Neo_7b-merge19 is a merge of the following models using
LazyMergekit:
1# Define the slices for the model merging process
2slices:
3 - sources:
4 # First part: merge layer 0 with layer 3
5 - model: DewEfresh/neo_7b
6 layer_range: [0, 1]
7 - model: m-a-p/neo_7b
8 layer_range: [3, 4]
9 - sources:
10 # Second part: merge layer 1 with layer 3
11 - model: DewEfresh/neo_7b
12 layer_range: [1, 2]
13 - model: m-a-p/neo_7b
14 layer_range: [3, 4]
15 - sources:
16 # Third part: merge layer 2 with layer 3
17 - model: DewEfresh/neo_7b
18 layer_range: [2, 3]
19 - model: m-a-p/neo_7b
20 layer_range: [3, 4]
21 - sources:
22 # Fourth part: merge layer 4 with layer 7
23 - model: DewEfresh/neo_7b
24 layer_range: [4, 5]
25 - model: m-a-p/neo_7b
26 layer_range: [7, 8]
27 - sources:
28 # Fifth part: merge layer 5 with layer 7
29 - model: DewEfresh/neo_7b
30 layer_range: [5, 6]
31 - model: m-a-p/neo_7b
32 layer_range: [7, 8]
33 - sources:
34 # Sixth part: merge layer 6 with layer 7
35 - model: DewEfresh/neo_7b
36 layer_range: [6, 7]
37 - model: m-a-p/neo_7b
38 layer_range: [7, 8]
39 - sources:
40 # Seventh part: merge layer 8 with layer 11
41 - model: DewEfresh/neo_7b
42 layer_range: [8, 9]
43 - model: m-a-p/neo_7b
44 layer_range: [11, 12]
45 - sources:
46 # Eighth part: merge layer 9 with layer 11
47 - model: DewEfresh/neo_7b
48 layer_range: [9, 10]
49 - model: m-a-p/neo_7b
50 layer_range: [11, 12]
51 - sources:
52 # Ninth part: merge layer 10 with layer 11
53 - model: DewEfresh/neo_7b
54 layer_range: [10, 11]
55 - model: m-a-p/neo_7b
56 layer_range: [11, 12]
57 - sources:
58 # Tenth part: merge layer 12 with layer 15
59 - model: DewEfresh/neo_7b
60 layer_range: [12, 13]
61 - model: m-a-p/neo_7b
62 layer_range: [15, 16]
63 - sources:
64 # Eleventh part: merge layer 13 with layer 15
65 - model: DewEfresh/neo_7b
66 layer_range: [13, 14]
67 - model: m-a-p/neo_7b
68 layer_range: [15, 16]
69 - sources:
70 # Twelfth part: merge layer 14 with layer 15
71 - model: DewEfresh/neo_7b
72 layer_range: [14, 15]
73 - model: m-a-p/neo_7b
74 layer_range: [15, 16]
75 - sources:
76 # Thirteenth part: merge layer 16 with layer 19
77 - model: DewEfresh/neo_7b
78 layer_range: [16, 17]
79 - model: m-a-p/neo_7b
80 layer_range: [19, 20]
81 - sources:
82 # Fourteenth part: merge layer 17 with layer 19
83 - model: DewEfresh/neo_7b
84 layer_range: [17, 18]
85 - model: m-a-p/neo_7b
86 layer_range: [19, 20]
87 - sources:
88 # Fifteenth part: merge layer 18 with layer 19
89 - model: DewEfresh/neo_7b
90 layer_range: [18, 19]
91 - model: m-a-p/neo_7b
92 layer_range: [19, 20]
93 - sources:
94 # Sixteenth part: merge layer 20 with layer 23
95 - model: DewEfresh/neo_7b
96 layer_range: [20, 21]
97 - model: m-a-p/neo_7b
98 layer_range: [23, 24]
99 - sources:
100 # Seventeenth part: merge layer 21 with layer 23
101 - model: DewEfresh/neo_7b
102 layer_range: [21, 22]
103 - model: m-a-p/neo_7b
104 layer_range: [23, 24]
105 - sources:
106 # Eighteenth part: merge layer 22 with layer 23
107 - model: DewEfresh/neo_7b
108 layer_range: [22, 23]
109 - model: m-a-p/neo_7b
110 layer_range: [23, 24]
111 - sources:
112 # Nineteenth part: merge layer 24 with layer 27
113 - model: DewEfresh/neo_7b
114 layer_range: [24, 25]
115 - model: m-a-p/neo_7b
116 layer_range: [26, 27]
117 - sources:
118 # Twentieth part: merge layer 25 with layer 27
119 - model: DewEfresh/neo_7b
120 layer_range: [25, 26]
121 - model: m-a-p/neo_7b
122 layer_range: [26, 27]
123 - sources:
124 # Twenty-first part: merge layer 26 with layer 27
125 - model: DewEfresh/neo_7b
126 layer_range: [26, 27]
127 - model: m-a-p/neo_7b
128 layer_range: [26, 27]
129# Specify the merging method for the slices
130merge_method: slerp
131base_model: DewEfresh/neo_7b
132parameters:
133 t: 1 # Set global interpolation value to 33.33%
134dtype: bfloat16
135
1!pip install -qU transformers accelerate
2
3from transformers import AutoTokenizer
4import transformers
5import torch
6
7model = "DewEfresh/Neo_7b-merge19"
8messages = [{"role": "user", "content": "What is a large language model?"}]
9
10tokenizer = AutoTokenizer.from_pretrained(model)
11prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
12pipeline = transformers.pipeline(
13 "text-generation",
14 model=model,
15 torch_dtype=torch.float16,
16 device_map="auto",
17)
18
19outputs = pipeline(prompt, max_new_tokens=256, do_sample=True, temperature=0.7, top_k=50, top_p=0.95)
20print(outputs[0]["generated_text"])
