Merge to try to align the model results on the UGI leaderboard, to create the source for the modifications.
In general, just a tentative description.
Объединение, чтобы попытаться выровнять результаты модели в таблице лидеров UGI, для создания исходника для модификаций.
В целом просто пробное описание.
MT-Gen4_gemma-3-12B_flatten is a merge of the following models using
LazyMergekit:
1models:
2 - model: zelk12/26_05_2025_Test_LazyMergekit_gemma-3-12B
3 #no parameters necessary for base model
4 - model: zelk12/26_05_2025_Test_LazyMergekit_gemma-3-12B
5 parameters:
6 density: 0.438
7 weight: 0.561
8
9 - model: zelk12/MT4-gemma-3-12B
10 parameters:
11 density: 0.561
12 weight: 0.438
13
14merge_method: dare_ties
15base_model: zelk12/26_05_2025_Test_LazyMergekit_gemma-3-12B
16parameters:
17 normalize: true
18dtype: bfloat16
1!pip install -qU transformers accelerate
2
3from transformers import AutoTokenizer
4import transformers
5import torch
6
7model = "zelk12/MT-Gen4_gemma-3-12B_flatten"
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"])
