--- license: apache-2.0 base_model: mlabonne/Qwen3-4B-abliterated base_model_relation: quantized pipeline_tag: text-generation tags: [gguf, llama.cpp, ollama, lm-studio, on-device, offline, private, abliterated, uncensored, qwen3] --- # TC Mini 4B — GGUF (Ollama · LM Studio · llama.cpp) Private AI that runs fully on your own computer. Best for: desktops and laptops (GPU or Apple Silicon). Part of **TC Mini**: also runs in your phone's browser — https://truthprotectors.github.io/TC-Mini/app/ · code https://github.com/truthprotectors/TC-Mini · all sizes https://huggingface.co/truthcoder/TC-Mini ```bash ollama run hf.co/truthcoder/TC-Mini-4B-GGUF:Q4_K_M llama-cli -hf truthcoder/TC-Mini-4B-GGUF:Q4_K_M ``` LM Studio: search `truthcoder/TC-Mini-4B-GGUF`. Add `/no_think` to a message to skip the thinking step. | File | Quant | Size | |---|---|---| | `TC-Mini-4B-F16.gguf` | F16 | 8.05 GB | | `TC-Mini-4B-Q4_K_M.gguf` | Q4_K_M | 2.50 GB | | `TC-Mini-4B-Q8_0.gguf` | Q8_0 | 4.28 GB | Q4_K_M = smallest, recommended · Q8_0 = near-lossless · F16 = full precision (for re-quantizing). **Lineage (honest):** [mlabonne/Qwen3-4B-abliterated](https://huggingface.co/mlabonne/Qwen3-4B-abliterated) (Qwen3 with refusal directions removed by mlabonne), converted with llama.cpp `convert_hf_to_gguf.py` and quantized with `llama-quantize`. Not fine-tuned by us. 0/8 refusals on our 8-prompt probe (browser build of the same weights). You are responsible for how you use it. SHA-256: ``` 7e91d02b780f1b071bc062d37d659a3c9ac6b90407b80ad8bfd15ac01b839f17 TC-Mini-4B-F16.gguf e2c1a8a8b9a54e915a7138ecbc0c1590ea692e57350054d16d104c445417c6d1 TC-Mini-4B-Q4_K_M.gguf a0833fd945174f78b37c2ff805df819a10bacdef70da280f525b6141f362c10e TC-Mini-4B-Q8_0.gguf ``` License: Apache-2.0 (Qwen3).