GPUburnout-1B-90K – AI Model by GPUburnout | AlphaNeural AI

You can deploy this model and start earning money today!

GPUburnout/
GPUburnout-1B-90K

like

0

text-generation

Views

No views yet

GPUburnout-1B

A 1.04 billion parameter Llama-style language model trained from scratch on 11.8B tokens for $175.

Model Details

Architecture: Llama-style decoder-only transformer
Parameters: 1.04B
Hidden dim: 2048
Layers: 16
Attention: GQA (32 query heads, 8 KV heads)
FFN: SwiGLU (intermediate 8192)
Position encoding: RoPE (theta=500000)
Context length: 2048 tokens
Vocabulary: 32,005 tokens (BPE + 5 special tokens)
Weight tying: Yes (embedding + LM head)

Training

Data: 11.8B tokens (FineWeb-Edu 85%, Python-Edu 4.2%, FineMath 10.8%)
Hardware: A100 SXM 80GB on RunPod
Steps: 90,000
Final loss: 2.494
Total cost: ~$175 GPU compute
Throughput: ~28,535 tokens/sec

Benchmarks (0-shot)

Benchmark	Metric	Score	Random
ARC-Easy	acc	47.1%	25%
HellaSwag	acc_norm	28.8%	25%
ARC-Challenge	acc_norm	23.3%	25%
MMLU	acc	23.0%	25%

Tokenizer

Includes ChatML special tokens for future SFT:

<|im_start|> (32000), <|im_end|> (32001)
<|system|> (32002), <|user|> (32003), <|assistant|> (32004)

Usage

python

from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("GPUburnout/GPUburnout-1B", torch_dtype="float16")
tokenizer = AutoTokenizer.from_pretrained("GPUburnout/GPUburnout-1B")

inputs = tokenizer("The capital of France is", return_tensors="pt")
outputs = model.generate(**inputs, max_new_tokens=50)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

Blog

Full training journey documented at gpuburnout.com

Author

Jun Park (@GPUburnout)