gemma-instruct-merge-test_

gemma-instruct-merge-test_

gemma-instruct-merge-test_ is a merge of the following models using LazyMergekit:

• google/gemma-2-2b-it

🧩 Configuration

- model: google/gemma-2-2b
- model: google/gemma-2-2b-it
  parameters:
    density:
    - filter: model.layers.1.self_attn.q_proj
      value: 0.00539
    - filter: model.layers.2.self_attn.q_proj
      value: 0.03843
    - filter: model.layers.6.self_attn.q_proj
      value: 0.03716
    - filter: model.layers.24.self_attn.q_proj
      value: 0.04552
    - filter: model.layers.25.self_attn.q_proj
      value: 0.03919
    - filter: model.layers.0.self_attn.k_proj
      value: 0.00592
    - filter: model.layers.2.self_attn.k_proj
      value: 0.02603
    - filter: model.layers.3.self_attn.k_proj
      value: 0.07283
    - filter: model.layers.8.self_attn.k_proj
      value: 0.08753
    - filter: model.layers.10.self_attn.k_proj
      value: 0.07783
    - filter: model.layers.23.self_attn.k_proj
      value: 0.05987
    - filter: model.layers.24.self_attn.k_proj
      value: 0.02903
    - filter: model.layers.25.self_attn.k_proj
      value: 0.08715
    - filter: model.layers.0.self_attn.v_proj
      value: 0.03025
    - filter: model.layers.2.self_attn.v_proj
      value: 0.00286
    - filter: model.layers.3.self_attn.v_proj
      value: 0.09155
    - filter: model.layers.6.self_attn.v_proj
      value: 0.06811
    - filter: model.layers.7.self_attn.v_proj
      value: 0.01334
    - filter: model.layers.10.self_attn.v_proj
      value: 0.04
    - filter: model.layers.13.self_attn.v_proj
      value: 0.09347
    - filter: model.layers.24.self_attn.v_proj
      value: 0.07956
    - filter: model.layers.0.self_attn.o_proj
      value: 0.03265
    - filter: model.layers.2.self_attn.o_proj
      value: 0.06134
    - filter: model.layers.4.self_attn.o_proj
      value: 0.07924
    - filter: model.layers.6.self_attn.o_proj
      value: 0.09982
    - filter: model.layers.7.self_attn.o_proj
      value: 0.02826
    - filter: model.layers.8.self_attn.o_proj
      value: 0.03906
    - filter: model.layers.19.self_attn.o_proj
      value: 0.09507
    - filter: model.layers.23.self_attn.o_proj
      value: 0.00282
    - filter: model.layers.24.self_attn.o_proj
      value: 0.09864
    - filter: model.layers.25.self_attn.o_proj
      value: 0.00961
    - filter: model.layers.1.mlp.gate_proj
      value: 0.08775
    - filter: model.layers.2.mlp.gate_proj
      value: 0.0001
    - filter: model.layers.6.mlp.gate_proj
      value: 0.06577
    - filter: model.layers.12.mlp.gate_proj
      value: 0.02651
    - filter: model.layers.13.mlp.gate_proj
      value: 0.04687
    - filter: model.layers.15.mlp.gate_proj
      value: 0.03147
    - filter: model.layers.16.mlp.gate_proj
      value: 0.05726
    - filter: model.layers.17.mlp.gate_proj
      value: 0.04511
    - filter: model.layers.23.mlp.gate_proj
      value: 0.08641
    - filter: model.layers.1.mlp.up_proj
      value: 0.06887
    - filter: model.layers.6.mlp.up_proj
      value: 0.07411
    - filter: model.layers.7.mlp.up_proj
      value: 0.05424
    - filter: model.layers.12.mlp.up_proj
      value: 0.08044
    - filter: model.layers.13.mlp.up_proj
      value: 0.0021
    - filter: model.layers.14.mlp.up_proj
      value: 0.26389
    - filter: model.layers.15.mlp.up_proj
      value: 0.06886
    - filter: model.layers.23.mlp.up_proj
      value: 0.02931
    - filter: model.layers.0.mlp.down_proj
      value: 0.06756
    - filter: model.layers.1.mlp.down_proj
      value: 0.03746
    - filter: model.layers.2.mlp.down_proj
      value: 0.09104
    - filter: model.layers.3.mlp.down_proj
      value: 0.06643
    - filter: model.layers.4.mlp.down_proj
      value: 0.05003
    - filter: model.layers.5.mlp.down_proj
      value: 0.0406
    - filter: model.layers.6.mlp.down_proj
      value: 0.01609
    - filter: model.layers.7.mlp.down_proj
      value: 0.09629
    - filter: model.layers.8.mlp.down_proj
      value: 0.08912
    - filter: model.layers.10.mlp.down_proj
      value: 0.04635
    - filter: model.layers.11.mlp.down_proj
      value: 0.0099
    - filter: model.layers.12.mlp.down_proj
      value: 0.03487
    - filter: model.layers.13.mlp.down_proj
      value: 0.04977
    - filter: model.layers.14.mlp.down_proj
      value: 0.00393
    - filter: model.layers.15.mlp.down_proj
      value: 0.00748
    - filter: model.layers.16.mlp.down_proj
      value: 0.06696
    - filter: model.layers.17.mlp.down_proj
      value: 0.02067
    - filter: model.layers.19.mlp.down_proj
      value: 0.009
    - filter: model.layers.20.mlp.down_proj
      value: 0.0215
    - filter: model.layers.21.mlp.down_proj
      value: 0.04196
    - filter: model.layers.22.mlp.down_proj
      value: 0.06326
    - filter: model.layers.25.mlp.down_proj
      value: 0.04905
    weight:
    - value: 1
merge_method: ties
base_model: google/gemma-2-2b
parameters:
  normalize: true
  int8_mask: true
dtype: bfloat16
tokenizer_source: union

💻 Usage

!pip install -qU transformers accelerate

from transformers import AutoTokenizer
import transformers
import torch

model = "choprahetarth/gemma-instruct-merge-test_"
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"])