Instructions to use groxaxo/MagiSeek-V1.1 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use groxaxo/MagiSeek-V1.1 with Transformers:

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

pipe = pipeline("text-generation", model="groxaxo/MagiSeek-V1.1")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("groxaxo/MagiSeek-V1.1")
model = AutoModelForCausalLM.from_pretrained("groxaxo/MagiSeek-V1.1", 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 groxaxo/MagiSeek-V1.1 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "groxaxo/MagiSeek-V1.1"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "groxaxo/MagiSeek-V1.1",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/groxaxo/MagiSeek-V1.1

SGLang

How to use groxaxo/MagiSeek-V1.1 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 "groxaxo/MagiSeek-V1.1" \
    --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": "groxaxo/MagiSeek-V1.1",
		"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 "groxaxo/MagiSeek-V1.1" \
        --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": "groxaxo/MagiSeek-V1.1",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use groxaxo/MagiSeek-V1.1 with Docker Model Runner:
```
docker model run hf.co/groxaxo/MagiSeek-V1.1
```
Browse Quantizations to use this model in llama.cpp, Ollama, LM Studio, or any compatible app.

MagiSeek V1.1

Overview

MagiSeek-V1.1 is a model checkpoint packaged for compatible Hugging Face runtimes, published by groxaxo. It is intended for open-source evaluation, reproducible experimentation, and compatible local or hosted inference workflows. The wording below is deliberately limited to what can be verified from this repository's metadata and artifacts.

At a glance

Field	Details
Format	Transformers
Source / base	`groxaxo/MagiSeek-Pro-V1`
Intended task	`text-generation`
License	`apache-2.0`

What is included

*.safetensors (10 files)
config.json
generation_config.json
tokenizer.json
tokenizer_config.json
chat_template.jinja
Additional configuration, tokenizer, processor, or shard files (17 visible artifacts total)

Quick start

Getting started

Start with the upstream library named in the repository metadata and keep all configuration, tokenizer, processor, and weight files together. This repository is an artifact release, so the source project remains the authoritative reference for task-specific loading code.

Compatibility and responsible use

Use a runtime that explicitly supports this format, architecture, and modality.
Keep configuration, tokenizer, processor, projection, and weight files from the same revision together.
Review the source model card and license before redistribution or deployment.
Hardware needs depend on parameter count, context length, cache precision, quantization, and concurrency.
Report reproducible issues with the runtime version, hardware, launch command, and a minimal example.

Generated outputs may be inaccurate or unsuitable for a given use case. Users are responsible for testing behavior, applying appropriate safeguards, and complying with applicable licenses and laws.

A field-tested continuation of MagiSeek-Pro-V1

MagiSeek V1.1 is a carefully merged continuation of MagiSeek-Pro-V1, shaped around the kind of work that happens after the demo: reading a messy request, keeping the thread of intent, reasoning through a change, and producing a useful next action.

The story behind this release is deliberately simple. MagiSeek-Pro-V1 was the strong starting voice. We then placed it in a long, demanding practice loop of cleaned coding-agent and tool-aware conversations. The model learned through a rank-64 rsLoRA continuation, and this release captures the first verified checkpoint of that journey: step 800, before the run was intentionally stopped.

That makes V1.1 a snapshot with character, not a claim of finality. It is the moment where the new habits are visible while the original MagiSeek style is still close to the surface. Use it as an experimental open model, a base for evaluation, or a reference point for future training.

What is included

Full merged model weights; no adapter is required.
bfloat16 safetensors, split into 10 shards.
Native MagiSeek tokenizer and chat template.
Merge source manifest in MERGE_SOURCE.json.

Training snapshot

Base: groxaxo/MagiSeek-Pro-V1
Base revision: c7a0a437ca849bf4e3286fe5aae7951185974651
Training method: QLoRA / rsLoRA, rank 64, alpha 64
Target modules: attention and MLP projections
Context length: 8,192 tokens
Objective: assistant-only supervised fine-tuning
Checkpoint: global step 800, approximately 0.187 epoch of the planned run
Hardware: three NVIDIA RTX 3090 GPUs

The run was stopped before its planned 10 epochs and before a final adapter was written. No benchmark numbers are claimed for this checkpoint yet. Please treat it as experimental and evaluate it for your own workload.

Quick start

from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "groxaxo/MagiSeek-V1.1"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype="auto",
    device_map="auto",
)

messages = [
    {"role": "user", "content": "Explain what you can help me build."},
]
inputs = tokenizer.apply_chat_template(
    messages, add_generation_prompt=True, return_tensors="pt"
).to(model.device)
outputs = model.generate(inputs, max_new_tokens=256, temperature=0.7)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

Intended use

MagiSeek V1.1 is intended for research, experimentation, coding assistance, structured tool-use prototypes, and conversational applications where a model should turn context into a practical next step.

Do not use it as an unsupervised authority for medical, legal, financial, security, or other high-impact decisions. Generated code and tool arguments must be reviewed and sandboxed before execution. As an early checkpoint, it may be less stable than the base model on tasks outside its training emphasis.

Limitations and honest status

This is not a benchmark-certified release. The training run ended early, and the model card does not claim that V1.1 is better than MagiSeek-Pro-V1 on every task. It may overfit to coding-agent patterns, imitate incorrect tool calls, hallucinate facts, or produce confident but unsafe instructions. Always test with held-out prompts and keep human approval in the loop for external actions.

The training corpus was cleaned and structurally filtered locally. This card does not reproduce private source conversations or claim that every generated example is error-free. Users and downstream distributors remain responsible for their own data, safety, and license review.

License

The base model is published under Apache-2.0, and this merged release follows that license. See the base model card for upstream attribution and terms.