I’m Md Shakib Molla, known as mo-shakib on Ollama. I publish Gemma 4-based local AI models, including a general-purpose 8B variant and focused writing and career assistants. I published gemma4-e4b-uncensored on Ollama for people who want to experiment with a more direct Gemma 4 8B model on their own machine. In roughly its first two months, the Ollama listing passed 6,000 downloads (6,111 when I checked on October 15, 2026).
The model is configured for minimal built-in refusal behavior. That means it may answer prompts more directly, but it does not make every answer correct, suitable, or safe to use. I treat it as a tool for experimentation and review its outputs before relying on them.
Which Gemma 4 models does mo-shakib publish on Ollama?
My Ollama profile brings together three Gemma 4-based models for different local AI workflows:
- gemma4-e4b-uncensored is an 8B Q4_K_M variant with minimal built-in refusal behavior and a 131,072-token context window.
- ClearWriter is a Gemma 4 E2B writing assistant for rewriting, proofreading, grammar correction, tone adjustment, and email drafting.
- ResumePilot is a Gemma 4 E2B assistant for resume review, job matching, cover letters, and interview preparation.
These are published model variants and assistants. Their Ollama pages describe how to run and use them; this article does not claim that each one was separately fine-tuned or retrained.
What is gemma4-e4b-uncensored?
This is my Gemma 4 8B variant packaged for local inference with Ollama. The published q4_k_m tag uses Q4_K_M quantization, which helps balance model quality and memory use. The Ollama model listing reports a 9.6 GB download, a 128K context window, and text and image inputs. Its README specifies a context length of 131,072 tokens and an embedding size of 2,560.
| Specification | Published value |
|---|---|
| Base architecture | Gemma 4 |
| Parameters | 8B |
| Ollama tag | mo-shakib/gemma4-e4b-uncensored:q4_k_m |
| Quantization | Q4_K_M |
| Context length | 131,072 tokens |
| Default temperature / top-K / top-P | 1.0 / 64 / 0.95 |
The README also describes reasoning, tool calling, vision understanding, and audio understanding. The current Ollama tag explicitly labels the downloadable model as Text, Image, so check audio support in your particular runtime and workflow before depending on it.
Run the model locally with Ollama
With Ollama installed, open a terminal and run:
ollama run mo-shakib/gemma4-e4b-uncensored:q4_k_mOllama downloads the model on the first run. After that, you can chat in the terminal or call its local API. For example:
curl http://localhost:11434/api/chat \
-d '{
"model": "mo-shakib/gemma4-e4b-uncensored:q4_k_m",
"messages": [{"role": "user", "content": "Explain a practical use for a 128K context window."}],
"stream": false
}'The 128K window makes longer conversations and document workflows possible, but the usable length and response speed still depend on the machine and Ollama configuration. The published model size is a helpful starting point when checking available storage and memory.
Where it fits in an AI engineering workflow
For AI engineers, a local model can be useful for prototyping prompts, testing tool-based workflows, exploring vision tasks, and comparing responses across model configurations. This release is aimed at general chat, creative writing, coding help, research, brainstorming, roleplay, and long-document analysis.
If you work on AI training or evaluation, it is important to separate those activities from simply running a published model. The Ollama page documents this model’s published configuration and capabilities; it does not provide a training recipe or benchmark results. A sensible evaluation is to create your own task set, record the model tag and generation settings, and check outputs for accuracy, instruction following, and unwanted content. That gives you evidence about how it performs for your use case without assuming that fewer refusals mean better quality.
Minimal refusal behavior also calls for careful use. The model may produce offensive, explicit, biased, or inaccurate content. Verify factual claims and keep qualified professionals involved in medical, legal, financial, safety-critical, and other high-risk decisions.
Try gemma4-e4b-uncensored
If you want to experiment with the Gemma 4 8B variant described here, start with the gemma4-e4b-uncensored model page and check its current tags and README before running it.