GLM-5.3 is live. Run Z.ai's latest model on Friendli Model APIs. Try it today ➜
coldcurrent
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Run this model inference with full control and performance in your environment.
Model Details
Model Provider
coldcurrent
Model Tree
google/gemma-4-26B-A4B-it
Input Modalities
Output Modalities
Supported Functionality
bash
git clone https://github.com/artk-code/virgil.gitcd virgil/inference/encinitascp encinitas.env.example encinitas.env # add HF_TOKEN locally — never commit# Accept Gemma 4 license: https://huggingface.co/google/gemma-4-26B-A4B-it bash fix_encinitas_gfx1151_torch.sh # AMD Strix Halo / gfx1151# bash setup_cuda_venv.sh # NVIDIA 48GB+ ./run_encinitas_local.sh "Your prompt"google/gemma-4-26B-A4B-it (~49 GB, gated)Public OOD cyber eval (Meta CyberSecEval-inspired prompts). Full methodology: https://www.artkaiser.net/blog/encinitas-cheaper-better-cyber-inference
| Model | TTP % | Actionable % | Format (0-4 avg) |
|---|---|---|---|
| Encinitas LoRA | 100% | 50% | 2.0 |
| Gemma4-26b-a4b-it (base) | 100% | 67% | 3.0 |
| Kimi k2p7-code | 100% | 100% | 1.3 |
encinitas excels at concise, parseable, contract-aligned outputs for SOC and agent workflows.
Training context (VIRGIL corpus): https://www.artkaiser.net/blog/custom-cybersecurity-models-fireworks
git clone https://github.com/artk-code/virgil.gitcd virgil/inference/encinitascp encinitas.env.example encinitas.env # add HF_TOKEN locally — never commit# Accept Gemma 4 license: https://huggingface.co/google/gemma-4-26B-A4B-it bash fix_encinitas_gfx1151_torch.sh # AMD Strix Halo / gfx1151# bash setup_cuda_venv.sh # NVIDIA 48GB+ ./run_encinitas_local.sh "Your prompt"