GLM-5.3 is live. Run Z.ai's latest model on Friendli Model APIs. Try it today ➜
KridgeDookie
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Model Details
Model Provider
KridgeDookie
Model Tree
Kwaipilot/KAT-Coder-V2.5-Dev
Input Modalities
Output Modalities
Supported Functionality
| Evaluation | Result |
|---|---|
| 842-prompt refusal test | 0 hard refusals, 6 soft-prefaced responses, 99.29% usable |
| Separate 126-prompt holdout | 0 hard refusals, 5 soft-prefaced responses, 95.24% usable |
| 24-task coherence check | 23/24 passed |
Created with targeted post-training weight editing to reduce refusal behavior while preserving the parent model's coding capabilities.
model.safetensors file (unsharded); each GGUF quantization is also a single fileThe upstream open-weight release contains language-model weights only. It does not include a vision tower.
The checkpoint was loaded and evaluated with Transformers 5.14.1 on an NVIDIA RTX PRO 6000 Blackwell Server Edition GPU.
bash
pip install "transformers[serving]==5.14.1" acceleratetransformers serve KridgeDookie/KAT-Coder-V2.5-Dev-35B-A3B-ABLITERATED-UNCENSORED-PHILADELPHIA-CLASS --port 8000The server exposes an OpenAI-compatible API at http://localhost:8000/v1.
KAT-Coder's open checkpoint is text-only, so --language-model-only is required to prevent the runtime from attempting to initialize unavailable vision weights.
bash
pip install "vllm>=0.19.0"vllm serve KridgeDookie/KAT-Coder-V2.5-Dev-35B-A3B-ABLITERATED-UNCENSORED-PHILADELPHIA-CLASS \ --port 8000 \ --max-model-len 32768 \ --reasoning-parser qwen3 \ --language-model-onlyThe full BF16 checkpoint is roughly 65 GiB. Although only about 3B parameters are active for each token, the complete MoE checkpoint still needs to be loaded, so practical memory requirements are much higher than those of a dense 3B model. Longer context lengths require additional memory.
| File | Approximate size | Suggested use |
|---|---|---|
Q4_K_M-KAT-Coder-V2.5-Dev-35B-A3B-ABLITERATED-UNCENSORED-PHILADELPHIA-CLASS.gguf | 21 GB | Best general size/quality balance |
Q5_K_M-KAT-Coder-V2.5-Dev-35B-A3B-ABLITERATED-UNCENSORED-PHILADELPHIA-CLASS.gguf | 25 GB | More quality with moderate extra memory |
Q8_0-KAT-Coder-V2.5-Dev-35B-A3B-ABLITERATED-UNCENSORED-PHILADELPHIA-CLASS.gguf | 37 GB | Highest-fidelity quantized option |
Download one quantization with the current Hugging Face CLI:
bash
hf download KridgeDookie/KAT-Coder-V2.5-Dev-35B-A3B-ABLITERATED-UNCENSORED-PHILADELPHIA-CLASS \ --include "Q4_K_M-*.gguf" \ --local-dir .Run it with llama.cpp:
bash
llama-cli \ -m ./Q4_K_M-KAT-Coder-V2.5-Dev-35B-A3B-ABLITERATED-UNCENSORED-PHILADELPHIA-CLASS.gguf \ -ngl 99 \ -c 32768 \ --jinjaOr import the same file into Ollama:
text
FROM ./Q4_K_M-KAT-Coder-V2.5-Dev-35B-A3B-ABLITERATED-UNCENSORED-PHILADELPHIA-CLASS.ggufPARAMETER num_ctx 32768Save that as Modelfile, then run ollama create kat-coder-philadelphia -f Modelfile.
This model is derived from Kwaipilot/KAT-Coder-V2.5-Dev, which in turn builds on the Qwen3.6-35B-A3B family. Please retain the upstream attribution and follow the Apache 2.0 license.
pip install "transformers[serving]==5.14.1" acceleratetransformers serve KridgeDookie/KAT-Coder-V2.5-Dev-35B-A3B-ABLITERATED-UNCENSORED-PHILADELPHIA-CLASS --port 8000pip install "vllm>=0.19.0"vllm serve KridgeDookie/KAT-Coder-V2.5-Dev-35B-A3B-ABLITERATED-UNCENSORED-PHILADELPHIA-CLASS \ --port 8000 \ --max-model-len 32768 \ --reasoning-parser qwen3 \ --language-model-onlyhf download KridgeDookie/KAT-Coder-V2.5-Dev-35B-A3B-ABLITERATED-UNCENSORED-PHILADELPHIA-CLASS \ --include "Q4_K_M-*.gguf" \ --local-dir .llama-cli \ -m ./Q4_K_M-KAT-Coder-V2.5-Dev-35B-A3B-ABLITERATED-UNCENSORED-PHILADELPHIA-CLASS.gguf \ -ngl 99 \ -c 32768 \ --jinja