Boris-250M

Boris-250M is a 250 million-parameter language model created by New Millennium Artificial Intelligence (NMAI).

This is a base (pretrained) model. It has not been instruction-tuned and does not follow instructions or hold a conversation — it continues text. For an instruction-following version, see KSP-NMAI/Boris-250M-Instruct.

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer

tok = AutoTokenizer.from_pretrained("KSP-NMAI/Boris-250M")
model = AutoModelForCausalLM.from_pretrained("KSP-NMAI/Boris-250M")

ids = tok("The ocean is", return_tensors="pt").input_ids
out = model.generate(ids, max_new_tokens=40, do_sample=True, top_p=0.95)
print(tok.decode(out[0], skip_special_tokens=True))

Details

Architecture	GPT-2 (pre-LN, learned positional embeddings, tied embeddings)
Layers / heads / d_model	12 / 18 / 1152
Context length	1024
Vocab	50304 (GPT-NeoX-20B BPE, padded)
Tokenizer	EleutherAI/gpt-neox-20b
Precision	trained in bf16 autocast with fp32 master weights

Base model training

Trained on 5.01B tokens for ~105:56:17 on one RTX 3060.

Final loss	3.0693
Final grad norm	0.225
Final learning rate	6.00e-05

Limitations

A base model of this size will produce text that is frequently inaccurate, inconsistent, or offensive. It has received no alignment or safety tuning and should not be used for factual reference or deployed without supervision.

Copyright & License

Copyright 2026 Joseph Jones

This project and all associated files (the "Work") are licensed under the Apache License, Version 2.0 (the "License"); you may not use this project except in compliance with the License. You may obtain a copy of the License at:

http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License.