--- license: apache-2.0 language: - en base_model: - Qwen/Qwen3-32B datasets: - SicariusSicariiStuff/UBW_Tapestries widget: - text: "Assistant_Pepe_32B" output: url: https://huggingface.co/SicariusSicariiStuff/Assistant_Pepe_32B/resolve/main/Images/Assistant_Pepe_32B.png ---
---
---
> What happens if we fry Qwen's brain?
"This pile of tensors definitely needs therapy" đž??? âđ©đ»ââïžđ§
---
# This finetune almost didnât happen
This finetune was an absolute pain to make. I wanted to scrap it (multiple times!) and move on. ButâŠ
I got a (very) generous donation to make a 32B version, so I said: **âI will 100% make it happen!â**âhence, **I was committed**. And this was after the 8B and 70B were received with much more positivity than common sense could have divined, so sure, why not make a 32B version? **I was compelled to**. Surely a 32B version would be a walk in the park. Surely. Well, **it wasnât**. I really did want to scrap it.
Iâll just start with the clichĂ© âItâs not about the moneyâ, because it really wasnât. If I had trained this degenerate pile of tensors in the cloud, the bill wouldâve been well into the **four figures** by now. Easily. Not to mention the sheer amount of time (and pain) this took; even putting the compute aside, it was absolutely absurd. But Iâm a man of my word, and I did make a promise. And Iâm so glad I did.
# âQwen is a bad base for creative stuffâ
Many very talented and experienced tuners (rightfully) complain about **Qwen being extremely hard to train**. Not on STEM, but on anything⊠other than STEM. **Especially creative stuff**. Qwen has a very strong, distinct âQweninessâ. Itâs stubborn, rigid, and very dry. Tuning it for any creative endeavor seemed like a lost cause, and to the best of my knowledge, thereâs only a **tiny amount** of creative tunes for it, and they are just not especially popular vs. their Mistral/Llama counterparts from the same authors, while using similar datasets. For creative stuff, **Mistral** and **Llama** simply (consistently) give better results; thatâs a known fact.
But holy guacamole, did this model cook! The result was extraordinarily unique, fresh, hilarious, and very, VERY much unhinged. I had to **push very hard** to get the âQweninessâ and thinking/assistant slop out of it. This model is obviously not as smart as the 70B variant, and maybe even less smart than the 8B variant (in some aspectsâlike coding), but it is **more special** than both. Iâm serious, this is not cope; **something unique and unpredictable sometimes happens** when you take a model with very specific priors, persona, and vibe (in this case, STEM with a focus on thinking) and make it do something completely alien (see [Phi-Lthy4](https://huggingface.co/SicariusSicariiStuff/Phi-lthy4) for example).
This is the first model that gave me the **uncanny âself-aware**â facades of [TenebrÄ-30B](https://huggingface.co/SicariusSicariiStuff/Tenebra_30B_Alpha01) and its smaller variant, the very first models I published on HuggingFace at the **end of 2023** and the **very beginning of 2024**. Ever since then, people occasionally asked for a **new TenebrÄ**, but the dataset for those was, unfortunately, forever lost. Interestingly, this model shares the same size as TenebrÄ, but has a modern architecture. While TenebrÄ was based on old Llama-1, this one is Qwen-3-based, has superb context, and Iâd argue, the same **uncanny, fun quirks** that TenebrÄ had.
One of the driest, most âroboticâ base models has birthed one of the arguably **most human-like** finetunes. It is **exceptionally fun to talk to** and is simply amazing at any creative/brainstorming endeavor. I couldâve served this to 1,000 people, and probably at least 95% of them wouldnât believe this is Qwen underneath. Based đ€
---
### TL;DR
- Easily the most **human** Qwen-3 tune.
- **No-thinking!** think haters, rejoice!
- Can still think though, if explicitly prompted.
- **NO SYSTEM PROMPT REQUIRED!** The persona is baked into the weights :)
- Will **go HARD roasting you**! AND **itself**, thanks to a ~~not so subtle~~ **negativity bias infusion**.
- Excellent and **extremely creative** writer ([See examples](https://huggingface.co/SicariusSicariiStuff/Assistant_Pepe_32B#chat-examples-click-below-to-expand)).
- Unique writing patterns, **absolutely minimal amount of slop!**
- Superb **long context**, thanks to Qwen-3 32B base (see [nVidia's RULER](https://github.com/NVIDIA/RULER) for more details). Expect superb coherency at 32K, and **very good coherency even at 64K!**
- Possibly **more unhinged and degenerate** than the previous **8B** & **70B** version.
- Massive **swipe-diversity**.
- **Exceptionally** fun to talk to!
- **VERY** human-like
- Does **NOT** feel like Qwen-3, at all!
- Inclusive toward amphibians.
---
## Model Details
- Intended use: **Shitposting**, **Creative Writing**, **Brain-storming**, **Chat**.
- Censorship level: Low - Very low
- **7.5 / 10** (10 completely uncensored)
## UGI score:
---
## One of the absolutely best models for spicy writing:
---
## Available quantizations:
- Original: [FP16](https://huggingface.co/SicariusSicariiStuff/Assistant_Pepe_32B)
- GGUF: [Static Quants](https://huggingface.co/SicariusSicariiStuff/Assistant_Pepe_32B_GGUF)
- EXL3: [3.0 bpw](https://huggingface.co/dr-housemd/Assistant-Pepe-32B-exl3-3.00bpw) | [3.5 bpw](https://huggingface.co/dr-housemd/Assistant-Pepe-32B-exl3-3.50bpw)
- GPTQ: [4-Bit-128 AutoRound](https://huggingface.co/SicariusSicariiStuff/Assistant_Pepe_32B_GPTQ_AR_4-bit-128)
- Mobile (ARM): [Q4_0](https://huggingface.co/SicariusSicariiStuff/Assistant_Pepe_32B_ARM)
# Generation settings
**Recommended settings for assistant mode:**