djinn is a merge of the following models using
LazyMergekit:
1merge_method: linear
2parameters:
3 weight: 1.0
4slices:
5 - sources:
6 - model: CultriX/NeuralTrix-7B-dpo # embed_tokens comes along with the ride with whatever is the first layer
7 layer_range: [0, 1]
8 - model: paulml/DPOB-INMTOB-7B # add dummy second model with 0 weight so tokenizer-based merge routine is invoked for embed_tokens
9 layer_range: [0, 1]
10 parameters:
11 weight: 0
12 - sources:
13 - model: cognitivecomputations/dolphin-2.1-mistral-7b
14 layer_range: [0, 8]
15 - sources:
16 - model: bardsai/jaskier-7b-dpo-v5.6
17 layer_range: [8, 16]
18 - sources:
19 - model: paulml/OGNO-7B
20 layer_range: [16, 24]
21 - sources:
22 - model: argilla/distilabeled-OpenHermes-2.5-Mistral-7B
23 layer_range: [24, 31]
24 - sources: # same as above, but for lm_head with the last layer
25 - model: CultriX/NeuralTrix-7B-dpo
26 layer_range: [31, 32]
27 - model: paulml/DPOB-INMTOB-7B
28 layer_range: [31, 32]
29 parameters:
30 weight: 0
31dtype: float16
32tokenizer_source: model:cognitivecomputations/dolphin-2.1-mistral-7b
33
1!pip install -qU transformers accelerate
2
3from transformers import AutoTokenizer
4import transformers
5import torch
6
7model = "mayacinka/djinn"
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"])
