video

INT4-MIXED

11

56.1

55.5

18.01801802

qwen3.5-9b

INT4-MIXED

1024

956.8

55.5

18.01801802

phi-4-multimodal-instruct

INT8-CW

1570

1361.9

56

17.85714286

minicpm3-4b

INT8-CW

32

193.7

56.3

17.76198934

deepseek-r1-distill-llama-8b

INT4-MIXED

35

75.8

56.4

17.73049645

gemma-7b-it

INT4-MIXED

1024

757.9

56.4

17.73049645

llama-3.1-8b-instruct

INT4-MIXED

32

74.5

56.5

17.69911504

ltx-video

INT8-CW

11

57

56.5

17.69911504

phi-4-multimodal-instruct

INT8-CW

1362

1248.4

56.9

17.57469244

phi-2

FP16

32

64.9

57.3

17.45200698

gemma-4-e2b-it

FP16

274

348.7

57.6

17.36111111

qwen3.5-9b

INT4-MIXED

83

218.2

57.8

17.30103806

gemma-2-9b-it

INT4-MIXED

32

78.8

58.3

17.15265866

qwen3.5-9b

INT4-MIXED

1024

1143.4

58.7

17.03577513

deepseek-r1-distill-llama-8b

INT4-MIXED

1135

760.8

58.8

17.00680272

gemma-2-9b-it

INT4-MIXED

32

82.2

59.5

16.80672269

llama-3.1-8b-instruct

INT4-MIXED

1024

727.5

59.8

16.72240803

gemma-2-9b-it

INT4-MIXED

32

83.5

60

16.66666667

phi-2

FP16

1024

415.6

61.1

16.36661211

gemma-2-9b-it

INT4-MIXED

1024

687.2

61.6

16.23376623

minicpm3-4b

INT8-CW

1024

901.2

61.6

16.23376623

gemma-2-9b-it

INT4-MIXED

1024

723.3

63.1

15.84786054

gemma-2-9b-it

INT4-MIXED

1024

788.1

63.8

15.67398119

stable-diffusion-xl-1.0-inpainting-0.1

FP16

32

66.5

64.3

15.55209953

gemma-4-e2b-it

FP16

1024

580

65.4

15.29051988

lcm-dreamshaper-v7

INT8-HYBRID

1024

67.8

66.1

15.12859304

lcm-dreamshaper-v7

INT8-HYBRID

32

68.7

66.3

15.08295626

llama-3.2-3b-instruct

FP16

32

72.7

66.7

14.99250375

videochat-flash-qwen2_5-7b_internvideo2-1b

INT4-MIXED

14531

84815.8

67.6

14.79289941

dolly-v2-12b

INT4-MIXED

32

102.2

67.8

14.74926254

llama-3.2-3b-instruct

FP16

1024

384.5

68.8

14.53488372

llama-2-13b-chat-hf

INT4-MIXED

32

100.5

69.7

14.3472023

dolly-v2-12b

INT4-MIXED

1024

1253.8

72

13.88888889

smollm3-3b

FP16

32

78.1

73

13.69863014

gemma-3-12b-it

INT4-MIXED

32

128.5

73.2

13.66120219

qwen2.5-coder-3b-instruct

FP16

32

78.2

73.9

13.53179973

llama-2-13b-chat-hf

INT4-MIXED

1024

962

74

13.51351351

qwen2.5-7b-instruct

INT8-CW

32

95

74.7

13.38688086

qwen2.5-7b-instruct-1m

INT8-CW

32

94.6

74.8

13.36898396

qwen2-7b-instruct

INT8-CW

32

94.9

74.9

13.35113485

smollm3-3b

FP16

1024

308.1

75

13.33333333

deepseek-r1-distill-qwen-7b

INT8-CW

32

95

75.3

13.28021248

falcon-7b-instruct

INT8-CW

32

90.8

75.5

13.24503311

qwen2.5-7b-instruct

INT8-CW

32

94.9

75.8

13.19261214

fara-7b

INT8-CW

32

143.3

76

13.15789474

qwen2.5-coder-3b-instruct

FP16

1024

309.3

76

13.15789474

qwen2.5-vl-7b-instruct

INT8-CW

32

115.7

76.1

13.14060447

codellama-7b

INT8-CW

32

88.7

76.2

13.12335958

minicpm-o-2_6

INT8-CW

238

555.3

76.2

13.12335958

qwen2-7b-instruct

INT8-CW

1024

565.8

76.3

13.1061599

qwen2.5-7b-instruct-1m

INT8-CW

1024

565.5

76.3

13.1061599

qwen2.5-7b-instruct

INT8-CW

1024

574.4

76.4

13.08900524

gemma-3-12b-it

INT4-MIXED

32

134.5

76.9

13.00390117

phi-3-mini-128k-instruct

FP16

32

84.6

77

12.98701299

gemma-3-12b-it

INT4-MIXED

1024

1046.4

77.2

12.95336788

minicpm-v-2_6

INT8-CW

228

552.6

77.3

12.93661061

phi-3.5-mini-instruct

FP16

32

85.3

77.4

12.91989664

qwen2.5-vl-7b-instruct

INT8-CW

1024

650.8

77.4

12.91989664

phi-3-mini-4k-instruct

FP16

32

84.7

77.5

12.90322581

qwen2.5-7b-instruct

INT8-CW

1024

563.6

77.5

12.90322581

deepseek-r1-distill-qwen-7b

INT8-CW

1024

572.5

77.6

12.88659794

falcon-7b-instruct

INT8-CW

1024

793

77.6

12.88659794

fara-7b

INT8-CW

1024

786.1

77.6

12.88659794

lcm-dreamshaper-v7

INT8-CW

1024

78.9

77.6

12.88659794

phi-4

INT4-MIXED

32

112

77.6

12.88659794

phi-4-reasoning

INT4-MIXED

32

112

77.6

12.88659794

llama-2-7b-chat-hf

INT8-CW

32

88.6

78.4

12.75510204

lcm-dreamshaper-v7

INT8-CW

32

80.5

78.8

12.69035533

lcm-dreamshaper-v7

FP16

32

80.9

79.5

12.57861635

codellama-7b

INT8-CW

1024

556.1

79.8

12.53132832

lcm-dreamshaper-v7

FP16

1024

79.8

79.9

12.51564456

lcm-dreamshaper-v7

INT8-CW

1024

82.3

80

12.5

phi-4-mini-reasoning

FP16

32

89.4

80.3

12.45330012

lcm-dreamshaper-v7

INT8-CW

32

79.4

80.4

12.43781095

phi-4-mini-instruct

FP16

32

88.8

80.4

12.43781095

neural-chat-7b-v3-3

INT8-CW

32

94.7

80.6

12.40694789

zephyr-7b-beta

INT8-CW

32

95.5

80.7

12.39157373

phi-3.5-vision-instruct

FP16

802

702.4

80.8

12.37623762

gemma-3-12b-it

INT4-MIXED

1024

1194.5

81.1

12.33045623

biomistral-7b-slerp

INT8-CW

7

85.9

81.3

12.300123

phi-3-mini-128k-instruct

FP16

1024

553.8

81.3

12.300123

phi-4

INT4-MIXED

1024

1094.3

81.3

12.300123

qwen1.5-14b-chat

INT4-MIXED

32

115.9

81.3

12.300123

qwen3_8b_draft_eagle3

INT8-CW

1024

635.1

81.3

12.300123

mistral-7b-instruct-v0.2

INT8-CW

32

93.8

81.4

12.28501229

mistral-7b-instruct-v0.1

INT8-CW

32

93.8

81.5

12.26993865

phi-3-mini-4k-instruct

FP16

1024

562.4

81.6

12.25490196

phi-3.5-mini-instruct

FP16

1024

552.1

81.8

12.22493888

phi-3.5-vision-instruct

FP16

1032

877.6

81.9

12.21001221

phi-4-reasoning

INT4-MIXED

1024

1072.1

81.9

12.21001221

mistral-7b-instruct-v0.3

INT8-CW

32

93.8

82

12.19512195

llama-2-7b-chat-hf

INT8-CW

1024

566.1

82.2

12.16545012

phi-4-mini-instruct

FP16

1024

539.6

82.3

12.15066829

phi-4-reasoning

INT4-MIXED

32

123.5

82.3

12.15066829

phi-4-mini-reasoning

FP16

1024

539.1

82.6

12.10653753

phi-4

INT4-MIXED

32

125

82.8

12.07729469

neural-chat-7b-v3-3

INT8-CW

1024

590.9

83

12.04819277

deepseek-r1-distill-qwen-14b

INT4-MIXED

32

114.3

83.1

12.03369434

qwen3_8b_draft_eagle3

INT8-CW

32

105.6

83.2

12.01923077

zephyr-7b-beta

INT8-CW

1024

584

83.2

12.01923077

gemma-3-4b-it

FP16

32

93.4

83.3

12.00480192

bloomz-7b1

INT8-CW

32

96.3

83.9

11.91895113

mistral-7b-instruct-v0.1

INT8-CW

1025

634.5

83.9

11.91895113

mistral-7b-instruct-v0.2

INT8-CW

1024

586.8

84

11.9047619

gemma-3-4b-it

FP16

1024

515.8

84.1

11.89060642

mistral-7b-instruct-v0.3

INT8-CW

1024

590.4

84.2

11.87648456

qwen1.5-14b-chat

INT4-MIXED

1024

1152.9

84.2

11.87648456

deepseek-r1-distill-llama-8b

INT8-CW

35

108.6

84.9

11.77856302

llama-3.1-8b-instruct

INT8-CW

32

99

85

11.76470588

phi-4-reasoning

INT4-MIXED

1024

1177.6

85.8

11.65501166

phi-4-reasoning

INT4-MIXED

32

124.2

86.2

11.60092807

phi-4

INT4-MIXED

1024

1301.1

86.4

11.57407407

qwen3-4b

FP16

32

93.5

86.6

11.54734411

llama-3-8b-instruct

INT8-CW

32

100.1

86.7

11.53402537

llama-3.1-8b-instruct

INT8-CW

1024

600.5

87.2

11.46788991

bloomz-7b1

INT8-CW

1024

792.2

87.4

11.4416476

deepseek-r1-distill-qwen-14b

INT4-MIXED

1024

1175.1

87.5

11.42857143

llama-3-8b-instruct

INT8-CW

1024

610.4

88.3

11.32502831

llava-v1.6-mistral-7b-hf

INT8-CW

2944

2875.9

88.3

11.32502831

deepseek-r1-distill-llama-8b

INT8-CW

1135

663.6

88.5

11.29943503

llama-2-13b-chat-hf

INT4-MIXED

32

121.3

88.9

11.24859393

starcoder

INT4-MIXED

32

139.1

89.5

11.17318436

phi-4-reasoning

INT4-MIXED

1024

1347.6

90.1

11.09877913

afm-4.5b

FP16

32

99.5

90.3

11.07419712

qwen3-4b

FP16

1024

476.6

90.3

11.07419712

qwen3-8b

INT8-CW

32

102.1

91.8

10.89324619

minicpm-v-4_5

INT8-CW

217

559.8

92.2

10.84598698

afm-4.5b

FP16

1024

483.9

92.4

10.82251082

llama-2-13b-chat-hf

INT4-MIXED

1024

1172.7

93.3

10.71811361

starcoder

INT4-MIXED

1024

1388.6

93.6

10.68376068

gemma-7b-it

INT8-CW

32

115

93.9

10.64962726

qwen3-8b

INT8-CW

1024

644.9

94

10.63829787

minicpm4-8b

INT8-CW

32

107.2

94.5

10.58201058

minicpm3-4b

FP16

32

171.3

94.8

10.54852321

phi-4-multimodal-instruct

FP16

786

764.1

95.1

10.51524711

phi-4-multimodal-instruct

FP16

578

637.4

95.6

10.46025105

phi-4-multimodal-instruct

FP16

1362

1535.3

96.2

10.3950104

minicpm4-8b

INT8-CW

1024

698.9

96.6

10.35196687

starcoder2-15b

INT4-MIXED

32

146.2

97.5

10.25641026

gemma-7b-it

INT8-CW

1024

744.1

98.2

10.18329939

phi-4-multimodal-instruct

FP16

1570

1678.8

98.7

10.13171226

Topology

Precision

Input Size

1st latency (ms)

2nd latency (ms)

2nd token per sec

minicpm4-0.5b

INT4-MIXED

32

22

7.3

136.9863014

minicpm4-0.5b

INT4-MIXED

32

21.1

7.5

133.3333333

minicpm4-0.5b

INT4-MIXED

32

23

7.7

129.8701299

gemma-3-270m

INT4-MIXED

1024

67.1

7.9

126.5822785

gemma-3-270m

INT4-MIXED

32

23.2

8

125

lfm2.5-350m

INT4-MIXED

32

24.7

8.2

121.9512195

gemma-3-270m

INT8-CW

32

22.9

8.3

120.4819277

lfm2.5-350m

INT4-MIXED

1024

155

8.4

119.047619

minicpm4-0.5b

INT4-MIXED

1024

100.2

8.4

119.047619

gemma-3-270m

INT8-CW

1024

63.2

8.7

114.9425287

minicpm4-0.5b

INT4-MIXED

1024

131.5

8.7

114.9425287

minicpm4-0.5b

INT4-MIXED

1024

103.7

8.7

114.9425287

qwen2.5-coder-0.5b-instruct

INT4-MIXED

32

22.4

8.9

112.3595506

nuextract-tiny

INT4-MIXED

32

23.7

9

111.1111111

qwen2.5-coder-0.5b-instruct

INT4-MIXED

32

22.5

9.1

109.8901099

qwen2.5-coder-0.5b-instruct

INT4-MIXED

32

23.2

9.2

108.6956522

lfm2.5-350m

INT8-CW

32

26.2

9.7

103.0927835

nuextract-tiny

INT4-MIXED

1024

125.7

9.8

102.0408163

qwen2.5-coder-0.5b-instruct

INT4-MIXED

1024

105

9.8

102.0408163

lfm2.5-350m

INT8-CW

1024

145.1

10

100

qwen2.5-coder-0.5b-instruct

INT4-MIXED

1024

108.5

10

100

qwen2.5-coder-0.5b-instruct

INT4-MIXED

1024

134.9

10.1

99.00990099

minicpm4-0.5b

INT8-CW

32

23

10.4

96.15384615

whisper-small

INT4-MIXED

32

139.8

10.6

94.33962264

whisper-small

INT4-MIXED

1024

208.6

11

90.90909091

nuextract-tiny

INT8-CW

32

24.9

11.1

90.09009009

whisper-small

INT4-MIXED

32

141.5

11.2

89.28571429

whisper-small

INT4-MIXED

1024

200.3

11.2

89.28571429

minicpm4-0.5b

INT8-CW

1024

116.1

11.5

86.95652174

distil-large-v2

INT4-MIXED

32

441.9

11.6

86.20689655

distil-large-v2

INT8-CW

32

411.3

11.7

85.47008547

distil-large-v2

INT8-CW

1024

467.8

11.7

85.47008547

whisper-small

INT4-MIXED

32

154.8

11.7

85.47008547

whisper-small

INT4-MIXED

1024

225.2

11.7

85.47008547

qwen2.5-coder-0.5b-instruct

INT8-CW

32

25.4

11.8

84.74576271

distil-large-v2

INT4-MIXED

1024

496

11.9

84.03361345

whisper-small

INT8-CW

32

150.7

11.9

84.03361345

whisper-small

INT8-CW

1024

226.4

11.9

84.03361345

gemma-3-270m

FP16

32

22.5

12.3

81.30081301

whisper-large-v3-turbo

INT8-CW

1024

481.2

12.4

80.64516129

nanollava

INT4-MIXED

760

249.8

12.5

80

nuextract-tiny

INT8-CW

1024

110.1

12.6

79.36507937

qwen2.5-coder-0.5b-instruct

INT8-CW

1024

124.3

12.6

79.36507937

whisper-large-v3-turbo

INT4-MIXED

1024

527.3

12.7

78.74015748

whisper-large-v3-turbo

INT8-CW

32

423.8

12.7

78.74015748

gemma-3-270m

FP16

1024

72.2

12.8

78.125

whisper-large-v3-turbo

INT4-MIXED

32

450.8

12.8

78.125

tiny-llama-1.1b-chat

INT4-MIXED

32

28.7

13.1

76.33587786

nanollava

INT8-CW

760

239.6

13.7

72.99270073

minicpm5-1b

INT4-MIXED

32

28.2

13.9

71.94244604

qwen3.5-0.8b

INT4-MIXED

85

61.5

14.1

70.92198582

nanollava

INT4-MIXED

1752

481.1

14.3

69.93006993

qwen3.5-0.8b

INT4-MIXED

85

65.1

14.3

69.93006993

qwen3.5-0.8b

INT4-MIXED

1024

347

14.5

68.96551724

whisper-small

FP16

32

139.7

14.7

68.02721088

whisper-small

FP16

1024

212.9

14.7

68.02721088

qwen3.5-0.8b

INT4-MIXED

85

70.4

14.8

67.56756757

qwen3.5-0.8b

INT4-MIXED

1024

349.5

14.9

67.11409396

tiny-llama-1.1b-chat

INT4-MIXED

1024

219.5

14.9

67.11409396

qwen3.5-0.8b

INT4-MIXED

1024

404.5

15.2

65.78947368

lfm2.5-1.2b-instruct

INT4-MIXED

32

34

15.4

64.93506494

lfm2-1.2b

INT4-MIXED

32

33.7

15.5

64.51612903

nanollava

INT8-CW

1752

458

15.7

63.69426752

distil-large-v2

FP16

32

437.5

15.8

63.29113924

distil-large-v2

FP16

1024

494.2

15.8

63.29113924

tiny-llama-1.1b-chat

INT4-MIXED

32

34

15.8

63.29113924

lfm2.5-1.2b-instruct

INT4-MIXED

1024

333.7

15.9

62.89308176

lfm2.5-1.2b-instruct

INT4-MIXED

32

40.4

15.9

62.89308176

lfm2-1.2b

INT4-MIXED

1024

316.1

16

62.5

lfm2-1.2b

INT4-MIXED

32

39.8

16

62.5

minicpm5-1b

INT4-MIXED

1024

224.3

16

62.5

lfm2-1.2b

INT4-MIXED

32

39.8

16.1

62.11180124

lfm2.5-1.2b-instruct

INT4-MIXED

32

41.3

16.1

62.11180124

llama-3.2-1b-instruct

INT4-MIXED

32

31.2

16.1

62.11180124

lfm2.5-350m

FP16

32

28

16.2

61.72839506

lfm2-1.2b

INT4-MIXED

1024

361.6

16.3

61.34969325

lfm2.5-1.2b-instruct

INT4-MIXED

1024

376.8

16.3

61.34969325

llama-3.2-1b-instruct

INT4-MIXED

32

32.7

16.3

61.34969325

lfm2.5-350m

FP16

1024

159.1

16.4

60.97560976

whisper-large-v3-turbo

FP16

32

457.2

16.4

60.97560976

whisper-large-v3-turbo

FP16

1024

515.6

16.4

60.97560976

lfm2-1.2b

INT4-MIXED

1024

410.4

16.5

60.60606061

lfm2.5-1.2b-instruct

INT4-MIXED

1024

422.2

16.7

59.88023952

gemma-3-1b-it

INT4-MIXED

32

33.1

17.1

58.47953216

llama-3.2-1b-instruct

INT4-MIXED

1024

210.5

17.4

57.47126437

exaone-4.0-1.2b

INT4-MIXED

32

33.5

17.5

57.14285714

llama-3.2-1b-instruct

INT4-MIXED

1024

257.9

17.6

56.81818182

tiny-llama-1.1b-chat

INT4-MIXED

1024

299.1

17.6

56.81818182

qwen3.5-0.8b

INT8-CW

85

71.7

17.8

56.17977528

gemma-3-1b-it

INT4-MIXED

32

34.5

17.9

55.86592179

exaone-4.0-1.2b

INT4-MIXED

32

37.8

18

55.55555556

gemma-3-1b-it

INT4-MIXED

1024

206.2

18.1

55.24861878

exaone-4.0-1.2b

INT4-MIXED

32

38.1

18.2

54.94505495

gemma-3-1b-it

INT4-MIXED

32

34.7

18.2

54.94505495

qwen3.5-0.8b

INT8-CW

1024

363.7

18.6

53.76344086

gemma-3-1b-it

INT4-MIXED

1024

209.8

19

52.63157895

minicpm5-1b

INT8-CW

32

35.9

19.2

52.08333333

gemma-3-1b-it

INT4-MIXED

1024

244.7

19.3

51.8134715

minicpm4-0.5b

FP16

32

23.7

19.3

51.8134715

nuextract-tiny

FP16

32

28.7

19.3

51.8134715

minicpm-1b-sft

INT4-MIXED

31

55.6

19.4

51.54639175

exaone-4.0-1.2b

INT4-MIXED

1024

336.1

19.9

50.25125628

minicpm4-0.5b

FP16

1024

133.5

20.3

49.26108374

exaone-4.0-1.2b

INT4-MIXED

1024

344.9

20.5

48.7804878

exaone-4.0-1.2b

INT4-MIXED

1024

378.4

20.7

48.30917874

minicpm5-1b

INT8-CW

1024

210.1

20.7

48.30917874

qwen2.5-coder-1.5b-instruct

INT4-MIXED

32

41.2

20.7

48.30917874

minicpm-1b-sft

INT4-MIXED

31

45.2

20.8

48.07692308

deepseek-r1-distill-qwen-1.5b

INT4-MIXED

32

43.2

21.3

46.94835681

minicpm-1b-sft

INT4-MIXED

31

47.5

21.4

46.72897196

tiny-llama-1.1b-chat

INT8-CW

32

38.2

21.4

46.72897196

nuextract-tiny

FP16

1024

131.4

21.5

46.51162791

qwen2.5-coder-1.5b-instruct

INT4-MIXED

32

44.9

21.6

46.2962963

qwen2.5-1.5b-instruct

INT4-MIXED

32

44.2

21.7

46.08294931

nanollava

FP16

760

284.9

21.8

45.87155963

qwen2.5-coder-0.5b-instruct

FP16

32

27.8

21.9

45.66210046

qwen2.5-coder-1.5b-instruct

INT4-MIXED

32

45.4

22.2

45.04504505

qwen2.5-coder-0.5b-instruct

FP16

1024

143.1

22.4

44.64285714

qwen2.5-coder-1.5b-instruct

INT4-MIXED

1024

292.2

22.4

44.64285714

minicpm-1b-sft

INT4-MIXED

1014

323.1

22.6

44.24778761

deepseek-r1-distill-qwen-1.5b

INT4-MIXED

1024

344.9

23

43.47826087

gemma-3-1b-it

INT8-CW

32

42.8

23.1

43.29004329

qwen2.5-coder-1.5b-instruct

INT4-MIXED

1024

347.2

23.2

43.10344828

tiny-llama-1.1b-chat

INT8-CW

1024

247.1

23.3

42.91845494

qwen2.5-1.5b-instruct

INT4-MIXED

1024

344.8

23.5

42.55319149

qwen2.5-1.5b-instruct

INT4-MIXED

32

47.7

23.6

42.37288136

nanollava

FP16

1752

508.9

23.9

41.84100418

qwen2.5-coder-1.5b-instruct

INT4-MIXED

1024

359.9

23.9

41.84100418

minicpm-1b-sft

INT4-MIXED

1014

335.7

24.2

41.32231405

gemma-3-1b-it

INT8-CW

1024

245.3

24.3

41.15226337

lfm2-8b-a1b

INT4-MIXED

32

381.5

24.5

40.81632653

minicpm-1b-sft

INT4-MIXED

1014

389.3

24.6

40.6504065

llama-3.2-1b-instruct

INT8-CW

32

41

24.7

40.48582996

lfm2-1.2b

INT8-CW

32

50.9

25.2

39.68253968

lfm2-8b-a1b

INT4-MIXED

1024

881.8

25.2

39.68253968

lfm2.5-1.2b-instruct

INT8-CW

32

50.9

25.3

39.5256917

qwen2.5-1.5b-instruct

INT4-MIXED

1024

369.1

25.4

39.37007874

lfm2-1.2b

INT8-CW

1024

410.1

25.9

38.61003861

lfm2.5-1.2b-instruct

INT8-CW

1024

391.1

26

38.46153846

llama-3.2-1b-instruct

INT8-CW

1024

243.1

26.2

38.16793893

exaone-4.0-1.2b

INT8-CW

32

47.5

27.2

36.76470588

exaone-4.0-1.2b

INT8-CW

1024

372.1

29.8

33.55704698

deepseek-r1-distill-qwen-1.5b

INT4-MIXED

32

59.9

30

33.33333333

minicpm-1b-sft

INT8-CW

31

56

31.1

32.15434084

qwen3.5-0.8b

FP16

85

74.7

31.2

32.05128205

qwen3.5-0.8b

FP16

1024

388.1

31.6

31.64556962

deepseek-r1-distill-qwen-1.5b

INT4-MIXED

1024

353.7

31.8

31.44654088

deepseek-r1-distill-qwen-1.5b

INT8-CW

32

59.6

32.3

30.95975232

qwen2.5-coder-1.5b-instruct

INT8-CW

32

62.4

32.9

30.39513678

stable-zephyr-3b-dpo

INT4-MIXED

32

62.3

32.9

30.39513678

qwen2.5-1.5b-instruct

INT8-CW

32

62.3

33

30.3030303

phi-2

INT4-MIXED

32

64.7

33.4

29.94011976

deepseek-r1-distill-qwen-1.5b

INT8-CW

1024

340.6

33.7

29.6735905

minicpm-1b-sft

INT8-CW

1014

369.3

34.1

29.3255132

qwen2.5-1.5b-instruct

INT8-CW

1024

350

34.3

29.15451895

qwen2.5-coder-1.5b-instruct

INT8-CW

1024

350.6

34.3

29.15451895

gemma-4-e2b-it

INT4-MIXED

274

720.1

34.7

28.8184438

minicpm5-1b

FP16

32

44

36.3

27.54820937

gemma-4-e2b-it

INT4-MIXED

274

758.7

36.4

27.47252747

whisper-large-v3

INT4-MIXED

32

578.5

36.6

27.32240437

stable-zephyr-3b-dpo

INT4-MIXED

32

77

37

27.02702703

gemma-4-e2b-it

INT4-MIXED

1024

1171.1

37.4

26.73796791

stable-zephyr-3b-dpo

INT4-MIXED

1024

787.5

37.5

26.66666667

minicpm5-1b

FP16

1024

257.7

37.6

26.59574468

phi-2

INT4-MIXED

1024

719.8

38

26.31578947

whisper-large-v3

INT4-MIXED

1024

636.2

38.2

26.17801047

llama-3.2-3b-instruct

INT4-MIXED

32

64.9

38.3

26.10966057

gemma-4-e2b-it

INT4-MIXED

1024

1311

39

25.64102564

llama-3.2-3b-instruct

INT4-MIXED

32

77.5

39.4

25.38071066

smollm3-3b

INT4-MIXED

32

71.9

39.5

25.3164557

phi-2

INT4-MIXED

32

84.8

39.7

25.18891688

llama-3.2-3b-instruct

INT4-MIXED

32

78.4

40

25

qwen2.5-coder-3b-instruct

INT4-MIXED

32

71.6

40.4

24.75247525

smollm3-3b

INT4-MIXED

32

82.3

40.7

24.57002457

tiny-llama-1.1b-chat

FP16

32

50.2

40.7

24.57002457

qwen2.5-coder-3b-instruct

INT4-MIXED

32

82.5

41.1

24.33090024

smollm3-3b

INT4-MIXED

32

83.5

41.1

24.33090024

stable-zephyr-3b-dpo

INT4-MIXED

1024

886

41.6

24.03846154

llama-3.2-3b-instruct

INT4-MIXED

1024

555.8

41.8

23.92344498

tiny-llama-1.1b-chat

FP16

1024

315.5

42

23.80952381

gemma-3-1b-it

FP16

32

52.5

42.3

23.64066194

smollm3-3b

INT4-MIXED

1024

538.5

42.5

23.52941176

llama-3.2-3b-instruct

INT4-MIXED

1024

595.4

43

23.25581395

whisper-large-v3

INT8-CW

32

546.6

43.2

23.14814815

gemma-3-1b-it

FP16

1024

293

43.4

23.04147465

llama-3.2-3b-instruct

INT4-MIXED

1024

677.8

43.4

23.04147465

qwen2.5-coder-3b-instruct

INT4-MIXED

1024

514.5

43.4

23.04147465

smollm3-3b

INT4-MIXED

1024

576.5

43.7

22.88329519

whisper-large-v3

INT8-CW

1024

609.2

43.7

22.88329519

smollm3-3b

INT4-MIXED

1024

684.3

44.1

22.67573696

qwen2.5-coder-3b-instruct

INT4-MIXED

1024

535.7

44.3

22.57336343

phi-2

INT4-MIXED

1024

973.5

44.6

22.42152466

lfm2.5-1.2b-instruct

FP16

32

63.6

47.6

21.00840336

lfm2-1.2b

FP16

32

64.1

47.8

20.92050209

qwen2.5-coder-3b-instruct

INT4-MIXED

32

95.6

48

20.83333333

lfm2.5-1.2b-instruct

FP16

1024

453.7

48.2

20.74688797

lfm2-1.2b

FP16

1024

440.9

48.3

20.70393375

llama-3.2-1b-instruct

FP16

32

58.4

48.4

20.66115702

qwen3-4b

INT4-MIXED

32

88.8

49.2

20.32520325

phi-4-mini-instruct

INT4-MIXED

32

86.6

49.8

20.08032129

phi-4-mini-reasoning

INT4-MIXED

32

85.6

49.9

20.04008016

llama-3.2-1b-instruct

FP16

1024

316.3

50

20

gemma-3-4b-it

INT4-MIXED

32

83.9

50.2

19.92031873

qwen3-4b

INT4-MIXED

32

103

50.8

19.68503937

phi-4-mini-instruct

INT4-MIXED

32

99.6

51

19.60784314

phi-4-mini-reasoning

INT4-MIXED

32

100.5

51.1

19.56947162

qwen2.5-coder-3b-instruct

INT4-MIXED

1024

1003.6

51.1

19.56947162

phi-4-mini-reasoning

INT4-MIXED

32

102.2

51.4

19.45525292

afm-4.5b

INT4-MIXED

32

103.3

51.7

19.34235977

gemma-3-4b-it

INT4-MIXED

32

98.3

51.8

19.30501931

gemma-4-e2b-it

INT8-CW

274

838.5

52.1

19.19385797

gemma-3-4b-it

INT4-MIXED

32

98.3

52.2

19.15708812

phi-3.5-vision-instruct

INT4-MIXED

802

1187.9

53.1

18.83239171

gemma-3-4b-it

INT4-MIXED

1024

848.8

53.8

18.58736059

phi-4-mini-instruct

INT4-MIXED

1024

867.6

54

18.51851852

phi-4-mini-reasoning

INT4-MIXED

1024

871

54

18.51851852

phi-3.5-vision-instruct

INT4-MIXED

1032

1427.3

54.8

18.24817518

exaone-4.0-1.2b

FP16

32

64.5

55

18.18181818

phi-4-mini-instruct

INT4-MIXED

1024

910.6

55.2

18.11594203

gemma-3-4b-it

INT4-MIXED

1024

877.6

55.3

18.08318264

gemma-4-e2b-it

INT8-CW

1024

1354.4

55.3

18.08318264

afm-4.5b

INT4-MIXED

1024

888

55.4

18.05054152

qwen3-4b

INT4-MIXED

1024

775.1

55.4

18.05054152

phi-4-mini-reasoning

INT4-MIXED

1024

913.9

55.5

18.01801802

phi-2

INT8-CW

32

105.9

55.6

17.98561151

phi-4-mini-reasoning

INT4-MIXED

1024

985.3

55.6

17.98561151

gemma-3-4b-it

INT4-MIXED

1024

946.1

55.8

17.92114695

minicpm-1b-sft

FP16

31

68.3

56.3

17.76198934

qwen3-4b

INT4-MIXED

1024

814.6

56.9

17.57469244

exaone-4.0-1.2b

FP16

1024

449

57.6

17.36111111

phi-4-mini-instruct

INT4-MIXED

32

116.1

57.9

17.27115717

minicpm3-4b

INT4-MIXED

32

125

58.8

17.00680272

minicpm-1b-sft

FP16

1014

453.8

59.4

16.83501684

phi-2

INT8-CW

1024

842.8

60.3

16.58374793

minicpm3-4b

INT4-MIXED

32

132.2

60.7

16.47446458

deepseek-r1-distill-qwen-1.5b

FP16

32

73.8

60.9

16.42036125

minicpm3-4b

INT4-MIXED

32

136

61.4

16.28664495

qwen2.5-coder-1.5b-instruct

FP16

32

74.6

61.6

16.23376623

qwen2.5-1.5b-instruct

FP16

32

74.3

61.7

16.20745543

smollm3-3b

INT8-CW

32

111.8

61.7

16.20745543

phi-4-mini-instruct

INT4-MIXED

1024

1197

62

16.12903226

qwen2.5-coder-3b-instruct

INT8-CW

32

111.3

62

16.12903226

deepseek-r1-distill-qwen-1.5b

FP16

1024

435.6

62.3

16.05136437

qwen2.5-coder-1.5b-instruct

FP16

1024

442.5

62.9

15.89825119

qwen2.5-1.5b-instruct

FP16

1024

440

63

15.87301587

phi-4-multimodal-instruct

INT4-MIXED

578

1039.3

63.4

15.77287066

phi-4-multimodal-instruct

INT4-MIXED

786

1257.1

64.1

15.60062402

smollm3-3b

INT8-CW

1024

671.4

64.1

15.60062402

qwen2.5-coder-3b-instruct

INT8-CW

1024

652.2

64.2

15.57632399

whisper-large-v3

FP16

32

583.8

64.2

15.57632399

llama-3.2-3b-instruct

INT8-CW

32

112.1

64.4

15.52795031

whisper-large-v3

FP16

1024

656

64.6

15.47987616

phi-4-multimodal-instruct

INT4-MIXED

1362

2538.4

66.2

15.10574018

phi-4-multimodal-instruct

INT4-MIXED

1570

2811.8

67

14.92537313

llama-3.2-3b-instruct

INT8-CW

1024

714.5

67.3

14.85884101

minicpm3-4b

INT4-MIXED

1024

2139.5

67.5

14.81481481

codellama-7b

INT4-MIXED

32

122.8

68.9

14.5137881

minicpm3-4b

INT4-MIXED

1024

2186.1

69.3

14.43001443

minicpm3-4b

INT4-MIXED

1024

2315.3

70

14.28571429

llama-2-7b-chat-hf

INT4-MIXED

32

121.9

70.1

14.26533524

biomistral-7b-slerp

INT4-MIXED

7

81.8

72.3

13.83125864

codellama-7b

INT4-MIXED

32

156.6

72.4

13.8121547

llama-2-7b-chat-hf

INT4-MIXED

32

154.8

72.9

13.71742112

neural-chat-7b-v3-3

INT4-MIXED

32

137.5

73.7

13.56852103

zephyr-7b-beta

INT4-MIXED

32

133.2

74.1

13.49527665

falcon-7b-instruct

INT4-MIXED

32

131.7

74.3

13.4589502

codellama-7b

INT4-MIXED

1024

1134.9

74.4

13.44086022

llama-2-7b-chat-hf

INT4-MIXED

1024

1132.6

75.8

13.19261214

biomistral-7b-slerp

INT4-MIXED

7

85.8

75.9

13.17523057

phi-4-multimodal-instruct

INT4-MIXED

578

1330.4

76

13.15789474

qwen3-vl-4b-instruct

INT4-MIXED

4937

14679.5

76

13.15789474

mistral-7b-instruct-v0.2

INT4-MIXED

32

134.5

76.3

13.1061599

mistral-7b-instruct-v0.3

INT4-MIXED

32

137.9

76.3

13.1061599

qwen3-vl-4b-instruct

INT4-MIXED

4907

13865.4

76.4

13.08900524

phi-4-multimodal-instruct

INT4-MIXED

786

1585.9

76.5

13.07189542

qwen3-vl-4b-instruct

INT4-MIXED

4907

14153.6

76.6

13.05483029

qwen3-vl-4b-instruct

INT4-MIXED

4937

14954.7

76.7

13.03780965

qwen3-vl-4b-instruct

INT4-MIXED

4937

15633.3

77.4

12.91989664

stable-diffusion-xl-1.0-inpainting-0.1

INT8-CW

32

81.4

77.5

12.90322581

qwen3-vl-4b-instruct

INT4-MIXED

4907

14616.5

77.6

12.88659794

codellama-7b

INT4-MIXED

1024

1369.1

78

12.82051282

llama-2-7b-chat-hf

INT4-MIXED

1024

1236.8

78.5

12.7388535

phi-4-mini-instruct

INT8-CW

32

138.7

78.7

12.7064803

phi-4-mini-reasoning

INT8-CW

32

140.4

78.8

12.69035533

mistral-7b-instruct-v0.2

INT4-MIXED

32

171.1

78.9

12.67427123

mistral-7b-instruct-v0.3

INT4-MIXED

32

168.6

79.1

12.64222503

neural-chat-7b-v3-3

INT4-MIXED

1024

1232.9

79.1

12.64222503

neural-chat-7b-v3-3

INT4-MIXED

32

166.8

79.4

12.59445844

zephyr-7b-beta

INT4-MIXED

1024

1209.6

79.5

12.57861635

gemma-3-4b-it

INT8-CW

32

143.8

79.6

12.56281407

phi-4-multimodal-instruct

INT4-MIXED

1362

3223.1

79.7

12.54705144

falcon-7b-instruct

INT4-MIXED

1024

1185.7

79.8

12.53132832

mistral-7b-instruct-v0.1

INT4-MIXED

32

171

79.8

12.53132832

mistral-7b-instruct-v0.2

INT4-MIXED

32

171.2

79.8

12.53132832

mistral-7b-instruct-v0.3

INT4-MIXED

32

171.8

79.8

12.53132832

phi-4-multimodal-instruct

INT4-MIXED

1570

3470.5

79.9

12.51564456

deepseek-r1-distill-llama-8b

INT4-MIXED

32

142.8

80.4

12.43781095

qwen2.5-7b-instruct-1m

INT4-MIXED

32

150.4

80.4

12.43781095

deepseek-r1-distill-qwen-7b

INT4-MIXED

32

169.1

80.5

12.42236025

qwen2.5-7b-instruct

INT4-MIXED

32

147.4

80.6

12.40694789

minicpm-v-2_6

INT4-MIXED

228

1189.5

81.3

12.300123

deepseek-r1-distill-qwen-7b

INT4-MIXED

32

174

81.4

12.28501229

qwen3-4b

INT8-CW

32

145.4

81.5

12.26993865

minicpm-o-2_6

INT4-MIXED

238

1190.3

81.6

12.25490196

mistral-7b-instruct-v0.3

INT4-MIXED

1024

1238.6

81.6

12.25490196

mistral-7b-instruct-v0.2

INT4-MIXED

1024

1221.5

81.7

12.23990208

stable-diffusion-xl-1.0-inpainting-0.1

INT8-CW

32

83.4

81.7

12.23990208

qwen2.5-vl-7b-instruct

INT4-MIXED

32

178.9

81.8

12.22493888

llama-3.1-8b-instruct

INT4-MIXED

32

142.3

81.9

12.21001221

phi-4-mini-instruct

INT8-CW

1024

1030.3

82

12.19512195

phi-4-mini-reasoning

INT8-CW

1024

1038.4

82.1

12.18026797

deepseek-r1-distill-llama-8b

INT4-MIXED

32

173

83.1

12.03369434

qwen2-7b-instruct

INT4-MIXED

32

172.2

83.2

12.01923077

qwen2.5-7b-instruct

INT4-MIXED

32

172.2

83.2

12.01923077

qwen2.5-vl-7b-instruct

INT4-MIXED

32

205.5

83.2

12.01923077

gemma-3-4b-it

INT8-CW

1024

1010.3

83.3

12.00480192

llama-3-8b-instruct

INT4-MIXED

32

152.8

83.3

12.00480192

bloomz-7b1

INT4-MIXED

32

137.9

83.6

11.96172249

phi-3.5-vision-instruct

INT8-CW

802

1284.4

84

11.9047619

qwen2.5-7b-instruct-1m

INT4-MIXED

32

177.5

84

11.9047619

qwen2-7b-instruct

INT4-MIXED

32

175.3

84.1

11.89060642

qwen2.5-7b-instruct

INT4-MIXED

32

176.3

84.1

11.89060642

mistral-7b-instruct-v0.2

INT4-MIXED

1024

1324.9

84.3

11.8623962

mistral-7b-instruct-v0.3

INT4-MIXED

1024

1316.3

84.4

11.84834123

deepseek-r1-distill-qwen-7b

INT4-MIXED

1024

1248.9

84.5

11.83431953

llama-3-8b-instruct

INT4-MIXED

32

177.3

84.5

11.83431953

minicpm-v-2_6

INT4-MIXED

228

1232.1

84.5

11.83431953

llama-3-8b-instruct

INT4-MIXED

32

175

84.6

11.82033097

llama-3.1-8b-instruct

INT4-MIXED

32

175

84.6

11.82033097

neural-chat-7b-v3-3

INT4-MIXED

1024

1501.5

84.6

11.82033097

qwen2.5-7b-instruct

INT4-MIXED

1024

1182

84.6

11.82033097

qwen2.5-7b-instruct-1m

INT4-MIXED

1024

1183.9

84.6

11.82033097

fara-7b

INT4-MIXED

32

218.4

84.9

11.77856302

qwen2.5-vl-7b-instruct

INT4-MIXED

1024

1380.4

85

11.76470588

mistral-7b-instruct-v0.3

INT4-MIXED

1024

1505

85.1

11.75088132

llama-3-8b-instruct

INT4-MIXED

32

178.6

85.3

11.72332943

mistral-7b-instruct-v0.2

INT4-MIXED

1024

1498.6

85.3

11.72332943

qwen2.5-vl-7b-instruct

INT4-MIXED

32

207

85.3

11.72332943

qwen3-8b

INT4-MIXED

32

144.7

85.3

11.72332943

deepseek-r1-distill-qwen-7b

INT4-MIXED

1024

1430.7

85.5

11.69590643

mistral-7b-instruct-v0.1

INT4-MIXED

1025

1610.8

85.5

11.69590643

deepseek-r1-distill-llama-8b

INT4-MIXED

1024

1220.4

85.6

11.68224299

minicpm-o-2_6

INT4-MIXED

238

1207.1

85.6

11.68224299

minicpm-v-2_6

INT4-MIXED

228

1354.5

85.7

11.66861144

phi-3.5-vision-instruct

INT8-CW

1032

1568.1

85.8

11.65501166

qwen3-4b

INT8-CW

1024

954.6

86.3

11.58748552

falcon-7b-instruct

INT4-MIXED

32

187.1

86.6

11.54734411

qwen2.5-vl-7b-instruct

INT4-MIXED

1024

1394.6

86.8

11.52073733

afm-4.5b

INT8-CW

32

156.7

87

11.49425287

minicpm4-8b

INT4-MIXED

32

154.5

87

11.49425287

qwen2-7b-instruct

INT4-MIXED

1024

1263.4

87.1

11.48105626

llama-3.1-8b-instruct

INT4-MIXED

1024

1220.1

87.2

11.46788991

qwen2.5-7b-instruct

INT4-MIXED

1024

1256

87.2

11.46788991

minicpm-v-4_5

INT4-MIXED

217

1237.4

87.5

11.42857143

bloomz-7b1

INT4-MIXED

32

162.6

87.7

11.40250855

fara-7b

INT4-MIXED

1024

1898.9

87.8

11.38952164

qwen2-7b-instruct

INT4-MIXED

1024

1438.9

88

11.36363636

qwen2.5-7b-instruct-1m

INT4-MIXED

1024

1437.4

88

11.36363636

qwen2.5-7b-instruct

INT4-MIXED

1024

1439.6

88.1

11.3507378

qwen3-8b

INT4-MIXED

32

182.2

88.1

11.3507378

deepseek-r1-distill-llama-8b

INT4-MIXED

1024

1324.6

88.4

11.31221719

llama-3-8b-instruct

INT4-MIXED

1024

1252.8

88.5

11.29943503

qwen2.5-vl-7b-instruct

INT4-MIXED

1024

1579.7

88.6

11.28668172

bloomz-7b1

INT4-MIXED

1024

1602.7

89.1

11.22334456

qwen3-8b

INT4-MIXED

32

186.5

89.7

11.14827202

llama-3-8b-instruct

INT4-MIXED

1024

1314.9

89.9

11.12347052

llama-3-8b-instruct

INT4-MIXED

1024

1353.3

89.9

11.12347052

llama-3.1-8b-instruct

INT4-MIXED

1024

1323.5

90

11.11111111

afm-4.5b

INT8-CW

1024

947.5

90.3

11.07419712

zephyr-7b-beta

INT4-MIXED

32

181.7

90.5

11.04972376

llama-3-8b-instruct

INT4-MIXED

1024

1497.3

90.7

11.02535832

minicpm4-8b

INT4-MIXED

32

196

90.7

11.02535832

minicpm-v-4_5

INT4-MIXED

217

1295.8

91.3

10.95290252

qwen3-8b

INT4-MIXED

1024

1267

91.3

10.95290252

falcon-7b-instruct

INT4-MIXED

1024

1538.5

92.1

10.8577633

minicpm4-8b

INT4-MIXED

32

190

92.1

10.8577633

minicpm4-8b

INT4-MIXED

1024

1285.7

92.3

10.83423619

bloomz-7b1

INT4-MIXED

1024

1935.5

93.3

10.71811361

qwen3-8b

INT4-MIXED

1024

1376.4

94.1

10.62699256

minicpm3-4b

INT8-CW

32

180.1

94.3

10.60445387

qwen3_8b_draft_eagle3

INT4-MIXED

32

192.3

94.4

10.59322034

qwen3-8b

INT4-MIXED

1024

1579

95.6

10.46025105

zephyr-7b-beta

INT4-MIXED

1024

1398

95.6

10.46025105

qwen3.5-9b

INT4-MIXED

83

306.6

95.9

10.42752868

minicpm4-8b

INT4-MIXED

1024

1622.4

96

10.41666667

gemma-7b-it

INT4-MIXED

32

175

96.2

10.3950104

phi-4-multimodal-instruct

INT8-CW

578

1263.3

96.4

10.37344398

phi-4-multimodal-instruct

INT8-CW

786

1515

96.9

10.31991744

minicpm4-8b

INT4-MIXED

1024

1424.4

97.5

10.25641026

qwen3.5-9b

INT4-MIXED

1024

1904.6

97.5

10.25641026

gemma-7b-it

INT4-MIXED

32

208.8

100

10

phi-4-multimodal-instruct

INT8-CW

1362

3044.9

100

10

All models listed here were tested with the following parameters:

  • Framework: PyTorch

  • Beam: 1

  • Batch size: 1