GLM-5.3 is live. Run Z.ai's latest model on Friendli Model APIs. Try it today ➜
pineapplesuperfan
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Model Details
Model Provider
pineapplesuperfan
Model Tree
TheDrummer/Artemis-31B-v1
zerofata/G4-MeroMero-31B
Input Modalities
Output Modalities
Supported Functionality
Merge of TheDrummer/Artemis-31B-v1 with zerofata/G4-MeroMero-31B, trying to keep the same character as sophosympatheia's 0.3.1 while fixing some bugs the Q8 merge may have caused. Since 0.3.1 was published, TheDrummer has published the full BF16 weights of v1h, so I took the opportunity to do this (same yaml as the original merge).
Not abliterated yet. I plan to do it at some point on q4 on a single 4090, but it would be better if someone could do a full bf16 abliteration on it. For now I use sh0ck0r/Mero-Artemis-31B-v0.3.1-heretic, which is excellent.
Sampler Settings: Try the ones sophosympatheia recommends:
Personally I use temp 1.0, mip-p 0.05 and dry 0.8, dry-base 1.75. Adaptive seemed to loosen creativity a touch, so I skip it.