Instructions to use Retreatcost/Irix-mpf-stock with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Retreatcost/Irix-mpf-stock with Transformers:

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

pipe = pipeline("text-generation", model="Retreatcost/Irix-mpf-stock")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("Retreatcost/Irix-mpf-stock")
model = AutoModelForCausalLM.from_pretrained("Retreatcost/Irix-mpf-stock")
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

vLLM

How to use Retreatcost/Irix-mpf-stock with vLLM:

Install from pip and serve model

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

Use Docker

docker model run hf.co/Retreatcost/Irix-mpf-stock

SGLang

How to use Retreatcost/Irix-mpf-stock 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 "Retreatcost/Irix-mpf-stock" \
    --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": "Retreatcost/Irix-mpf-stock",
		"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 "Retreatcost/Irix-mpf-stock" \
        --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": "Retreatcost/Irix-mpf-stock",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use Retreatcost/Irix-mpf-stock with Docker Model Runner:
```
docker model run hf.co/Retreatcost/Irix-mpf-stock
```
Browse Quantizations to use this model in llama.cpp, Ollama, LM Studio, or any compatible app.

Irix-mpf-stock

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

An experimental merge to improve long-form writing capabilities of Irix-12B-Model_Stock.

I merged LatitudeGames/Muse-12B and Trappu/Nemo-Picaro-12B using karcher with higher tolerance and max iterations - took quite a lot of time
I've created 3 derivative models using arcee_fusion - they were hand-picked from hundreds of similar merges that performed best on 3 tests:
- Adventure prompt
- Factual consistency + reasoning
- Long-form creative writing
Created model_stock using these high-performing merges.

PROS
- Shows more variety than base model and sometimes subverts expectations
- More creativity and granularity in world-building
- Better at complex, layered and ambitious scenarios
CONS
- Slightly faster, often using dialogue to advance the scene (Original Irix feels like a novelist, this model feels like a screen writer)
- Less intricate prose, writing feels modern, grounded and grim (might be a plus in some cases)
- Less introspective, pays more attention to external details (Pays more attention to sensory details rather than how character feels)

TL;DR;

Original Irix: Produces highly readable, satisfying, and complete stories.

This model: Produces stories that feel like the thrilling beginning of a much larger work.

Oh, and I am planning to use this model as an output layer for next KansenSakura update

Merge Details

Merge Method

This model was merged using the Model Stock merge method using DreadPoor/Irix-12B-Model_Stock as a base.

Models Merged

The following models were included in the merge:

Configuration

The following YAML configuration was used to produce this model:

merge_method: karcher
models:
  - model: LatitudeGames/Muse-12B
  - model: Trappu/Nemo-Picaro-12B
parameters:
  max_iter: 10000
  tol: 1e-9
dtype: bfloat16
tokenizer_source: LatitudeGames/Muse-12B

merge_method: arcee_fusion
base_model: DreadPoor/Irix-12B-Model_Stock
models:
  - model: DreadPoor/Irix-12B-Model_Stock
  - model: ./musepicaro
dtype: bfloat16
tokenizer_source: DreadPoor/Irix-12B-Model_Stock

merge_method: model_stock
base_model: DreadPoor/Irix-12B-Model_Stock
models:
  - model: ./irix_fusion3
  - model: ./irix_fusion2
  - model: ./irix_fusion
parameters:
  normalize: false
  t: 0.75
dtype: bfloat16