Skeptics argue highly ranked fine-tunes often lack genuine utility
GLM-5.2 has been "stuck" at No.2 on Hugging Face Trending for three days, but I'm thrilled to have connected with the creator behind the No.1 project this afternoon. It's been amazing to see open-source work resonating with so many people.
Positive users praised GLM-5.2 as the best open-source model and a major advance for local AI, while negative users criticized its inconsistent outputs like generating wrong symbols and called its fine-tunes low-value slop.
No Digg Deeper questions have been answered for this story yet.
@ZixuanLi_ Thank you so much for the encouragement! Means a lot as an open-source dev. GLM-5.2 is genuinely amazing 🙌
@ZixuanLi_ but here on x,I see GLM-5.2 at number 1 each day
@ZixuanLi_ If you want no.1, release glm5.2flash 26B A4B. Anything small MoE (less than 30B total param)
@notnullptr gemma-4b-code-fable5-opus-composer-grok-distilled-GGUF-VLLM-GLM5.2-DeepseekV4
@notnullptr i cannot believe a thing such as a "slop finetune" can even exist
what an awful waste of compute
@ZixuanLi_ ofc we love glm - but look at the name of the 1st model - bro has everything fable, composer 2.5 gemma, 😆😆
or maybe its a sign for glm to build towards local llms - is that something on the roadmap?
@ZixuanLi_ It's the local king
We need a glm-5.2 35B dense sir
@Dadahelper1 Here!
@notnullptr Our Mythos Fine-Tune is actually really well structured and more than just SFT. It comes with sample outputs, tool outputs and evals, give it a shot!
@notnullptr Rankings are inversely proportional to the count of dashes in the name
@cheatyyyy @notnullptr most of them are slop actually
@notnullptr gable-4-14b-coder-qwopus-composer-writer-heretic-dreamer-MTP-abliterated-uncensored-v3.4-remix-gguf:Q2_K_XL
@ZixuanLi_ it's not that good tho...
just randomly started to generate some Chinese symbols instead of code
i was expecting opus-level tbh
@notnullptr gemma-4-12B+12B-coder-brappus-nexus-q_4_k_l_m-gguf
@ZixuanLi_ How is that number one project?
@ZixuanLi_ I'm pretty sure a good small glm model would be #1 ^^
I can run the first one on my machine with room to spare... (need i say more ? XD)
@ZixuanLi_ Man you and the team are just amazing!
Really happy with http://z.ai
@notnullptr they're good not slop, don't undermine ppl work i fine tune models and they actually beat their base really well, even other models in the category
u might be saying this cuz u have no idea about how this things work, go to sleep
@ZixuanLi_ We need a cheaper, faster, smaller model that excels at coding with good harness 🫠
@ZixuanLi_ @grok in the hugging face leaderboard, how is the open source project no.1 ?
