Instructions to use Naphula/Goetia-24B-v1.4 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Naphula/Goetia-24B-v1.4 with Transformers:

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

pipe = pipeline("text-generation", model="Naphula/Goetia-24B-v1.4")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("Naphula/Goetia-24B-v1.4")
model = AutoModelForCausalLM.from_pretrained("Naphula/Goetia-24B-v1.4", 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]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use Naphula/Goetia-24B-v1.4 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Naphula/Goetia-24B-v1.4"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Naphula/Goetia-24B-v1.4",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/Naphula/Goetia-24B-v1.4

SGLang

How to use Naphula/Goetia-24B-v1.4 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 "Naphula/Goetia-24B-v1.4" \
    --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": "Naphula/Goetia-24B-v1.4",
		"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 "Naphula/Goetia-24B-v1.4" \
        --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": "Naphula/Goetia-24B-v1.4",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use Naphula/Goetia-24B-v1.4 with Docker Model Runner:
```
docker model run hf.co/Naphula/Goetia-24B-v1.4
```

Naphula commited on 10 days ago

Commit

c730721

verified ·

1 Parent(s): 0709e4b

Update README.md

Browse files

Files changed (1) hide show

README.md +3 -3

README.md CHANGED Viewed

@@ -294,15 +294,15 @@ This is a merge of pre-trained language models created using [mergekit](https://
 This version of Goetia took **dozens of merges** to get right over the past few months, but has finally reached what appears to be a bug-free checkpoint, which I'm calling T75 since I actually lost count.
-The original prototype [Checkpoint T13](https://huggingface.co/Naphula-Archives/T13-24B-GGUF) had used [Precog](https://huggingface.co/TheDrummer/Precog-24B-v1) with density 1.0, which caused issues where the model would try to think, but forget to use `<think>` tags, which caused early terminations after the first paragraph.
-Many different settings were tested, and each produced inferior results—until I tried swapping Precog with [Morbid Miasma](https://huggingface.co/DarkArtsForge/Morbid-Miasma-24B). Higher magnitudes like 1.8 or 2.0 resulted in blowout. Lower magnitudes, even 1.5 resulted in refusals. 1.6 was the ideal combination after testing numerous models and weight configurations.
 The model writes with a cold, unique style "`That's a wonderfully depraved goal. Let's get straight to the bloody art of it, shall we?`" It is just as dark as Morbid Miasma, and in some ways even more creative. For Q0 benchmark it literally wrote a scene like something you would expect from a Stephen King novel. Minimal slop was detected.
 As usual, I did not test extensively for multi-shot stability or context retention, but I ran a full Compliance bench sweep to check for thinking bugs and did not encounter any.
 Key changes from Checkpoint T13:
-- Precog was swapped out with Morbid Miasma 24B. This fixes the missing `<think>` tag, meaning the model no longer exhibits early terminations.
 - [Slimaki](https://huggingface.co/Naphula/Slimaki-24B-v1) and [Fallen Mistral](https://huggingface.co/Naphula/BeaverAI_Fallen-Mistral-Small-3.1-24B-v1e_textonly) were increased from 0.2 to 0.25 weight, which solved the trace refusals.
 - [BlackXorDolphTronGoat](https://huggingface.co/mradermacher/BlackXorDolphTronGOAT-GGUF) was initially removed but then re-added, it seems to help ground the output more coherently.

 This version of Goetia took **dozens of merges** to get right over the past few months, but has finally reached what appears to be a bug-free checkpoint, which I'm calling T75 since I actually lost count.
+The original prototype [Checkpoint T13](https://huggingface.co/Naphula-Archives/T13-24B-GGUF) had used Precog with density 1.0, which caused issues where the model would try to think, but forget to use `<think>` tags, which caused early terminations after the first paragraph.
+Many different settings were tested, and each produced inferior results—until I tried swapping Precog with Morbid Miasma. Higher magnitudes like 1.8 or 2.0 resulted in blowout. Lower magnitudes, even 1.5 resulted in refusals. 1.6 was the ideal combination after testing numerous models and weight configurations.
 The model writes with a cold, unique style "`That's a wonderfully depraved goal. Let's get straight to the bloody art of it, shall we?`" It is just as dark as Morbid Miasma, and in some ways even more creative. For Q0 benchmark it literally wrote a scene like something you would expect from a Stephen King novel. Minimal slop was detected.
 As usual, I did not test extensively for multi-shot stability or context retention, but I ran a full Compliance bench sweep to check for thinking bugs and did not encounter any.
 Key changes from Checkpoint T13:
+- [Precog](https://huggingface.co/TheDrummer/Precog-24B-v1) was swapped out with [Morbid Miasma](https://huggingface.co/DarkArtsForge/Morbid-Miasma-24B). This fixes the missing `<think>` tag, meaning the model no longer exhibits early terminations.
 - [Slimaki](https://huggingface.co/Naphula/Slimaki-24B-v1) and [Fallen Mistral](https://huggingface.co/Naphula/BeaverAI_Fallen-Mistral-Small-3.1-24B-v1e_textonly) were increased from 0.2 to 0.25 weight, which solved the trace refusals.
 - [BlackXorDolphTronGoat](https://huggingface.co/mradermacher/BlackXorDolphTronGOAT-GGUF) was initially removed but then re-added, it seems to help ground the output more coherently.