metadata

base_model:
  - win10/Qwen2.5-2B-Instruct
  - win10/Qwen2.5-2B-Instruct
  - win10/Qwen2.5-2B-Instruct
  - win10/Qwen2.5-2B-Instruct
  - win10/Qwen2.5-2B-Instruct
  - win10/Qwen2.5-2B-Instruct
  - win10/Qwen2.5-2B-Instruct
  - win10/Qwen2.5-2B-Instruct
  - win10/Qwen2.5-2B-Instruct
  - win10/Qwen2.5-2B-Instruct
  - win10/Qwen2.5-2B-Instruct
  - win10/Qwen2.5-2B-Instruct
  - win10/Qwen2.5-2B-Instruct
  - win10/Qwen2.5-2B-Instruct
  - win10/Qwen2.5-2B-Instruct
  - win10/Qwen2.5-2B-Instruct
  - win10/Qwen2.5-2B-Instruct
  - win10/Qwen2.5-2B-Instruct
  - win10/Qwen2.5-2B-Instruct
  - win10/Qwen2.5-2B-Instruct
  - win10/Qwen2.5-2B-Instruct
  - win10/Qwen2.5-2B-Instruct
  - win10/Qwen2.5-2B-Instruct
  - win10/Qwen2.5-2B-Instruct
  - win10/Qwen2.5-2B-Instruct
  - win10/Qwen2.5-2B-Instruct
  - win10/Qwen2.5-2B-Instruct
  - win10/Qwen2.5-2B-Instruct
  - win10/Qwen2.5-2B-Instruct
  - win10/Qwen2.5-2B-Instruct
  - win10/Qwen2.5-2B-Instruct
  - win10/Qwen2.5-2B-Instruct
  - win10/Qwen2.5-2B-Instruct
  - win10/Qwen2.5-2B-Instruct
  - win10/Qwen2.5-2B-Instruct
  - win10/Qwen2.5-2B-Instruct
  - win10/Qwen2.5-2B-Instruct
  - win10/Qwen2.5-2B-Instruct
  - win10/Qwen2.5-2B-Instruct
  - win10/Qwen2.5-2B-Instruct
  - win10/Qwen2.5-2B-Instruct
  - win10/Qwen2.5-2B-Instruct
  - win10/Qwen2.5-2B-Instruct
  - win10/Qwen2.5-2B-Instruct
  - win10/Qwen2.5-2B-Instruct
  - win10/Qwen2.5-2B-Instruct
  - win10/Qwen2.5-2B-Instruct
  - win10/Qwen2.5-2B-Instruct
  - win10/Qwen2.5-2B-Instruct
  - win10/Qwen2.5-2B-Instruct
  - win10/Qwen2.5-2B-Instruct
tags:
  - merge
  - mergekit
  - lazymergekit
  - win10/Qwen2.5-2B-Instruct

Qwen2.5-4B-Instruct

Qwen2.5-4B-Instruct is a merge of the following models using LazyMergekit:

🧩 Configuration

dtype: bfloat16
merge_method: passthrough
slices:
- sources:
  - layer_range: [0, 2]
    model: win10/Qwen2.5-2B-Instruct
- sources:
  - layer_range: [1, 3]
    model: win10/Qwen2.5-2B-Instruct
- sources:
  - layer_range: [2, 4]
    model: win10/Qwen2.5-2B-Instruct
- sources:
  - layer_range: [3, 5]
    model: win10/Qwen2.5-2B-Instruct
- sources:
  - layer_range: [4, 6]
    model: win10/Qwen2.5-2B-Instruct
- sources:
  - layer_range: [5, 7]
    model: win10/Qwen2.5-2B-Instruct
- sources:
  - layer_range: [6, 8]
    model: win10/Qwen2.5-2B-Instruct
- sources:
  - layer_range: [7, 9]
    model: win10/Qwen2.5-2B-Instruct
- sources:
  - layer_range: [8, 10]
    model: win10/Qwen2.5-2B-Instruct
- sources:
  - layer_range: [9, 11]
    model: win10/Qwen2.5-2B-Instruct
- sources:
  - layer_range: [10, 12]
    model: win10/Qwen2.5-2B-Instruct
- sources:
  - layer_range: [11, 13]
    model: win10/Qwen2.5-2B-Instruct
- sources:
  - layer_range: [12, 14]
    model: win10/Qwen2.5-2B-Instruct
- sources:
  - layer_range: [13, 15]
    model: win10/Qwen2.5-2B-Instruct
- sources:
  - layer_range: [14, 16]
    model: win10/Qwen2.5-2B-Instruct
- sources:
  - layer_range: [15, 17]
    model: win10/Qwen2.5-2B-Instruct
- sources:
  - layer_range: [16, 18]
    model: win10/Qwen2.5-2B-Instruct
- sources:
  - layer_range: [17, 19]
    model: win10/Qwen2.5-2B-Instruct
- sources:
  - layer_range: [18, 20]
    model: win10/Qwen2.5-2B-Instruct
- sources:
  - layer_range: [19, 21]
    model: win10/Qwen2.5-2B-Instruct
- sources:
  - layer_range: [20, 22]
    model: win10/Qwen2.5-2B-Instruct
- sources:
  - layer_range: [21, 23]
    model: win10/Qwen2.5-2B-Instruct
- sources:
  - layer_range: [22, 24]
    model: win10/Qwen2.5-2B-Instruct
- sources:
  - layer_range: [23, 25]
    model: win10/Qwen2.5-2B-Instruct
- sources:
  - layer_range: [24, 26]
    model: win10/Qwen2.5-2B-Instruct
- sources:
  - layer_range: [25, 27]
    model: win10/Qwen2.5-2B-Instruct
- sources:
  - layer_range: [26, 28]
    model: win10/Qwen2.5-2B-Instruct
- sources:
  - layer_range: [27, 29]
    model: win10/Qwen2.5-2B-Instruct
- sources:
  - layer_range: [28, 30]
    model: win10/Qwen2.5-2B-Instruct
- sources:
  - layer_range: [29, 31]
    model: win10/Qwen2.5-2B-Instruct
- sources:
  - layer_range: [30, 32]
    model: win10/Qwen2.5-2B-Instruct
- sources:
  - layer_range: [31, 33]
    model: win10/Qwen2.5-2B-Instruct
- sources:
  - layer_range: [32, 34]
    model: win10/Qwen2.5-2B-Instruct
- sources:
  - layer_range: [33, 35]
    model: win10/Qwen2.5-2B-Instruct
- sources:
  - layer_range: [34, 36]
    model: win10/Qwen2.5-2B-Instruct
- sources:
  - layer_range: [35, 37]
    model: win10/Qwen2.5-2B-Instruct
- sources:
  - layer_range: [36, 38]
    model: win10/Qwen2.5-2B-Instruct
- sources:
  - layer_range: [37, 39]
    model: win10/Qwen2.5-2B-Instruct
- sources:
  - layer_range: [38, 40]
    model: win10/Qwen2.5-2B-Instruct
- sources:
  - layer_range: [39, 41]
    model: win10/Qwen2.5-2B-Instruct
- sources:
  - layer_range: [40, 42]
    model: win10/Qwen2.5-2B-Instruct
- sources:
  - layer_range: [41, 43]
    model: win10/Qwen2.5-2B-Instruct
- sources:
  - layer_range: [42, 44]
    model: win10/Qwen2.5-2B-Instruct
- sources:
  - layer_range: [43, 45]
    model: win10/Qwen2.5-2B-Instruct
- sources:
  - layer_range: [44, 46]
    model: win10/Qwen2.5-2B-Instruct
- sources:
  - layer_range: [45, 47]
    model: win10/Qwen2.5-2B-Instruct
- sources:
  - layer_range: [46, 48]
    model: win10/Qwen2.5-2B-Instruct
- sources:
  - layer_range: [47, 49]
    model: win10/Qwen2.5-2B-Instruct
- sources:
  - layer_range: [48, 50]
    model: win10/Qwen2.5-2B-Instruct
- sources:
  - layer_range: [49, 51]
    model: win10/Qwen2.5-2B-Instruct
- sources:
  - layer_range: [50, 52]
    model: win10/Qwen2.5-2B-Instruct

💻 Usage

!pip install -qU transformers accelerate

from transformers import AutoTokenizer
import transformers
import torch

model = "win10/Qwen2.5-4B-Instruct"
messages = [{"role": "user", "content": "What is a large language model?"}]

tokenizer = AutoTokenizer.from_pretrained(model)
prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
pipeline = transformers.pipeline(
    "text-generation",
    model=model,
    torch_dtype=torch.float16,
    device_map="auto",
)

outputs = pipeline(prompt, max_new_tokens=256, do_sample=True, temperature=0.7, top_k=50, top_p=0.95)
print(outputs[0]["generated_text"])