Instructions to use DedeProGames/GPT-U-20M with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use DedeProGames/GPT-U-20M with Transformers:

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

pipe = pipeline("text-generation", model="DedeProGames/GPT-U-20M")

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("DedeProGames/GPT-U-20M")
model = AutoModelForCausalLM.from_pretrained("DedeProGames/GPT-U-20M", device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use DedeProGames/GPT-U-20M with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "DedeProGames/GPT-U-20M"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "DedeProGames/GPT-U-20M",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/DedeProGames/GPT-U-20M

SGLang

How to use DedeProGames/GPT-U-20M 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 "DedeProGames/GPT-U-20M" \
    --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": "DedeProGames/GPT-U-20M",
		"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 "DedeProGames/GPT-U-20M" \
        --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": "DedeProGames/GPT-U-20M",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use DedeProGames/GPT-U-20M with Docker Model Runner:
```
docker model run hf.co/DedeProGames/GPT-U-20M
```

GPT-U-20M / scripts /common.py

DedeProGames

GPT-U-20M: final weights, tokenizer, model card, scripts and logs

13caab9 verified 15 days ago

Raw History Blame Contribute Delete

1.69 kB

	"""Shared helpers for GPT-U-20M: repo paths, model config and model construction."""
	import json
	import math
	from pathlib import Path

	import torch
	from transformers import LlamaConfig, LlamaForCausalLM

	ROOT = Path(__file__).resolve().parents[1]
	CONFIG_PATH = ROOT / "configs" / "model.json"
	TOKENIZER_DIR = ROOT / "tokenizer"
	DATA_DIR = ROOT / "data" / "tokens"
	OUT_DIR = ROOT / "out"
	CKPT_DIR = OUT_DIR / "ckpt"
	LOG_DIR = ROOT / "logs"

	EXPECTED_PARAMS = 20_453_760
	INIT_STD = 0.02


	def load_config() -> LlamaConfig:
	return LlamaConfig.from_dict(json.loads(CONFIG_PATH.read_text()), attn_implementation="sdpa")


	def init_weights(model: LlamaForCausalLM) -> None:
	"""N(0, 0.02) everywhere; residual output projections (o_proj, down_proj) scaled by 1/sqrt(2 * n_layers)."""
	resid_std = INIT_STD / math.sqrt(2 * model.config.num_hidden_layers)
	with torch.no_grad():
	for name, module in model.named_modules():
	if name == "lm_head" and model.config.tie_word_embeddings:
	continue # shares its weight with embed_tokens
	if isinstance(module, torch.nn.Linear):
	module.weight.normal_(0.0, resid_std if name.endswith(("o_proj", "down_proj")) else INIT_STD)
	elif isinstance(module, torch.nn.Embedding):
	module.weight.normal_(0.0, INIT_STD)
	elif type(module).__name__.endswith("RMSNorm"):
	module.weight.fill_(1.0)


	def build_model(init: bool = True) -> LlamaForCausalLM:
	# Master weights stay in fp32; bf16 only enters through autocast.
	model = LlamaForCausalLM(load_config()).to(torch.float32)
	if init:
	init_weights(model)
	return model