GLM-5.2 is live. #1 throughput on OpenRouter, pay-per-token on FriendliAI. Try it today ➜
KSP-NMAI
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Run this model inference with full control and performance in your environment.
Model Details
Model Provider
KSP-NMAI
Model Tree
KSP-NMAI/Boris-75M
Input Modalities
Output Modalities
Supported Functionality
This model uses the Alpaca format. A chat template is included in
tokenizer_config.json, so apply_chat_template produces the correct prompt
automatically:
python
from transformers import AutoModelForCausalLM, AutoTokenizertok = AutoTokenizer.from_pretrained("KSP-NMAI/Boris-75M-Instruct")model = AutoModelForCausalLM.from_pretrained("KSP-NMAI/Boris-75M-Instruct")messages = [{"role": "user", "content": "What is the capital of France?"}]prompt = tok.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)ids = tok(prompt, return_tensors="pt").input_idsout = model.generate(ids, max_new_tokens=64, do_sample=True, top_p=0.95, temperature=0.7)print(tok.decode(out[0][ids.shape[1]:], skip_special_tokens=True))
If you are building the prompt by hand, the layout is:
markdown
### Instruction:{your instruction}### Response:
Generation should stop at ### Instruction: (or end-of-text, token id 0).
| 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 |
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 |
The table above describes the base model's pretraining run; the instruction tuning was applied on top of that checkpoint.
This is a very small instruction-tuned model. It will produce text that is frequently inaccurate, inconsistent, or offensive, and it has received no alignment, RLHF, or safety tuning beyond supervised fine-tuning on Alpaca. Do not rely on it for factual information or deploy it without supervision.
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:
markdown
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.