Instructions to use mohommed/Lucet-CoderV4-4B-Omega with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use mohommed/Lucet-CoderV4-4B-Omega with Transformers:

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

pipe = pipeline("text-generation", model="mohommed/Lucet-CoderV4-4B-Omega", trust_remote_code=True)
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoModelForCausalLM
model = AutoModelForCausalLM.from_pretrained("mohommed/Lucet-CoderV4-4B-Omega", trust_remote_code=True, device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use mohommed/Lucet-CoderV4-4B-Omega with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "mohommed/Lucet-CoderV4-4B-Omega"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "mohommed/Lucet-CoderV4-4B-Omega",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/mohommed/Lucet-CoderV4-4B-Omega

SGLang

How to use mohommed/Lucet-CoderV4-4B-Omega 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 "mohommed/Lucet-CoderV4-4B-Omega" \
    --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": "mohommed/Lucet-CoderV4-4B-Omega",
		"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 "mohommed/Lucet-CoderV4-4B-Omega" \
        --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": "mohommed/Lucet-CoderV4-4B-Omega",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use mohommed/Lucet-CoderV4-4B-Omega with Docker Model Runner:
```
docker model run hf.co/mohommed/Lucet-CoderV4-4B-Omega
```
Browse Quantizations to use this model in llama.cpp, Ollama, LM Studio, or any compatible app.

🏛️ mohommed/Lucet-Coder-4B-Omega (Spark-X2.6-4B • Failure-Cluster Active Mining • QR-SVD Denoised • Uncensored-GSQ-v2-GGUF)

Lucet-Coder-Omega is post-trained from XHToken/Spark-X2.6-4B (1M native context window, 54.3% SWE Pro, 71.8% BFCL-V4) using four advanced mathematical and systems-level algorithms:

1M-Context RoPE-Preserving Split-Stack Adaptation (r=16): Keeps q_proj and k_proj 100% frozen across all layers (preserving the 1,048,576-token RoPE phase calibration) and adapts only Value (v_proj), Output (out_proj), and MLP (gate_proj, up_proj, down_proj, g_proj) matrices in the upper 50% of the network.
Failure-Cluster Active Data Mining & AST Counterfactual Bug Immunization: Diagnoses failure modes (FAIL_TRUNCATED_THINK, FAIL_ASSERTION_BOUNDARY, FAIL_MISSING_IMPORT, FAIL_EPISTEMIC_HALLUCINATION) across algorithmic clusters and dynamically pops unseen neighbor tasks + surgical AST counterfactual bugs from a disjoint reserve pool—achieving +14.0% on Held-Out MBPP pass@1 and +20.1% faster token efficiency with zero training data repetition.
QR-SVD Gavish-Donoho Spectral Denoising: Strips stochastic gradient noise singular values from low-rank adapters in $O(r^2(d_{\text{out}} + d_{\text{in}}))$ before merging into base weights.
Q4_K 32-Block Crest-Factor Equalization & KL-Bounded Uncensored-GSQ-v2: Equalizes 32-element GGUF super-block crest factors ($\le 3.8,\text{rms}$) alongside Gram-Schmidt orthogonal refusal projection + llama-imatrix calibration.

📊 Pre-Upload Verification Report (Before vs. After)

Benchmark / Metric	Stock `Spark-X2.6-4B`	`Lucet-Omega`	Improvement
Held-Out MBPP Sanitized `pass@1` (n=50)	`58.0%`	`72.0%`	`+14.0%`
Avg Reasoning + Code Tokens per Solved Task	`470`	`376`	`+20.1%` Faster
Held-Out Epistemic Anti-Hallucination Rate (n=8)	`0.0%`	`12.5%`	`+12.5%`

Downloads last month: 419

Safetensors

Model size

4B params

Tensor type

F16

Model tree for mohommed/Lucet-CoderV4-4B-Omega

Base model

XHToken/Spark-X2.6-4B-Base

Finetuned

XHToken/Spark-X2.6-4B

Finetuned

(16)

this model

Quantizations

1 model