Cosmos-T2-80M-Test

Universal Kaggle-ready training notebook for the Cosmos-T2 series.

Notebook-generated card. Final metrics are filled after the Kaggle training run. This notebook is designed to stay Kaggle-friendly on 2x T4 GPUs. The goal is a reusable training recipe, not a production assistant.

Model Details

Model class	CosmosT2_LLM
Architecture	Decoder-only Transformer with RoPE, RMSNorm, SwiGLU, GQA, and a configurable Engram memory path
Parameters	~87.60 M
Layers	12
Attention heads	8
KV heads	2
d_model	384
FFN hidden	1536
Positional encoding	RoPE (rope_base=10000)
Normalization	RMSNorm
MLP	SwiGLU
Memory	Engram (use_engram=True, every 2 blocks)
Context length	1028
Training block size	1028
Tokenizer	Qwen/Qwen2.5-0.5B
Dataset	wop/XXXXXL-chain-of-thought
License	Apache-2.0

Why these choices

• RoPE keeps positional handling compact and avoids learned absolute embeddings.
• RMSNorm is cheaper and more stable than LayerNorm for this small decoder-only model.
• SwiGLU usually gives a better quality/compute tradeoff than a plain GELU MLP.
• GQA reduces KV cost while keeping multi-head query capacity.
• Engram gives the stack a lightweight explicit memory path for repeated reasoning patterns.

Training Summary

Metric	Value
Rows used	1000
Approx. packed tokens	177,844
Epochs	50
Batch size	6
Peak LR	3.00e-04
Weight decay	0.1
Gradient clipping	1.0
Wall-clock time	14m 14s
Final training loss	0.0522
Final training perplexity	1.05
Final validation loss	4.2545
Final validation perplexity	70.43
Best validation loss	3.1329
Best epoch	8

Loss and perplexity

The notebook shows live loss and perplexity plots every 20 epochs and does not save the graph to disk.

How to Use

Quick start

import torch
from transformers import AutoTokenizer

from app import CosmosT2_LLM

tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen2.5-0.5B")
if tokenizer.pad_token is None:
    tokenizer.pad_token = tokenizer.eos_token

ckpt = torch.load("$CHECKPOINT_NAME", map_location="cpu")
model = CosmosT2_LLM(**ckpt["config"])
model.load_state_dict(ckpt["model_state"])
model.eval()

prompt = tokenizer.apply_chat_template(
    [
        {"role": "system", "content": "Enable thinking features: INTUITION, COLD START, HOT START"},
        {"role": "user", "content": "What is 12 * 7?"},
    ],
    tokenize=False,
    add_generation_prompt=True,
)
ids = tokenizer(prompt, return_tensors="pt", add_special_tokens=False).input_ids
out = model.generate(ids, max_new_tokens=120, temperature=0.8, top_k=50)
print(tokenizer.decode(out[0], skip_special_tokens=False))

Prompt format

Use the Qwen2.5 chat template. The default system prompt is:

Enable thinking features: INTUITION, COLD START, HOT START

The model will then emit a <think> block followed by an answer when it has enough signal.

Limitations

• The model is intentionally small and is still a research/demo artifact.
• Training on chain-of-thought data can overfit quickly if the corpus is tiny.
• Long-context behavior is limited by the configured block size.
• The model is not safety-aligned and should not be exposed as a public assistant without additional work.

Intended Use

• Research into small-scale pretraining and reasoning-style formatting
• Educational demos for decoder-only Transformer training
• Hugging Face Spaces or local inference demos
• Not for production use

Cosmos-T2 Series

This notebook is designed to train future Cosmos-T2 variants by changing only the config block at the top.

Citation

@misc{cosmos-t2-80m,
  author    = {wop},
  title     = {Cosmos-T2-80M: A small from-scratch chain-of-thought Transformer},
  year      = {2026},
  publisher = {Hugging Face},
  url       = {https://huggingface.co/wop/Cosmos-T2-80M}
}

Acknowledgements

• Tokenizer from Qwen2.5 by Alibaba Cloud
• Training data from wop/XXXXXL-chain-of-thought
• Trained on Kaggle T4 GPUs