Not-For-All-Audiences

Instructions to use Tarek07/Dungeonmaster-V2.2-Expanded-LLaMa-70B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Tarek07/Dungeonmaster-V2.2-Expanded-LLaMa-70B with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="Tarek07/Dungeonmaster-V2.2-Expanded-LLaMa-70B")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("Tarek07/Dungeonmaster-V2.2-Expanded-LLaMa-70B")
model = AutoModelForCausalLM.from_pretrained("Tarek07/Dungeonmaster-V2.2-Expanded-LLaMa-70B", device_map="auto")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=40)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Inference
Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use Tarek07/Dungeonmaster-V2.2-Expanded-LLaMa-70B with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Tarek07/Dungeonmaster-V2.2-Expanded-LLaMa-70B"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Tarek07/Dungeonmaster-V2.2-Expanded-LLaMa-70B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/Tarek07/Dungeonmaster-V2.2-Expanded-LLaMa-70B

SGLang

How to use Tarek07/Dungeonmaster-V2.2-Expanded-LLaMa-70B with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "Tarek07/Dungeonmaster-V2.2-Expanded-LLaMa-70B" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Tarek07/Dungeonmaster-V2.2-Expanded-LLaMa-70B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "Tarek07/Dungeonmaster-V2.2-Expanded-LLaMa-70B" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Tarek07/Dungeonmaster-V2.2-Expanded-LLaMa-70B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use Tarek07/Dungeonmaster-V2.2-Expanded-LLaMa-70B with Docker Model Runner:
```
docker model run hf.co/Tarek07/Dungeonmaster-V2.2-Expanded-LLaMa-70B
```
Browse Quantizations to use this model in llama.cpp, Ollama, LM Studio, or any compatible app.

V2.2 of Dungeonmaster (Very good at following prompts and quite unhinged), I decided to move away from the R1 base here, because I feel it the pros dont necessarily outweigh the cons. For the V2.X series I decided to go for a custom uncensored base.

Shoutout to Thana Alt from the Beaver AI discord who thoroughly tested this model and was able to get some interesting results (spoiler below)

⚠️ Warning: Gratuitous violence

(Thana's Silly Tavern advanced formatting settings are in the models files)

The sweet spot for the important sampler settings seems to be around:

Temp: 0.8
Min P: 0.02

Dungeonmaster is meant to be specifically for creative roleplays with stakes and consequences using the following curated models:

LatitudeGames/Wayfarer-Large-70B-Llama-3.3 - A fine-tuned model specifically designed for this very application.
ArliAI/Llama-3.3-70B-ArliAI-RPMax-v1.4 - Another fine-tune trained on RP datasets.
Sao10K/70B-L3.3-mhnnn-x1 - For some extra unhinged creativity
TheDrummer/Anubis-70B-v1 - Another excellent RP fine-tune to help balance things out.
EVA-UNIT-01/EVA-LLaMA-3.33-70B-v0.1 - For it's strong descriptive writing.
SicariusSicariiStuff/Negative_LLAMA_70B - To assist with the darker undertones.
TheDrummer/Fallen-Llama-3.3-R1-70B-v1 - The secret sauce, a completely unhinged thinking model that turns things up to 11.

This is a merge of pre-trained language models created using mergekit.

Merge Details

Merge Method

This model was merged using the Linear DELLA merge method using TareksLab/L3.3-TRP-BASE-80-70B as a base.

Models Merged

The following models were included in the merge:

Configuration

The following YAML configuration was used to produce this model:

models:
  - model: TheDrummer/Fallen-Llama-3.3-R1-70B-v1
    parameters:
      weight: 0.12
      density: 0.7
  - model: ArliAI/Llama-3.3-70B-ArliAI-RPMax-v1.4
    parameters:
      weight: 0.12
      density: 0.7
  - model: EVA-UNIT-01/EVA-LLaMA-3.33-70B-v0.1
    parameters:
      weight: 0.12
      density: 0.7
  - model: TheDrummer/Anubis-70B-v1
    parameters:
      weight: 0.12
      density: 0.7
  - model: LatitudeGames/Wayfarer-Large-70B-Llama-3.3
    parameters:
      weight: 0.13
      density: 0.7
  - model: SicariusSicariiStuff/Negative_LLAMA_70B
    parameters:
      weight: 0.13
      density: 0.7
  - model: Sao10K/70B-L3.3-mhnnn-x1
    parameters:
      weight: 0.13
      density: 0.7
  - model: TareksLab/L3.3-TRP-BASE-80-70B
    parameters:
      weight: 0.13
      density: 0.7
merge_method: della_linear
base_model: TareksLab/L3.3-TRP-BASE-80-70B
parameters:
  epsilon: 0.2
  lambda: 1.1
  normalize: false
  int8_mask: true
dtype: bfloat16
chat_template: llama3
tokenizer:
 source: base