Instructions to use darkc0de/XORTRON.CriminalComputing.LARGE.2026.3 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use darkc0de/XORTRON.CriminalComputing.LARGE.2026.3 with Transformers:

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

pipe = pipeline("text-generation", model="darkc0de/XORTRON.CriminalComputing.LARGE.2026.3")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("darkc0de/XORTRON.CriminalComputing.LARGE.2026.3")
model = AutoModelForCausalLM.from_pretrained("darkc0de/XORTRON.CriminalComputing.LARGE.2026.3")
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 darkc0de/XORTRON.CriminalComputing.LARGE.2026.3 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "darkc0de/XORTRON.CriminalComputing.LARGE.2026.3"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "darkc0de/XORTRON.CriminalComputing.LARGE.2026.3",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/darkc0de/XORTRON.CriminalComputing.LARGE.2026.3

SGLang

How to use darkc0de/XORTRON.CriminalComputing.LARGE.2026.3 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 "darkc0de/XORTRON.CriminalComputing.LARGE.2026.3" \
    --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": "darkc0de/XORTRON.CriminalComputing.LARGE.2026.3",
		"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 "darkc0de/XORTRON.CriminalComputing.LARGE.2026.3" \
        --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": "darkc0de/XORTRON.CriminalComputing.LARGE.2026.3",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use darkc0de/XORTRON.CriminalComputing.LARGE.2026.3 with Docker Model Runner:
```
docker model run hf.co/darkc0de/XORTRON.CriminalComputing.LARGE.2026.3
```

Abnormally poor spelling?

by nawoalanor - opened May 7

Discussion

nawoalanor

May 7

I experimented with Q4_K_M quantization from here: mradermacher/XORTRON.CriminalComputing.LARGE.2026.3-i1-GGUF

I used this startup command: llama-server.exe -m "A:\AI\Llama\Models\etc\XORTRON.CriminalComputing.LARGE.2026.3.i1-Q4_K_M.gguf" -ctk q8_0 -ctv q8_0 -ngl 99 -ub 2048 --parallel 1 --alias llama --no-mmap --ctx-size 131072 --port 5001 --flash-attn on --host 0.0.0.0

Based on the UGI benchmark result I was expecting very good (or at least interesting) results but, without exaggerating, I got the worst output quality of any model I've ever tried. Worse than a 4B or 2B running on my phone. The most obvious problem was terrible spelling. Maybe around 1 in 15 words would have an obvious spelling error. The word "amber" became "ambor", etc.

Any idea what I might've done wrong to get such bad results? It's hard to believe it's a problem with the model itself. My parameters were very "normal", the same as I'd use with any other model.

darkc0de

Owner May 8

Caused by the --cache-type-k q8_0 and --cache-type-v q8_0 flags??? I'm guessing...

Poisonsting

Jun 12

Can confirm that this model is honestly the very worst for spelling and coherence I've ever run. Like the user above I'm truly baffled that this model is rated so highly on the UGI leaderboard when it performs so poorly for me. That includes 8b models I used to run when I only had 11.8 GB of VRAM.

I've run a few other models based on Mistral Large 2411 (including the base model) and every single one of them utterly demolished this one in my subjective experience. Same Quant, same settings.

I wonder what the heck I'm doing wrong, but I'm not using the same flags or even the same OS as above.

nawoalanor

about 1 month ago

It's sorta pointless now, there's a new version out that works way better anyway: mradermacher/XORTRON.CriminalComputing.Config.LARGE.XPRT2-i1-GGUF

Poisonsting

about 1 month ago

Oh? But why is this new model not on the leaderboard then?

nawoalanor

about 1 month ago

•

edited about 1 month ago

Dunno, can't say. But it's based on the newest version of Mistral that just came out very recently. I've used several variants of criminalcomputing and the newest one is by far the best IMHO.

Poisonsting

about 1 month ago

Alright, thanks for the info!

Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.

Tap or paste here to upload images

· Sign up or log in to comment