Neo_7b-merge3 is a merge of the following models using
LazyMergekit:
1slices:
2 # Group 1
3 - sources:
4 - model: m-a-p/neo_7b
5 layer_range: [0, 0]
6 - model: m-a-p/neo_7b
7 layer_range: [3, 3]
8 - sources:
9 - model: m-a-p/neo_7b
10 layer_range: [1, 1]
11 - model: m-a-p/neo_7b
12 layer_range: [3, 3]
13 - sources:
14 - model: m-a-p/neo_7b
15 layer_range: [2, 2]
16 - model: m-a-p/neo_7b
17 layer_range: [3, 3]
18 # Group 2
19 - sources:
20 - model: m-a-p/neo_7b
21 layer_range: [4, 4]
22 - model: m-a-p/neo_7b
23 layer_range: [7, 7]
24 - sources:
25 - model: m-a-p/neo_7b
26 layer_range: [5, 5]
27 - model: m-a-p/neo_7b
28 layer_range: [7, 7]
29 - sources:
30 - model: m-a-p/neo_7b
31 layer_range: [6, 6]
32 - model: m-a-p/neo_7b
33 layer_range: [7, 7]
34 # Group 3
35 - sources:
36 - model: m-a-p/neo_7b
37 layer_range: [8, 8]
38 - model: m-a-p/neo_7b
39 layer_range: [11, 11]
40 - sources:
41 - model: m-a-p/neo_7b
42 layer_range: [9, 9]
43 - model: m-a-p/neo_7b
44 layer_range: [11, 11]
45 - sources:
46 - model: m-a-p/neo_7b
47 layer_range: [10, 10]
48 - model: m-a-p/neo_7b
49 layer_range: [11, 11]
50 # Group 4
51 - sources:
52 - model: m-a-p/neo_7b
53 layer_range: [12, 12]
54 - model: m-a-p/neo_7b
55 layer_range: [15, 15]
56 - sources:
57 - model: m-a-p/neo_7b
58 layer_range: [13, 13]
59 - model: m-a-p/neo_7b
60 layer_range: [15, 15]
61 - sources:
62 - model: m-a-p/neo_7b
63 layer_range: [14, 14]
64 - model: m-a-p/neo_7b
65 layer_range: [15, 15]
66 # Group 5
67 - sources:
68 - model: m-a-p/neo_7b
69 layer_range: [16, 16]
70 - model: m-a-p/neo_7b
71 layer_range: [19, 19]
72 - sources:
73 - model: m-a-p/neo_7b
74 layer_range: [17, 17]
75 - model: m-a-p/neo_7b
76 layer_range: [19, 19]
77 - sources:
78 - model: m-a-p/neo_7b
79 layer_range: [18, 18]
80 - model: m-a-p/neo_7b
81 layer_range: [19, 19]
82 # Group 6
83 - sources:
84 - model: m-a-p/neo_7b
85 layer_range: [20, 20]
86 - model: m-a-p/neo_7b
87 layer_range: [23, 23]
88 - sources:
89 - model: m-a-p/neo_7b
90 layer_range: [21, 21]
91 - model: m-a-p/neo_7b
92 layer_range: [23, 23]
93 - sources:
94 - model: m-a-p/neo_7b
95 layer_range: [22, 22]
96 - model: m-a-p/neo_7b
97 layer_range: [23, 23]
98 # Group 7 (last group)
99 - sources:
100 - model: m-a-p/neo_7b
101 layer_range: [24, 24]
102 - model: m-a-p/neo_7b
103 layer_range: [27, 27]
104 - sources:
105 - model: m-a-p/neo_7b
106 layer_range: [25, 25]
107 - model: m-a-p/neo_7b
108 layer_range: [27, 27]
109 - sources:
110 - model: m-a-p/neo_7b
111 layer_range: [26, 26]
112 - model: m-a-p/neo_7b
113 layer_range: [27, 27]
114merge_method: slerp
115base_model: m-a-p/neo_7b
116parameters:
117 t: 0.3333 # Apply 1/3 of the 4th layer to each of the previous 3 layers
118dtype: bfloat16
119output_path: ./merged_redistributed_neo_7b
1!pip install -qU transformers accelerate
2
3from transformers import AutoTokenizer
4import transformers
5import torch
6
7model = "DewEfresh/Neo_7b-merge3"
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"])
