Model Requests & Suggestions
Suggest a model for a future handcrafted MiniPlus or NanoPlus release.
Please include:
- Upstream model link
- Intended use or strengths
- Why it would be valuable at a MiniPlus or NanoPlus footprint
I will review requests based on demand, architecture, local-inference practicality, and whether the model would benefit from a custom tensor-by-tensor quantization.
CyberTiel
- Link: peculiar-ragdoll/Cyber-Tiel-Coder-35B-A3B-GGUF-MTP (https://huggingface.co/peculiar-ragdoll/Cyber-Tiel-Coder-35B-A3B-GGUF-MTP)
- Just like Tiel , agentic coding tasks, but stronger, less thinking(in my opinion) and uncensored
- Miniplus like what you did on Tiel
The MiniPlus V2.1 and NanoPlus quantizations are already underway; they will be available within an hour at mostβfollow me to get notified.
Wow
Thanks for your hard works and very fast reply, really impressive, this is the fastest model request i've seen on hf (or maybe you had already worked on this model from before :))
I will definitely check and test for if the Apex Miniplus CyberTiel work good too
Request for next model.
Reason: I've personally used k2 and it's the closest thing to match or exceed Qwen intelligence:
IFM/K2-Horizon-MoVA-36B-A4B
I personally tried the K2-Horizon, the smaller versions and that MoVA, and they didn't convince me because of the cost of the KV Cache.
I could barely get it to 8K context in Q4_0, not enough to properly test it, is it really any good?
For me, the fact that it had such an absurdly expensive KV cache was enough to completely rule it out. Did you guys test it in real-world use?
You were using Q4_0 KV?
How much VRAM on the rig?
I assume your models were tested on 24gb ram?
I need a 35b a3b that even beat qwen 3.8 27b.
Qwen 3.6 35b a3b is easy to beat
I need a 35b a3b that even beat qwen 3.8 27b.
Qwen 3.6 35b a3b is easy to beat
I don't think easy to beat is true, every MoE that comes out is comparing to Qwen3.6 35B A3B, the 40B A2B Flint couldn't even beat it, finetunes like Occamy 1.0 and Nex2.5 Mini drop it hard on coding, so the choice is hard, the finetunes that actually beat it:
1- KAT Coder Dev v2.5
2- Ornith 1.0/Ornith 1.5 and their respective coding focused quants Tiel Coder/Cyber Tiel.
3- Sage 1.5
And the ones that aren't finetunes:
1- Clover 35B A3B
2- Maxwell V1 Flash
These are the ones as far as I remember, oh put in mind, I treated you as someone who wants a coding model (most do so), if not correct me so I can recommend a model/finetune in your field.
Hi IsValorum,
I'm wondering if you might consider creating a MiniPlus of Gemma 4 26B-A4B Instruct (https://huggingface.co/google/gemma-4-26B-A4B-it)
I'm aware you have already created a ThinkingCap finetune of this model, but I am a fan of original models rather than finetunes. I would like to use Google's original Instruct model at a higher quality level than the QAT version for its strong conversation and narrative abilities. I believe your APEX MiniPlus would provide that quality and speed for Gemma 4 26B-A4B, based on what I've seen of your Qwen 3.6 35B-A3B MiniPlus when I compared it against regular Q4_K quants. It really is a very strong model for conversation and creative tasks, and for my vram-poor system, Apex quants are my only way of hopefully increasing quality without sacrificing speed or context length.
Regardless if you consider my request or not, I do want to say thank you. I've replaced most of my MoE collection with your Apex quants.
@IsValorum thanks for the T.C. Gemma quants! If it's useful at all I can easily apply your templates from it to the -it model for SpmBx000.
@madc read JOURNEY.md from the repo, you might change your mind.
Hi IsValorum,
I'm wondering if you might consider creating a MiniPlus of Gemma 4 26B-A4B Instruct (https://huggingface.co/google/gemma-4-26B-A4B-it)
I'm aware you have already created a ThinkingCap finetune of this model, but I am a fan of original models rather than finetunes. I would like to use Google's original Instruct model at a higher quality level than the QAT version for its strong conversation and narrative abilities. I believe your APEX MiniPlus would provide that quality and speed for Gemma 4 26B-A4B, based on what I've seen of your Qwen 3.6 35B-A3B MiniPlus when I compared it against regular Q4_K quants. It really is a very strong model for conversation and creative tasks, and for my vram-poor system, Apex quants are my only way of hopefully increasing quality without sacrificing speed or context length.
Regardless if you consider my request or not, I do want to say thank you. I've replaced most of my MoE collection with your Apex quants.
You can try Air-26B-A4B if you want close to the original. It's pretty good, honeslty
βI know most fine-tuning improvements boost some benchmarks while making others awful, or at least that's usually how it goes.
βThe Gemma family is a prime example of a model that urgently needed a fine-tune to fix its habit of overthinking and breaking everything down.
βTry ThinkingCap Gemma. Personally, I never used Gemma as my everyday model because of that it would dissect a simple "Hi" as if it were studying it.
βAlso, before calling a tool, it would do trial runs while thinking, and sometimes it wouldn't even call it and would just keep going in circles.
βI haven't noticed any drop in capability or intelligence where Gemma with ThinkingCap already shines compared to the original, they perform just as well.
Maxwell V1 Flash β
MiniPlus https://huggingface.co/IsValorum/Maxwell-V1-Flash-APEX-I-MiniPlus-V2.1-GGUF
NanoPlus https://huggingface.co/IsValorum/Maxwell-V1-Flash-APEX-I-NanoPlus-V2.1-GGUF