Skyfall-31B-v4.2-heretic-exl3-4.0bpw

meddle86/Skyfall-31B-v4.2-heretic-exl3-4.0bpw

Skyfall-31B-v4.2-heretic-exl3-4.0bpw is a machine learning model from meddle86. Use it for the machine learning task on the model card, and read the license before you ship it in a product.

This is a Mul1 quant, compatible with Exllama v1.0.0+ (but should also work with older versions also)

Downloads · 30 days

11% of all-time downloads

All-time downloads

100

Public

Parameters

8.5B

17.1 GB on disk

Likes

Public

Hugging Face

Repo makeup

Click a slice to open those files.

.safetensors17 GB · 100%

Parameter types

How the weights are stored.

I167.8B · 92%

Base models

TheDrummer/Skyfall-31B-v4.2

Model card

From the Hugging Face model README

This is a Mul1 quant, compatible with Exllama v1.0.0+ (but should also work with older versions also)

This is a decensored version of TheDrummer/Skyfall-31B-v4.2, made using Heretic v1.2.0

Abliteration parameters

Parameter	Value
direction_index	21.36
attn.o_proj.max_weights.0	0: 1.48
attn.o_proj.max_weight_position	36.76
attn.o_proj.min_weights.0	0: 1.48
attn.o_proj.min_weight_distance	27.61
mlp.down_proj.max_weights.0	0: 0.99
mlp.down_proj.max_weight_position	38.21
mlp.down_proj.min_weights.0	0: 0.46
mlp.down_proj.min_weight_distance	30.90

Performance

Metric	This model	Original model (TheDrummer/Skyfall-31B-v4.2)
KL divergence	0.0050	0 (by definition)
Refusals	6/100	89/100

Drummer is open for new opportunities (I'm a Software Engineer). Contact me through any of these channels: https://linktr.ee/thelocaldrummer

Thank you to everyone who subscribed through Patreon. Your support helps me chug along in this brave new world.

FAQ for those out-of-the-loop

<details> <summary>🐶 Who is Drummer?</summary>

Hi! I'm Drummer. I'm a Software Engineer with experience in JavaScript, Golang, Python, and generally engineering the crap out of things.

Why I'm in the AI space:

Exploration: Everyone is trying to figure out how AI works and what it's capable of. I am too - just not in creating the smartest, safest model at all costs.
Upskill: The world is headed towards AI. It is here to stay. This has been my way of brushing up in this new form of computing challenge.
Value: I yearn to create value. I feel satisfaction and fulfillment in providing something meaningful for others.
Fun: It's just fun using and making models. It's also fun coming up with theories and realizing them in practice (training AI).

I started my tuning venture back in mid-2024 when I wanted to improve its literary capabilities. I've come a long way since then and I have branched out and specialized. Foundational models today are optimized for non-creative uses, and I believe there is a place for AI in creativity and entertainment.

I am here to take the road less traveled by.

</details> <details> <summary>❓ What are my models like?</summary>

Bottomline: My models are usually geared towards creativity, usability, and entertainment!

While intelligence, correctness, and problem solving are not my priority, they are still one of many qualities I want in my models.

The primary goal is to enhance the experience for users looking to use models for creative uses, and other use cases which require no alignment.

In an effort to make it clear to myself and to others what I'm aiming for, I've identified certain qualities that my users often want:

Creativity

Writing: Does it string together words and sentences in a pleasant & effective way? Does it feel like a writer?
Dynamism: How good is the AI at being compelling and intriguing in its storytelling?
Imagination: Can the AI navigate through a plethora of possibilities? Can it skirt incoherence and rise up to absolute coherence at the end of it?

(Dis)alignment

Attitude: Does it refuse in both soft or hard ways? Does it lean towards certain corporate/religious/political ethics & beliefs? How does it see the user and itself?
Morality: Does it know ethics? Is its language infected with forced positivity? If not, can it still moralize over difficult & dubious themes?
Formatting: How stubborn is it with its established formatting? Can it create effective and novel formats to answer the prompt?

Intelligence

Adherence: Can it follow instructions? Is it sticking to the prompt? Can it understsand you?
Knowledge: Does it know about the world in both fictional and non-fictional way?
Perception: Can it handle nuance, complexity, and logic?

If it doesn't excel in one of these qualities, or if it's overall mediocre for its size, then I would most likely reiterate until I get something right.

</details> <details> <summary>💡 Philosophy</summary>

A person is defined by the language they use. Not whether they speak in English or German, but in how they perceive reality.

Just like how we associate a serial killer as a mind that can't map 'murder' to 'evil', an innocent person is a mind that simply can't imagine 'murder'. They get confused when forced to deal with such subjects.

AI's use of language speaks volumes about their 'perception' of reality. If a language model has been skewed and limited to a positive perception, then it's ability to imagine is also limited.

Finetuning is an opportunity to adjust and broaden the language. Corporations use it to achieve safety and compliance. I'm here to

</details>

Drummer proudly presents...

Skyfall 31B v4.2

Supported Chat Templates

Mistral v7 Tekken

Reviews

Surprisingly good inference of complex emotions. I asked it to deduce what a character is feeling at a certain point in my RP, and it interpreted correctly where GLM 4.6 failed.

This one's really good! Using it with Rimtalk, zero problems with json output. Dialogues are creative and coherent. Been noticing a lack of narration/description, but that may be my prompting. It also feels less horny, the spice seems more veiled. I'm used to playing with v4p which had good amount of sauciness I think

I’ve spent a couple hours playing with it, I’m up to about 14k context and it’s enjoyable.

Yeah, really surprised by the dialogues with this one too. They feel more natural, direct, less awkward.

🤩 I adore v4y, it plays the card better, it's wittier, warmer and it does seem smarter.

Because it's more consistent while still being more creative than most local models I've tried so far. It made me appreciate longer replies without it suddenly speaking for me. This model really knows, at least in my chats, that it's {{char}} and not {{user}}. I love the humor of this model, I love how it understands a lot of (personality)tropes, jokes, etc... Sure. The replies can be insanely long (over 500 tokens no think) but it somehow get's the pacing right imo by not advancing the story too hasty and not too slowly, especially for the long replies. I barely had to edit any messages after the first few as it stayed pretty consistent until it doesn't, and swipes make the same mistake once it does make one.

Finally got some time to try this Skyfall. This is really good, the writing is very naturalistic, and this model is very good with prompt adherence and using the appropriate tone/writing style for a given card

This model is great, i am at 20k context 1 char with 3 regular npc's and different narrated ones. It holds strong without mixing. I am the Narrator as user and tell the story after i let my {{user}} die in the first 6k context.

I've been using this model for a couple days and have really started to love it. It's funny and clever. Def verbose, but surprisingly not very repetitive compared to other models that start to get wordy. It might become my main model over a gemma-based one I've liked for a long time. It's not utterly perfect in every way, but it feels like it's punching above it's weight class.

This skyfall feels really good to use btw. No hard numbers as I did not do serious testing, but both understanding and writing style feels like a nice upgrade from both previous skyfalls and recent cydonias.

This Skyfall is sooo good. The word choices it makes :chef2: . Damn shame this hasn't been released, this one is in a different league from earlier Skyfalls and they were excellent models. It has a fresh vocabulary and it doesn't need everything spelled out in your cards to get what you are aiming for.

I'll say this one is the overall best Skyfall so far for RP. Really good characterization ability. I'd really like to see this one released if you are taking a break from training.

Honestly this skyfall is cinema.

This model here is Drummer's magnum opus. Unbelievably good

Special Thanks

Thank you to each and everyone who donated and subscribed in Patreon and Ko-Fi to make our venture a little bit easier.
Subscribe to my Patreon!