huihui-ai/Llama-3.3-70B-Instruct-abliterated

Text Generation • 71B • Updated Dec 12, 2024 • 1.1k • 62

Note Yet to use.

Babsie/BrownLoafers-70B

Text Generation • 71B • Updated Nov 27 • 48 • 3

Note RP version of Hermes 4 - combined with Arli RP MAX. Bench scores were near identical, thought it would be a good merge. Apparently others think so - 5k downloads including ggufs. I personally don't care for it's writing. I find it tropey and horrifically purple. But, you know, I like boxed mac and cheese and pot noodles, so who am I to judge.

Note Nous re-training for YaRN stretches. Gutenberg press library, full novels. Produced brilliant result on retraining the 13B 64K. Will be using on Openhermes 2 13B 32K and Hermes2 Yi 34B 40K

Note re-training for YaRN stretch in Cappy.

Babsie/Hermes2Yi-34B-40k

Text Generation • 34B • Updated Oct 16 • 14

Note RoPE stretched to 40K from 4K. tested responses in pod with 500K token responses. seems ok. Needs to be tested in chat. May need re-training with YaRN dataset.

Babsie/CapyberaHermesYi-34B-ChatML-200K

Updated Oct 22 • 5

Note Chat template added as there was none. Very briefly tested in chat to 2K tokens to see if template worked. Will be testing further.

Babsie/OpenHermes2-13B-32K

Text Generation • 13B • Updated Oct 15 • 6

Note RoPE stretched to 32K. Language drunk. Needs fine tuning re-training. Will be doing with a subset of Nous "deepmind/pg19" re-training for YaRN dataset.

Babsie/NousYarnFlashLlama-13B-64k

Text Generation • Updated Oct 24 • 16

Note Flash attention Added. Tested briefly in pod to ensure chat template and stability with 1k token response. Still need to test in chat.

NousResearch/Hermes-4-70B

Text Generation • 71B • Updated Sep 2 • 6.78k • • 160

Note - model included in merges

NousResearch/Hermes-2-Theta-Llama-3-70B

Text Generation • 71B • Updated Sep 8, 2024 • 373 • • 80

Note - model included in Trickster Theta merge

Babsie/Trickster-Theta-4-70B

71B • Updated Nov 13 • 27 • 5

Note Finished! Finally. 🎉 Big personality. RP/writer/co-creator AI. Handles large, complex RP cards. NSFW. At a guess? 8.8-9/10. Leans a little heavy. I had to slap it once; it stropped off in a screeching RADA huff. Called it back, had a gentle word. Settled. Probably an artifact of RHITL conditioning that favours “0-100 in 0.7 of a paragraph” at any whiff of attraction. *Eyeroll.* Long live slow burn intellectual flirtation. UPDATE: Base Model for upcoming ThetaPhysis and LogiVitrar

Babsie/Trickser_Theta_4_70B_calibrated_gguf

71B • Updated Nov 4 • 60 • 1

Note i1_Q GGUFs - calibrated using specific set of data made for trickster. details on model card.

Babsie/TrckstrExtDimenBrainDamageNO_USE_RefernceOnly

Text Generation • 71B • Updated Nov 8 • 4

Note NOTE: define base number in all future builds with SCE+TIES hybrid merge. Never use "all crimes" again.