Boris-75M
Boris-75M is a 75 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-75M-Instruct.
Usage
from transformers import AutoModelForCausalLM, AutoTokenizer
tok = AutoTokenizer.from_pretrained("KSP-NMAI/Boris-75M")
model = AutoModelForCausalLM.from_pretrained("KSP-NMAI/Boris-75M")
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 / 9 / 576 |
| 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 1.55B tokens for 14:49:08 on one RTX 3060.
| Final loss | 3.6356 |
| Final grad norm | 0.328 |
| 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.