Skip to content

GitLab

Switch branch/tag
Find file Normal viewHistoryPermalink
bench_autoawq_autogptq.py 4.31 KB
Newer Older
yangql's avatar
Initial commit  
yangql committed
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179 
import torch


try:
    from awq.modules.linear import WQLinear_GEMM, WQLinear_GEMV
except ModuleNotFoundError as e:
    raise ModuleNotFoundError(
        f"AutoAWQ package (https://github.com/casper-hansen/AutoAWQ) is required to run this benchmark. {e}"
    )

import numpy as np

from auto_gptq.modeling._utils import autogptq_post_init
from auto_gptq.nn_modules.qlinear.qlinear_exllamav2 import QuantLinear
from auto_gptq.utils.import_utils import dynamically_import_QuantLinear


group_size = 128
bits = 4

# Yi 34B down_proj
k = 20480
n = 7168

device = torch.device("cuda:0")

linear_class = dynamically_import_QuantLinear(use_triton=False, desc_act=False, group_size=group_size, bits=4)

linear_gptq = linear_class(
    bits=bits,
    group_size=group_size,
    infeatures=k,
    outfeatures=n,
    bias=False,
)

assert isinstance(linear_gptq, QuantLinear)

linear_gptq = linear_gptq.eval()
linear_gptq = linear_gptq.to(device)

linear_gptq = autogptq_post_init(linear_gptq, use_act_order=False)

num_runs = 60

lines = []

seqlens = [
    1,
    2,
    3,
    4,
    5,
    6,
    7,
    8,
    12,
    16,
    24,
    32,
    48,
    64,
    80,
    120,
    250,
    512,
    1024,
    2048,
    4000,
    8000,
]

print(f"in_features={k}, out_features={n}")
for query_length in seqlens:
    # batch_size, query_length, hidden_size
    inp = torch.rand(1, query_length, k, dtype=torch.float16).to(device)

    torch.cuda.empty_cache()

    # Warmup Exllama v2
    with torch.no_grad():
        res = linear_gptq(inp)

    latencies = []
    torch.cuda.synchronize()
    for _ in range(num_runs):
        start_event = torch.cuda.Event(enable_timing=True)
        end_event = torch.cuda.Event(enable_timing=True)
        torch.cuda.synchronize()
        start_event.record()

        res = linear_gptq(inp)

        end_event.record()
        torch.cuda.synchronize()

        latency_ms = start_event.elapsed_time(end_event)
        latencies.append(latency_ms)

    # print("-------")
    # print(f"Latency GPTQ Exllama v2 (query_length={query_length}): {np.mean(latencies):.3f} ms, p10={np.percentile(latencies, 10):.3f}, p90={np.percentile(latencies, 90):.3f}")

    exllamav2_mean_latency = np.mean(latencies)
    exllamav2_p10 = np.percentile(latencies, 10)
    exllamav2_p90 = np.percentile(latencies, 90)

    torch.cuda.empty_cache()

    total_seqlen = inp.shape[:-1].numel()
    if total_seqlen <= 8:
        awq_kernel = "GEMV"
        linear_awq = WQLinear_GEMV(
            w_bit=bits,
            group_size=group_size,
            in_features=k,
            out_features=n,
            bias=False,
            dev=device,
        )
    else:
        awq_kernel = "GEMM"
        linear_awq = WQLinear_GEMM(
            w_bit=bits,
            group_size=group_size,
            in_features=k,
            out_features=n,
            bias=False,
            dev=device,
        )

    # Warmup AWQ
    with torch.no_grad():
        res = linear_awq(inp)

    latencies = []
    torch.cuda.synchronize()
    for _ in range(num_runs):
        start_event = torch.cuda.Event(enable_timing=True)
        end_event = torch.cuda.Event(enable_timing=True)
        torch.cuda.synchronize()
        start_event.record()

        res = linear_awq(inp)

        end_event.record()
        torch.cuda.synchronize()

        latency_ms = start_event.elapsed_time(end_event)
        latencies.append(latency_ms)

    awq_mean_latency = np.mean(latencies)
    awq_p10 = np.percentile(latencies, 10)
    awq_p90 = np.percentile(latencies, 90)

    exllama_speedup = awq_mean_latency / exllamav2_mean_latency

    # print(f"Latency AWQ (query_length={query_length}, kernel={awq_kernel}): {np.mean(latencies):.3f} ms, p10={np.percentile(latencies, 10):.3f}, p90={np.percentile(latencies, 90):.3f}")

    line = "{},{},{},{},{},{},{},{},{},{},{}".format(
        bits,
        group_size,
        total_seqlen,
        awq_kernel,
        f"{awq_mean_latency:.3f}",
        f"{exllamav2_mean_latency:.3f}",
        f"{awq_p10:.3f}",
        f"{awq_p90:.3f}",
        f"{exllamav2_p10:.3f}",
        f"{exllamav2_p90:.3f}",
        f"{exllama_speedup:.3f}",
    )
    lines.append(line)


header = "bits, group_size, total_seqlen, awq_kernel, awq_mean_latency (ms), exllamav2_mean_latency (ms), awq_p10, awq_p90, exllamav2_p10, exllamav2_p90, exllama_speedup"

print(header)
for line in lines:
    print(line)