---
license: apache-2.0
language:
- en
library_name: transformers
pipeline_tag: text-generation
datasets:
- HuggingFaceFW/fineweb-edu
- mlfoundations/dclm-baseline-1.0-parquet
- HuggingFaceTB/smol-smoltalk
tags:
- boris
- nmai
- gpt2
- 125M
- instruct
---

![Boris](Boris-1.3-125M-Instruct.png)

# Boris-1.3-125M-Instruct

Boris-1.3-125M-Instruct is a 125 million-parameter language model created by New
Millennium Artificial Intelligence (NMAI).

This is an **instruction-tuned model**. It follows instructions and holds a
conversation. For the base (pretrained-only) version, see
[KSP-NMAI/Boris-1.3-125M](https://huggingface.co/KSP-NMAI/Boris-1.3-125M).

## Usage

```python
from transformers import AutoModelForCausalLM, AutoTokenizer

tok = AutoTokenizer.from_pretrained("KSP-NMAI/Boris-1.3-125M-Instruct")
model = AutoModelForCausalLM.from_pretrained("KSP-NMAI/Boris-1.3-125M-Instruct")

messages = [{"role": "user", "content": "What's a good way to start learning C?"}]
ids = tok.apply_chat_template(messages, add_generation_prompt=True, return_tensors="pt")
out = model.generate(ids, max_new_tokens=120, do_sample=True, top_p=0.95)
print(tok.decode(out[0][ids.shape[-1]:], skip_special_tokens=True))
```

## Details

| | |
|---|---|
| Architecture | GPT-2 (pre-LN, learned positional embeddings, tied embeddings) |
| Layers / heads / d_model | 12 / 12 / 768 |
| Context length | 1024 |
| Vocab | 50304 (GPT-NeoX-20B BPE, padded) |
| Tokenizer | `EleutherAI/gpt-neox-20b` |
| Precision | trained in bf16 autocast with fp32 master weights |

## Fine-tuning

Fine-tuned on [smol-smoltalk](https://huggingface.co/datasets/HuggingFaceTB/smol-smoltalk)
and [OpenAssistant](https://huggingface.co/datasets/OpenAssistant) conversational
data (OASST1/OASST2 — specific version not recorded) for *(tokens)* tokens over
*(time)* on one RTX 3060.

| | |
|---|---|
| Final loss | *(fill in)* |
| Final grad norm | *(fill in)* |
| Final learning rate | *(fill in)* |

![Benchmarks](benchmarks.png)

## Limitations

A 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.