Instructions to use N8Programs/BestTerm-440M-Checkpts with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use N8Programs/BestTerm-440M-Checkpts with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="N8Programs/BestTerm-440M-Checkpts")

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("N8Programs/BestTerm-440M-Checkpts")
model = AutoModelForCausalLM.from_pretrained("N8Programs/BestTerm-440M-Checkpts")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use N8Programs/BestTerm-440M-Checkpts with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "N8Programs/BestTerm-440M-Checkpts"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "N8Programs/BestTerm-440M-Checkpts",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/N8Programs/BestTerm-440M-Checkpts

SGLang

How to use N8Programs/BestTerm-440M-Checkpts with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "N8Programs/BestTerm-440M-Checkpts" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "N8Programs/BestTerm-440M-Checkpts",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "N8Programs/BestTerm-440M-Checkpts" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "N8Programs/BestTerm-440M-Checkpts",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use N8Programs/BestTerm-440M-Checkpts with Docker Model Runner:
```
docker model run hf.co/N8Programs/BestTerm-440M-Checkpts
```

BestTerm-440M Checkpoint

Tentative checkpoint for BestTerm-440M.

This is a global parameter-vector SLERP between the released N8Programs/NextTerm-440M base model and the hot b-file-only continued-pretraining checkpoint at 500M tokens, with interpolation t=0.80.

Quick Scores

Ryskina & Knight sequence completion, exact next-term accuracy:

Greedy, old PyTorch/Transformers sanity path: 38/57 (66.67%).
Beam search, num_beams=4, comma stop and EOS/PAD suppressed: 40/57 (70.18%).

Other preservation checks from the same sweep:

OEIS-Eval-Neo: 6532/19034 (34.318%).
M1 Competition 111 macro MAPE: 17.582548.
Polynomial continuation: arithmetic 94.5625%, quadratic 86.3043%, cubic 74.5682%, quartic 67.9524%.

Notes

This checkpoint is tentative and was selected as the aggressive Ryskina/M1 point on the base-to-hot500 SLERP sweep. The more conservative preservation point was t=0.60.

Downloads last month: -

Safetensors

Model size

0.4B params

Tensor type

BF16

Model tree for N8Programs/BestTerm-440M-Checkpts

Base model

N8Programs/NextTerm-440M

Finetuned

(1)

this model