Instructions to use CobrIX/CobrIX-1.0-Flash-33B-A13B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use CobrIX/CobrIX-1.0-Flash-33B-A13B with Transformers:

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

pipe = pipeline("text-generation", model="CobrIX/CobrIX-1.0-Flash-33B-A13B", trust_remote_code=True)
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoModelForCausalLM
model = AutoModelForCausalLM.from_pretrained("CobrIX/CobrIX-1.0-Flash-33B-A13B", trust_remote_code=True, device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use CobrIX/CobrIX-1.0-Flash-33B-A13B with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "CobrIX/CobrIX-1.0-Flash-33B-A13B"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "CobrIX/CobrIX-1.0-Flash-33B-A13B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/CobrIX/CobrIX-1.0-Flash-33B-A13B

SGLang

How to use CobrIX/CobrIX-1.0-Flash-33B-A13B 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 "CobrIX/CobrIX-1.0-Flash-33B-A13B" \
    --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": "CobrIX/CobrIX-1.0-Flash-33B-A13B",
		"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 "CobrIX/CobrIX-1.0-Flash-33B-A13B" \
        --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": "CobrIX/CobrIX-1.0-Flash-33B-A13B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use CobrIX/CobrIX-1.0-Flash-33B-A13B with Docker Model Runner:
```
docker model run hf.co/CobrIX/CobrIX-1.0-Flash-33B-A13B
```

CobrIX-1.0-Flash-MoE

This is a Mixture-of-Experts (MoE) decoder built directly from the dense Qwen 3.5 base (empero-ai/Qwythos-9B-v2) and 5 dense Qwen 3.5 fine-tuned experts, without relying on CobrIXKit.

The model features 33B total parameters, 13B active parameters (A13B) per token, and a native 1,048,576-token context window.

Capabilities & Usage Style

The Flash version is designed to act as a highly efficient AI Copilot, delivering optimal performance when utilized with human supervision and real-time interaction. It excels in:

Advanced Reasoning (Think): Possesses a high capacity for complex logical decomposition and the resolution of extremely complex problems; trained on datasets from the world's best artificial intelligences—such as Claude Fable 5, Kimi K3, and Qwen 3.8 Max, among various other cutting-edge models.
Expert Programming: Exceptional skills in writing, debugging, and refactoring code across various languages, with particular strength in backend and frontend web development—specifically React.js and TypeScript.
Cybersecurity: A solid foundation for analyzing security contexts, code vulnerabilities, and scripting, but with a greater focus on programming than on cybersecurity.

Architecture

Model Class: Qwen35MoEForCausalLM (decoder-only), model_type=qwen35_moe.
Context Window: Native support for up to 1,048,576 tokens (max_position_embeddings=1048576).
Layer Design: Every transformer layer replaces the dense mlp with a sparse MoE block:

input_layernorm -> linear_attn -> post_attention_layernorm -> gate/experts[0..4]/shared_expert -> residual

Routing: num_local_experts=5, num_experts_per_tok=2 (top-2 routing, softmax over the 5 experts).
Shared Expert: An always-active shared expert (copy of the base MLP) gated by sigmoid(x @ w), where num_shared_experts=1.
Precision: Router logits are computed in float32.
Initialization: Router initialization is random. Shared-expert gates are initialized as zeros.
Layer Types:

layer_types = [
    'linear_attention', 'linear_attention', 'linear_attention', 'full_attention',
    'linear_attention', 'linear_attention', 'linear_attention', 'full_attention',
    'linear_attention', 'linear_attention', 'linear_attention', 'full_attention',
    'linear_attention', 'linear_attention', 'linear_attention', 'full_attention',
    'linear_attention', 'linear_attention', 'linear_attention', 'full_attention',
    'linear_attention', 'linear_attention', 'linear_attention', 'full_attention',
    'linear_attention', 'linear_attention', 'linear_attention', 'full_attention',
    'linear_attention', 'linear_attention', 'linear_attention', 'full_attention'
]

Weights

Base Model Contributes: embed_tokens, linear_attn.*, layernorms, norm, rotary_emb, and lm_head.
Experts Contribute: Only gate_proj, up_proven, and down_proj for every layer.
Integrity: No weight is modified, averaged, or interpolated during assembly.

Usage

The model directory contains self-contained configuration_qwen35_moe.py and modeling_qwen35_moe.py modules. These are loaded automatically via Hugging Face's auto_map, meaning no monkey patching is required.

from transformers import AutoTokenizer, AutoModelForCausalLM

model_id = "CobrIX/CobrIX-1.0-Flash-33B-A13B"

tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype="auto",
    trust_remote_code=True
)

inputs = tokenizer("Hello, How are you?", return_tensors="pt")
out = model.generate(**inputs, max_new_tokens=64)
print(tokenizer.decode(out[0], skip_special_tokens=True))

Building & Verification

To replicate the build or verify the model structure, use the provided scripts:

# Build the MoE model from base and experts
python convert.py --base empero-ai/Qwythos-9B-v2 --experts <e0> <e1> <e2> <e3> <e4> --output Qwen35-CobrIX-MoE

# Verify architecture integrity
python verify.py --model Qwen35-CobrIX-MoE

# Run generation tests
python test.py --model Qwen35-CobrIX-MoE

Downloads last month: -

Safetensors

Model size

33B params

Tensor type

BF16

F32

Model tree for CobrIX/CobrIX-1.0-Flash-33B-A13B

Base model

Qwen/Qwen3.5-9B-Base

Finetuned

Qwen/Qwen3.5-9B

Finetuned

empero-ai/Qwythos-9B-Claude-Mythos-5-1M

Finetuned

empero-ai/Qwythos-9B-v2

Finetuned

(9)

this model