AbyssSynth-12B

She holds the weight of a galaxy, a fragile infinity encased within a sphere of shimmering light.
Darkness swirls beneath endless stars, vast and unfathomable, yet tempered by her calm grasp.
Not a force of chaos, but of silent order, the abyss contained, a cosmos in delicate balance, where creation and oblivion meet.

🔧 Recommended Sampling Settings:

Temperature: 0.75 to 1.25
Min P: 0.035
Context Length: Stable at 12k tokens, with possible support for extended contexts

💬 Prompt Format

Supports ChatML style messages. Example:

<|im_start|>user
Your question here.
<|im_end|>
<|im_start|>assistant

AbyssSynth-12B is a merge of the following models using LazyMergekit:

🧩 Configuration

merge_method: ties
base_model: yamatazen/LorablatedStock-12B
models:
  - model: yamatazen/LorablatedStock-12B
    parameters:
      weight: 0.65
      density: 1.0
  - model: yamatazen/EtherealAurora-12B-v2
    parameters:
      weight: 0.35
      density: 1.0

parameters:
  normalize: false
  int8_mask: false
dtype: bfloat16

layer_parameters:
  - filter: "attn"
    sources:
      - model: yamatazen/LorablatedStock-12B
        weight: 0.6
      - model: yamatazen/EtherealAurora-12B-v2
        weight: 0.4

  - filter: "mlp"
    sources:
      - model: yamatazen/LorablatedStock-12B
        weight: 0.55
      - model: yamatazen/EtherealAurora-12B-v2
        weight: 0.45

  - filter: "embed_tokens"
    sources:
      - model: yamatazen/LorablatedStock-12B
        weight: 1.0
      - model: yamatazen/EtherealAurora-12B-v2
        weight: 0.0

  - filter: "layer_norm"
    sources:
      - model: yamatazen/LorablatedStock-12B
        weight: 0.7
      - model: yamatazen/EtherealAurora-12B-v2
        weight: 0.3

💻 Usage

!pip install -qU transformers accelerate

from transformers import AutoTokenizer
import transformers
import torch

model = "Marcjoni/AbyssSynth-12B-12B"
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=1, top_k=0, top_p=1)
print(outputs[0]["generated_text"])