llama-3-sqrt-crocodile-v0.0A

🧩 Configuration-moe

base_model: llama-3-sqrt-crocodile-v0.0A/Uninstruct-Uncensored
gate_mode: hidden
dtype: bfloat16
experts:
  - source_model: llama-3-sqrt-crocodile-v0.0A/sqrt-talker
    positive_prompts:
      - "Uncensored, creative, configurable, adapative"
  - source_model: llama-3-sqrt-crocodile-v0.0A/the-operator 
    positive_prompts:
      - "Function calling"
      - "Good at structured tasks"
      - "Programmatic instruction following"

🧩 Configuration-mega

models:
  - model: Orenguteng/Lexi-Llama-3-8B-Uncensored
    parameters:
      weight: [0.2, 0.3, 0.4, 0.6]
    layer_range: [0, 32]
  - model: NousResearch/Meta-Llama-3-8B
    parameters:
      weight: [0.6, 0.2, 0.2, 0.1]
    layer_range: [0, 32]
  - model: NousResearch/Meta-Llama-3-8B-Instruct
    parameters:
      weight: [0.2, 0.3, 0.85, 0.3]
    layer_range: [0, 32]
merge_method: dare_linear
base_model: NousResearch/Meta-Llama-3-8B-Instruct
dtype: bfloat16
name: Uninstruct-Uncensored
---
models:
  - model: cognitivecomputations/dolphin-2.9-llama3-8b
    parameters:
      weight: [0.25, 0.4, 0.35, 0.35]
      density: [0.3, 0.45, 0.2, 0.6]
    layer_range: [0, 32]
  - model: NousResearch/Meta-Llama-3-8B
    parameters: 
      weight: [0.15, 0.25, 0.05, 0]
      density: [0.2, 0.3, 0.4, 0.1]
  - model: Undi95/Llama-3-Unholy-8B
    parameters:
      weight: [0.4, 0.25, 0.45, 0.35]
      density: [0.2, 0.15, 1.5, 0.1]
    layer_range: [0, 32]
  - model: Uninstruct-Uncensored
    parameters:
      weight: [0.3, 0.1, 0.25, 0.3]
      density: [0.3, 0.15, 2.5, 0.2]
    layer_range: [0, 32]
merge_method: dare_ties
base_model: Uninstruct-Uncensored
dtype: bfloat16
name: augmented-dolphin-hap
---
models:
  - model: vicgalle/Configurable-Llama-3-8B-v0.3
    parameters:
      weight: [0.5, 0.3, 0.1]
  - model: hiieu/Meta-Llama-3-8B-Instruct-function-calling-json-mode
    parameters:
      weight: 0.5
  - model: Trelis/Meta-Llama-3-8B-Instruct-function-calling
    parameters:
      weight: 0.3
    layer_range: [0, 32]
  - model: Rookie/Llama-3-8B-Instruct-Chinese 
    parameters:
      weight: 0.2
    layer_range: [0, 32]
  - model: Uninstruct-Uncensored
    parameters:
      weight: [0.7, 0.4, 0.25, 0.1]
    layer_range: [0, 32]
merge_method: model_stock
base_model: Uninstruct-Uncensored
dtype: bfloat16
name: the-operator
---
models:
  - model: vicgalle/Configurable-Llama-3-8B-v0.3
    parameters:
      weight: 0.7
  - model: hiieu/Meta-Llama-3-8B-Instruct-function-calling-json-mode
    parameters:
      weight: 0.1
  - model: Trelis/Meta-Llama-3-8B-Instruct-function-calling
    parameters:
      weight: 0.03
    layer_range: [0, 32]
  - model: Rookie/Llama-3-8B-Instruct-Chinese
    parameters:
      weight: 0.07
    layer_range: [0, 32]
  - model: Uninstruct-Uncensored
    parameters:
      weight: 0.1
    layer_range: [0, 32]
merge_method: model_stock
base_model: Uninstruct-Uncensored
dtype: bfloat16
name: her-calculator
---
models:
  - model: her-calculator
    parameters:
      density: 0.7 # density gradient
      weight: [0.7, 0.5, 0.1, 0.8]
  - model: augmented-dolphin-hap
    parameters:
      weight: 0.7
merge_method: slerp
base_model: her-calculator
parameters:
  t:
    - filter: self_attn
      value: [0, 0.5, 0.3, 0.7, 1]
    - filter: mlp
      value: [1, 0.5, 0.7, 0.3, 0]
    - value: 0.5 # fallback for rest of tensors
dtype: float16
name: sqrt-talker

💻 Usage

!pip install -qU transformers bitsandbytes accelerate

from transformers import AutoTokenizer
import transformers
import torch

model = "Nhoodie/llama-3-sqrt-crocodile-v0.0A"

tokenizer = AutoTokenizer.from_pretrained(model)
pipeline = transformers.pipeline(
    "text-generation",
    model=model,
    model_kwargs={"torch_dtype": torch.float16, "load_in_4bit": True},
)

messages = [{"role": "user", "content": "Explain what a Mixture of Experts is in less than 100 words."}]
prompt = pipeline.tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
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"])