Skip to content

Commit 3149bdb

Browse files
author
niushengxiao
committed
feat: update calibration config files
1 parent f5c48da commit 3149bdb

10 files changed

+6412
-0
lines changed

‎test/test_kv_cache_calib_per_head_qwen2.5_14b.json‎

Lines changed: 879 additions & 0 deletions
Large diffs are not rendered by default.

‎test/test_kv_cache_calib_per_head_qwen2.5_32b.json‎

Lines changed: 1167 additions & 0 deletions
Large diffs are not rendered by default.

‎test/test_kv_cache_calib_per_head_qwen2.5_72b.json‎

Lines changed: 1455 additions & 0 deletions
Large diffs are not rendered by default.

‎test/test_kv_cache_calib_per_head_qwen3_235b.json‎

Lines changed: 1707 additions & 0 deletions
Large diffs are not rendered by default.

test/test_kv_cache_calib_per_head.json renamed to test/test_kv_cache_calib_per_head_qwen3_30b.json

File renamed without changes.

‎test/test_kv_cache_calib_per_tensor_qwen2.5_14b.json‎

Lines changed: 207 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,207 @@
1+
{
2+
"version": "1.0",
3+
"architectures": "Qwen2ForCausalLM",
4+
"quant_type": "per_tensor",
5+
"qmin": -448.0,
6+
"qmax": 448.0,
7+
"num_layers": 48,
8+
"num_head": 8,
9+
"scales_shape": [
10+
48,
11+
2
12+
],
13+
"scales": [
14+
[
15+
0.0574776828289032,
16+
0.01297433115541935
17+
],
18+
[
19+
0.02441406436264515,
20+
0.0031040736939758062
21+
],
22+
[
23+
0.0558035746216774,
24+
0.0076729916036129
25+
],
26+
[
27+
0.066964291036129,
28+
0.01492745615541935
29+
],
30+
[
31+
0.0323660746216774,
32+
0.008684431202709675
33+
],
34+
[
35+
0.0396205373108387,
36+
0.010811942629516125
37+
],
38+
[
39+
0.0387834832072258,
40+
0.012416294775903225
41+
],
42+
[
43+
0.0412946455180645,
44+
0.0143694207072258
45+
],
46+
[
47+
0.0429687537252903,
48+
0.013183594681322575
49+
],
50+
[
51+
0.0424107164144516,
52+
0.013811384327709675
53+
],
54+
[
55+
0.0412946455180645,
56+
0.0156947560608387
57+
],
58+
[
59+
0.0491071455180645,
60+
0.0143694207072258
61+
],
62+
[
63+
0.0477120541036129,
64+
0.014787946827709675
65+
],
66+
[
67+
0.0424107164144516,
68+
0.015206473879516125
69+
],
70+
[
71+
0.0407366082072258,
72+
0.014648438431322575
73+
],
74+
[
75+
0.0449218787252903,
76+
0.0164620541036129
77+
],
78+
[
79+
0.0482700914144516,
80+
0.0154854916036129
81+
],
82+
[
83+
0.0412946455180645,
84+
0.01883370615541935
85+
],
86+
[
87+
0.0477120541036129,
88+
0.0166015625
89+
],
90+
[
91+
0.0443638414144516,
92+
0.0164620541036129
93+
],
94+
[
95+
0.0491071455180645,
96+
0.01702008955180645
97+
],
98+
[
99+
0.0454799123108387,
100+
0.01981026865541935
101+
],
102+
[
103+
0.0627790242433548,
104+
0.0262276791036129
105+
],
106+
[
107+
0.0661272332072258,
108+
0.0191127248108387
109+
],
110+
[
111+
0.0471540205180645,
112+
0.01702008955180645
113+
],
114+
[
115+
0.0424107164144516,
116+
0.0181361623108387
117+
],
118+
[
119+
0.0493861623108387,
120+
0.0203683041036129
121+
],
122+
[
123+
0.0493861623108387,
124+
0.01590401865541935
125+
],
126+
[
127+
0.0435267873108387,
128+
0.037109375
129+
],
130+
[
131+
0.0700334832072258,
132+
0.0191127248108387
133+
],
134+
[
135+
0.0809151828289032,
136+
0.0193917416036129
137+
],
138+
[
139+
0.0608258955180645,
140+
0.0154854916036129
141+
],
142+
[
143+
0.0432477705180645,
144+
0.0184151791036129
145+
],
146+
[
147+
0.064453125,
148+
0.0205078125
149+
],
150+
[
151+
0.0463169664144516,
152+
0.0203683041036129
153+
],
154+
[
155+
0.0499441996216774,
156+
0.0202287957072258
157+
],
158+
[
159+
0.0532924123108387,
160+
0.02845982275903225
161+
],
162+
[
163+
0.0429687537252903,
164+
0.03055245615541935
165+
],
166+
[
167+
0.0502232164144516,
168+
0.0252511166036129
169+
],
170+
[
171+
0.0463169664144516,
172+
0.0281808041036129
173+
],
174+
[
175+
0.0452008955180645,
176+
0.02943638525903225
177+
],
178+
[
179+
0.0474330373108387,
180+
0.0325055830180645
181+
],
182+
[
183+
0.0463169664144516,
184+
0.0306919664144516
185+
],
186+
[
187+
0.0574776828289032,
188+
0.0401785746216774
189+
],
190+
[
191+
0.0496651791036129,
192+
0.0415736623108387
193+
],
194+
[
195+
0.063058041036129,
196+
0.0569196455180645
197+
],
198+
[
199+
0.0898437574505806,
200+
0.06640625
201+
],
202+
[
203+
0.0585937537252903,
204+
0.0731026828289032
205+
]
206+
]
207+
}

0 commit comments

Comments
 (0)