Views
No views yet
1from transformers import AutoTokenizer, AutoModelForCausalLM
2import torch
3
4model_id = "umiyuki/Llama-3-Umievo-itr014-Shizuko-8b"
5
6tokenizer = AutoTokenizer.from_pretrained(model_id)
7model = AutoModelForCausalLM.from_pretrained(
8 model_id,
9 torch_dtype=torch.bfloat16,
10 device_map="auto",
11)
12
13messages = [
14 {"role": "system", "content": "You must answer all responses in Japanese.あなたは役に立つ誠実な日本人のアシスタントです。あなたは全ての回答に日本語で答えなければならない。"},
15 {"role": "user", "content": "二人の少女が終末世界を旅する物語を書いてください。"},
16]
17
18input_ids = tokenizer.apply_chat_template(
19 messages,
20 add_generation_prompt=True,
21 return_tensors="pt"
22).to(model.device)
23
24terminators = [
25 tokenizer.eos_token_id,
26 tokenizer.convert_tokens_to_ids("<|eot_id|>")
27]
28
29outputs = model.generate(
30 input_ids,
31 max_new_tokens=256,
32 eos_token_id=terminators,
33 do_sample=True,
34 temperature=0.6,
35 top_p=0.9,
36)
37response = outputs[0][input_ids.shape[-1]:]
38print(tokenizer.decode(response, skip_special_tokens=True))1base_model: meta-llama/Meta-Llama-3-8B-Instruct
2dtype: bfloat16
3merge_method: linear
4parameters:
5 int8_mask: 1.0
6 normalize: 1.0
7slices:
8- sources:
9 - layer_range: [0, 4]
10 model: lightblue/suzume-llama-3-8B-multilingual
11 parameters:
12 weight: 0.4149739730274144
13 - layer_range: [0, 4]
14 model: meta-llama/Meta-Llama-3-8B-Instruct
15 parameters:
16 weight: 0.6781276007090549
17 - layer_range: [0, 4]
18 model: aixsatoshi/Llama-3-youko-8b-instruct-chatvector
19 parameters:
20 weight: 0.34616999273932425
21 - layer_range: [0, 4]
22 model: shisa-ai/shisa-v1-llama3-8b
23 parameters:
24 weight: 1.3720042419649354
25- sources:
26 - layer_range: [4, 8]
27 model: lightblue/suzume-llama-3-8B-multilingual
28 parameters:
29 weight: 0.07652836818139683
30 - layer_range: [4, 8]
31 model: meta-llama/Meta-Llama-3-8B-Instruct
32 parameters:
33 weight: 1.234379009181979
34 - layer_range: [4, 8]
35 model: aixsatoshi/Llama-3-youko-8b-instruct-chatvector
36 parameters:
37 weight: 1.0146729889059811
38 - layer_range: [4, 8]
39 model: shisa-ai/shisa-v1-llama3-8b
40 parameters:
41 weight: 0.5811532109389872
42- sources:
43 - layer_range: [8, 12]
44 model: lightblue/suzume-llama-3-8B-multilingual
45 parameters:
46 weight: 0.5551700273906248
47 - layer_range: [8, 12]
48 model: meta-llama/Meta-Llama-3-8B-Instruct
49 parameters:
50 weight: 0.7418501521559635
51 - layer_range: [8, 12]
52 model: aixsatoshi/Llama-3-youko-8b-instruct-chatvector
53 parameters:
54 weight: 1.442504375594772
55 - layer_range: [8, 12]
56 model: shisa-ai/shisa-v1-llama3-8b
57 parameters:
58 weight: 0.6475631873316974
59- sources:
60 - layer_range: [12, 16]
61 model: lightblue/suzume-llama-3-8B-multilingual
62 parameters:
63 weight: 0.4227647782669271
64 - layer_range: [12, 16]
65 model: meta-llama/Meta-Llama-3-8B-Instruct
66 parameters:
67 weight: 1.2969869792284983
68 - layer_range: [12, 16]
69 model: aixsatoshi/Llama-3-youko-8b-instruct-chatvector
70 parameters:
71 weight: 0.7818773805802817
72 - layer_range: [12, 16]
73 model: shisa-ai/shisa-v1-llama3-8b
74 parameters:
75 weight: 0.8007371182560976
76- sources:
77 - layer_range: [16, 20]
78 model: lightblue/suzume-llama-3-8B-multilingual
79 parameters:
80 weight: 0.10979010874744283
81 - layer_range: [16, 20]
82 model: meta-llama/Meta-Llama-3-8B-Instruct
83 parameters:
84 weight: 0.19009547180175693
85 - layer_range: [16, 20]
86 model: aixsatoshi/Llama-3-youko-8b-instruct-chatvector
87 parameters:
88 weight: 0.6064294349661996
89 - layer_range: [16, 20]
90 model: shisa-ai/shisa-v1-llama3-8b
91 parameters:
92 weight: 0.7630087852386511
93- sources:
94 - layer_range: [20, 24]
95 model: lightblue/suzume-llama-3-8B-multilingual
96 parameters:
97 weight: 0.219671192433268
98 - layer_range: [20, 24]
99 model: meta-llama/Meta-Llama-3-8B-Instruct
100 parameters:
101 weight: 0.6303503074132494
102 - layer_range: [20, 24]
103 model: aixsatoshi/Llama-3-youko-8b-instruct-chatvector
104 parameters:
105 weight: 0.46265431269055757
106 - layer_range: [20, 24]
107 model: shisa-ai/shisa-v1-llama3-8b
108 parameters:
109 weight: 1.4662350856064592
110- sources:
111 - layer_range: [24, 28]
112 model: lightblue/suzume-llama-3-8B-multilingual
113 parameters:
114 weight: 0.1400550380200451
115 - layer_range: [24, 28]
116 model: meta-llama/Meta-Llama-3-8B-Instruct
117 parameters:
118 weight: 1.031570135674053
119 - layer_range: [24, 28]
120 model: aixsatoshi/Llama-3-youko-8b-instruct-chatvector
121 parameters:
122 weight: 0.5760956440228217
123 - layer_range: [24, 28]
124 model: shisa-ai/shisa-v1-llama3-8b
125 parameters:
126 weight: 1.5264012437679564
127- sources:
128 - layer_range: [28, 32]
129 model: lightblue/suzume-llama-3-8B-multilingual
130 parameters:
131 weight: 1.2311282964552015
132 - layer_range: [28, 32]
133 model: meta-llama/Meta-Llama-3-8B-Instruct
134 parameters:
135 weight: 0.43811773040605967
136 - layer_range: [28, 32]
137 model: aixsatoshi/Llama-3-youko-8b-instruct-chatvector
138 parameters:
139 weight: 0.5150682019605872
140 - layer_range: [28, 32]
141 model: shisa-ai/shisa-v1-llama3-8b
142 parameters:
143 weight: 0.342193342214983