Banaxi-Tech (Banaxi)

reacted to their post with 🔥 about 15 hours ago

Post

1169

We're announcing our BananaMind 2.1 model series!
The models will include:
- BananaMind 2.1 Nano: 10M parameters with 60B tokens.
- BananaMind 2.1 Lite: 25M parameters with 40B tokens.
- BananaMind 2.1 Flash: 50M parameters with 55B tokens.
- BananaMind 2.1 Pro: 135M-145M parameters (still deciding) with 100B tokens.
These model will use a multi tower architecture (like BananaMind/BananaMind-2.1-Unified) with some more architectural changes.

BananaMind 2.1 Pro will probrably use 2 no output towers, instead of one!

We're currently training some experimental models based on this architecture to see its scaling!

Follow us:

BananaMind
@Banaxi-Tech
@vovaRL
@DedeProGames

bananamind-research-community

4 replies

·

posted an update about 16 hours ago

Post

1169

We're announcing our BananaMind 2.1 model series!
The models will include:
- BananaMind 2.1 Nano: 10M parameters with 60B tokens.
- BananaMind 2.1 Lite: 25M parameters with 40B tokens.
- BananaMind 2.1 Flash: 50M parameters with 55B tokens.
- BananaMind 2.1 Pro: 135M-145M parameters (still deciding) with 100B tokens.
These model will use a multi tower architecture (like BananaMind/BananaMind-2.1-Unified) with some more architectural changes.

BananaMind 2.1 Pro will probrably use 2 no output towers, instead of one!

We're currently training some experimental models based on this architecture to see its scaling!

Follow us:

BananaMind
@Banaxi-Tech
@vovaRL
@DedeProGames

bananamind-research-community

4 replies

·

replied to their post about 17 hours ago

Our BananaMind 2.1 Model Series will use a Unified style architecture with some more changes!

replied to their post about 17 hours ago

sometimes crazy ones fail but they might end up being very good i usualy just screw around till one works lol

Yeah! This one i thought wont work too but i just thought ill make it and it was

replied to their post about 20 hours ago

i cant write triton kernels bu sometimes i use codex to write them but not most of the time no

replied to ProCreations's post about 21 hours ago

Also @ProCreations I submitted but I realized I can train locally faster with optimization, if I won give it to someone else!

replied to their post 2 days ago

Not really, its still a test model but a test for out new architecture!

replied to juiceb0xc0de's post 2 days ago

Can you (if its possible) make a atlas of my huggingface.co/BananaMind/BananaMind-2.1-Unified, it would be fascinating to see!

replied to Felladrin's post 2 days ago

It's, gemini, worse than qwen3.5

replied to Felladrin's post 2 days ago

xD

reacted to their post with 👀🚀🔥 3 days ago

Post

3659

We're releasing BananaMind 2.1 Unified, a 35M three-tower relay model where the two output towers can only talk to each other through a silent middle tower that has no output head and no loss term.

Tower B trains entirely on indirect gradient. It was never told what to predict. It woke up anyway. Jacobian lens shows it carrying the correct answer ("Paris", "oxygen", "blue") at its deepest layer. Its bridge gates grew 4-47x from init. Feed it from only one side and the representations collapse to junk — it needs both outer towers to become semantic.

The PIQA result is the cleanest demonstration: Tower A alone scores 50.11 (chance). Tower C alone 52.12. Full system 61.75. All physical reasoning lives in the integration. The 35M three-tower beats the 50M single-tower BananaMind 2 Medium on PIQA.

Trained on 38B tokens in ~7.5 hours on 8x RTX PRO 6000. Ships with 7 ablation modes so you can surgically cut the model apart without retraining. Full training logs, J-lens fits, and eval outputs for every mode included.

This is a research model. It is interesting because you can take it apart.
BananaMind/BananaMind-2.1-Unified

Follow us for more:

BananaMind

@Banaxi-Tech

11 replies

·

posted an update 3 days ago

Post

3659

We're releasing BananaMind 2.1 Unified, a 35M three-tower relay model where the two output towers can only talk to each other through a silent middle tower that has no output head and no loss term.

Tower B trains entirely on indirect gradient. It was never told what to predict. It woke up anyway. Jacobian lens shows it carrying the correct answer ("Paris", "oxygen", "blue") at its deepest layer. Its bridge gates grew 4-47x from init. Feed it from only one side and the representations collapse to junk — it needs both outer towers to become semantic.

The PIQA result is the cleanest demonstration: Tower A alone scores 50.11 (chance). Tower C alone 52.12. Full system 61.75. All physical reasoning lives in the integration. The 35M three-tower beats the 50M single-tower BananaMind 2 Medium on PIQA.

Trained on 38B tokens in ~7.5 hours on 8x RTX PRO 6000. Ships with 7 ablation modes so you can surgically cut the model apart without retraining. Full training logs, J-lens fits, and eval outputs for every mode included.

This is a research model. It is interesting because you can take it apart.
BananaMind/BananaMind-2.1-Unified

Follow us for more:

BananaMind

@Banaxi-Tech

11 replies

·

replied to Felladrin's post 3 days ago

Its not bad but I wouldn't be calling it innovation.

replied to their post 3 days ago

https://github.com/BananaMind/BananaMindOS

reacted to vovaRL's post with 🔥 4 days ago

Post

4060

I made an AI agent for end-to-end ML research. It is based on Claude Agents Python SDK (you can use it with your existing Claude subscription). Also needs Voyage AI (optional, but highly recommended for quality paper analysis). The workflow is: you give the agent a research question (or topic/direction), it pulls papers from papers-with-code and arxiv, reranks them and picks top-15 (with Voyage), then it makes a research plan, writes pipelines (with tests, preflight checks, smoke-runs). Then it uses Kaggle, HF Jobs, or SSH to a GPU to execute the pipeline. Finally, it gets the results, makes visualizations, writes the paper (with citing from papers ingested after step-1) and makes a jupyter notebook. Also there is an auto-generated wiki (like DeepWiki) to quickly understand the generated code. The code is fully open-source under permssive MIT license (it is a personal project). Here is the github link: https://github.com/view321/Grad and a workspace screenshot is attached to the message.

2 replies

·

replied to their post 4 days ago

Probrably

replied to aiqualitylab's post 5 days ago

You could use Qwen3.5 0.8B, Its smarter and smaller.

posted an update 5 days ago

Post

136

New Architecture.
Training started. Few days and you'll see what it is exactly.
Its 2 models connected into 1 unified model.

Banaxi PRO

AI & ML interests

Recent Activity

Organizations

Banaxi PRO

AI & ML interests

Recent Activity

Organizations

Banaxi-Tech's activity