Skip to content

GitLab

Switch branch/tag
Find file Normal viewHistoryPermalink
ggml_test.go 4.96 KB
Newer Older
Michael Yang's avatar
next ollama runner (#7913)  
Michael Yang 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 
package ggml

import (
	"maps"
	"slices"
	"strings"
	"testing"

	"github.com/google/go-cmp/cmp"
)

func TestTensorLayers(t *testing.T) {
	tensors := make(map[string]*Tensor)
	for _, name := range []string{
		"token_embd.weight",
		"blk.0.attn_k.weight",
		"blk.0.attn_output.weight",
		"blk.0.attn_q.weight",
		"blk.0.attn_v.weight",
		"blk.0.attn_norm.weight",
		"blk.0.ffn_down.weight",
		"blk.0.ffn_gate.weight",
		"blk.0.ffn_up.weight",
		"blk.0.ffn_norm.weight",
		"output_norm.weight",
		"mm.0.bias",
		"mm.0.weight",
		"v.blk.0.attn_k.weight",
		"v.blk.0.attn_output.weight",
		"v.blk.0.attn_q.weight",
		"v.blk.0.attn_v.weight",
		"v.blk.0.attn_norm.weight",
		"v.blk.0.ffn_down.weight",
		"v.blk.0.ffn_gate.weight",
		"v.blk.0.ffn_up.weight",
		"v.blk.0.ffn_norm.weight",
		"v.patch_embd.weight",
		"v.position_embd.gate",
		"v.position_embd.weight",
	} {
		tensors[name] = &Tensor{Name: name}
	}

	cases := []struct {
		name  string
		items []*Tensor
		want  map[string]Layer
	}{
		{
			name: "text",
			items: slices.Collect(func(yield func(*Tensor) bool) {
				for k, v := range tensors {
					if !strings.HasPrefix(k, "mm.") && !strings.HasPrefix(k, "v.") {
						if !yield(v) {
							return
						}
					}
				}
			}),
			want: map[string]Layer{
				"blk.0": {
					"attn_k.weight":      tensors["blk.0.attn_k.weight"],
					"attn_q.weight":      tensors["blk.0.attn_q.weight"],
					"attn_v.weight":      tensors["blk.0.attn_v.weight"],
					"attn_output.weight": tensors["blk.0.attn_output.weight"],
					"attn_norm.weight":   tensors["blk.0.attn_norm.weight"],
					"ffn_down.weight":    tensors["blk.0.ffn_down.weight"],
					"ffn_gate.weight":    tensors["blk.0.ffn_gate.weight"],
					"ffn_up.weight":      tensors["blk.0.ffn_up.weight"],
					"ffn_norm.weight":    tensors["blk.0.ffn_norm.weight"],
				},
				"token_embd":  {"weight": tensors["token_embd.weight"]},
				"output_norm": {"weight": tensors["output_norm.weight"]},
			},
		},
		{
			name: "vision",
			items: slices.Collect(func(yield func(*Tensor) bool) {
				for k, v := range tensors {
					if strings.HasPrefix(k, "mm.") || strings.HasPrefix(k, "v.") {
						if !yield(v) {
							return
						}
					}
				}
			}),
			want: map[string]Layer{
				"mm.0": {
					"bias":   tensors["mm.0.bias"],
					"weight": tensors["mm.0.weight"],
				},
				"v.blk.0": {
					"attn_k.weight":      tensors["v.blk.0.attn_k.weight"],
					"attn_q.weight":      tensors["v.blk.0.attn_q.weight"],
					"attn_v.weight":      tensors["v.blk.0.attn_v.weight"],
					"attn_output.weight": tensors["v.blk.0.attn_output.weight"],
					"attn_norm.weight":   tensors["v.blk.0.attn_norm.weight"],
					"ffn_down.weight":    tensors["v.blk.0.ffn_down.weight"],
					"ffn_gate.weight":    tensors["v.blk.0.ffn_gate.weight"],
					"ffn_up.weight":      tensors["v.blk.0.ffn_up.weight"],
					"ffn_norm.weight":    tensors["v.blk.0.ffn_norm.weight"],
				},
				"v": {
					"patch_embd.weight":    tensors["v.patch_embd.weight"],
					"position_embd.gate":   tensors["v.position_embd.gate"],
					"position_embd.weight": tensors["v.position_embd.weight"],
				},
			},
		},
		{
			name:  "vision and text",
			items: slices.Collect(maps.Values(tensors)),
			want: map[string]Layer{
				"blk.0": {
					"attn_k.weight":      tensors["blk.0.attn_k.weight"],
					"attn_q.weight":      tensors["blk.0.attn_q.weight"],
					"attn_v.weight":      tensors["blk.0.attn_v.weight"],
					"attn_output.weight": tensors["blk.0.attn_output.weight"],
					"attn_norm.weight":   tensors["blk.0.attn_norm.weight"],
					"ffn_down.weight":    tensors["blk.0.ffn_down.weight"],
					"ffn_gate.weight":    tensors["blk.0.ffn_gate.weight"],
					"ffn_up.weight":      tensors["blk.0.ffn_up.weight"],
					"ffn_norm.weight":    tensors["blk.0.ffn_norm.weight"],
				},
				"token_embd":  {"weight": tensors["token_embd.weight"]},
				"output_norm": {"weight": tensors["output_norm.weight"]},
				"mm.0": {
					"bias":   tensors["mm.0.bias"],
					"weight": tensors["mm.0.weight"],
				},
				"v.blk.0": {
					"attn_k.weight":      tensors["v.blk.0.attn_k.weight"],
					"attn_q.weight":      tensors["v.blk.0.attn_q.weight"],
					"attn_v.weight":      tensors["v.blk.0.attn_v.weight"],
					"attn_output.weight": tensors["v.blk.0.attn_output.weight"],
					"attn_norm.weight":   tensors["v.blk.0.attn_norm.weight"],
					"ffn_down.weight":    tensors["v.blk.0.ffn_down.weight"],
					"ffn_gate.weight":    tensors["v.blk.0.ffn_gate.weight"],
					"ffn_up.weight":      tensors["v.blk.0.ffn_up.weight"],
					"ffn_norm.weight":    tensors["v.blk.0.ffn_norm.weight"],
				},
				"v": {
					"patch_embd.weight":    tensors["v.patch_embd.weight"],
					"position_embd.gate":   tensors["v.position_embd.gate"],
					"position_embd.weight": tensors["v.position_embd.weight"],
				},
			},
		},
	}

	for _, tt := range cases {
		t.Run(tt.name, func(t *testing.T) {
			got := Tensors{items: tt.items}.GroupLayers()
			if diff := cmp.Diff(got, tt.want); diff != "" {
				t.Errorf("unexpected layers (-got +want):\n%s", diff)
			}
		})
	}
}