metadata

tags:
  - merge
  - mergekit
  - lazymergekit
  - Locutusque/llama-3-neural-chat-v1-8b
  - DiscoResearch/Llama3_DiscoLM_German_8b_v0.1_experimental
base_model:
  - Locutusque/llama-3-neural-chat-v1-8b
  - DiscoResearch/Llama3_DiscoLM_German_8b_v0.1_experimental

llama3-discolm-orca

is a merge of the following models

This was mostly a proof of concept test. GGUF 4k quants here: cstr/llama3-discolm-orca-GGUF

🧩 Configuration

LazyMergekit config:

models:
  - model: Locutusque/Llama-3-Orca-1.0-8B
    # no parameters necessary for base model
  - model: Locutusque/llama-3-neural-chat-v1-8b
    parameters:
      density: 0.60
      weight: 0.15
  - model: DiscoResearch/Llama3_DiscoLM_German_8b_v0.1_experimental
    parameters:
      density: 0.65
      weight: 0.7
merge_method: dare_ties
base_model: Locutusque/Llama-3-Orca-1.0-8B
parameters:
  int8_mask: true
dtype: bfloat16
random_seed: 0
tokenizer_source: base

💻 Usage

!pip install -qU transformers accelerate

from transformers import AutoTokenizer
import transformers
import torch

model = "cstr/llama3-discolm-orpo-t2"
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"])