Instructions to use transformers-community/dola with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use transformers-community/dola with Transformers:

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

pipe = pipeline("text-generation", model="transformers-community/dola")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("transformers-community/dola")
model = AutoModelForCausalLM.from_pretrained("transformers-community/dola")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=40)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Inference
Local Apps Settings

vLLM

How to use transformers-community/dola with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "transformers-community/dola"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "transformers-community/dola",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/transformers-community/dola

SGLang

How to use transformers-community/dola 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 "transformers-community/dola" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "transformers-community/dola",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

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 "transformers-community/dola" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "transformers-community/dola",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use transformers-community/dola with Docker Model Runner:
```
docker model run hf.co/transformers-community/dola
```

Fix Transformers v5 cached decoding in DoLa

by lavrenko - opened 1 day ago

base: refs/heads/main

←

from: refs/pr/2

Discussion Files changed

+17

-2

lavrenko

1 day ago

•

edited 1 day ago

This PR addresses the cached-decoding regression reported in #1.

The issue appears to be the same Transformers v5 cache-handling compatibility problem previously fixed for transformers-community/group-beam-search: https://huggingface.co/transformers-community/group-beam-search/discussions/4

The goal is to make DoLa generation preserve deterministic behavior when switching between use_cache=False and use_cache=True, while keeping compatibility with current Transformers 4.x behavior.

fix(generation): support v5 cached DoLa decoding33028a28

lavrenko

1 day ago

I pushed the cache-compatibility fix for this PR.

The change updates the DoLa generation loop to follow the Transformers v5 cache protocol: the first decoding iteration is treated as prefill and uses the full prompt, while later cached decoding steps use only the newest token. Uncached decoding keeps the full-prefix behavior.

This addresses the regression reported in #1 and follows the same compatibility pattern already accepted for transformers-community/group-beam-search: https://huggingface.co/transformers-community/group-beam-search/discussions/4

I tested the PR in Colab with Qwen/Qwen3-0.6B, deterministic decoding, and the local PR checkout as custom_generate.

Transformers 5 test

python: 3.12.13
torch: 2.11.0+cu128
transformers: 5.12.0
cuda: True
model: Qwen/Qwen3-0.6B
custom_generate: /content/dola
device: cuda:0

use_cache=False:

ids: [304, 279, 2355, 7982, 1849, 11, 323, 1083, 304, 12785, 8357, 13, 576, 6028, 990, 374, 311, 8317, 9072, 4802, 504, 825, 1992, 311, 2441, 13, 80532, 646, 387, 1483]
text: ' in the power distribution system, and also in industrial applications. The primary use is to transfer electric energy from one place to another. Transformers can be used'

use_cache=True:

ids: [304, 279, 2355, 7982, 1849, 11, 323, 1083, 304, 12785, 8357, 13, 576, 6028, 990, 374, 311, 8317, 9072, 4802, 504, 825, 1992, 311, 2441, 13, 80532, 646, 387, 1483]
text: ' in the power distribution system, and also in industrial applications. The primary use is to transfer electric energy from one place to another. Transformers can be used'

Transformers 4 compatibility test

python: 3.12.12
torch: 2.9.0+cu126
transformers: 4.57.6
cuda: True
model: Qwen/Qwen3-0.6B
custom_generate: /content/dola
device: cuda:0

use_cache=False:

ids: [304, 279, 2355, 7982, 1849, 11, 323, 1083, 304, 12785, 8357, 13, 576, 6028, 990, 374, 311, 8317, 9072, 4802, 504, 825, 1992, 311, 2441, 13, 80532, 646, 387, 1483]
text: ' in the power distribution system, and also in industrial applications. The primary use is to transfer electric energy from one place to another. Transformers can be used'

use_cache=True:

ids: [304, 279, 2355, 7982, 1849, 11, 323, 1083, 304, 12785, 8357, 13, 576, 6028, 990, 374, 311, 8317, 9072, 4802, 504, 825, 1992, 311, 2441, 13, 80532, 646, 387, 1483]
text: ' in the power distribution system, and also in industrial applications. The primary use is to transfer electric energy from one place to another. Transformers can be used'

So the PR fixes the Transformers v5 cached-decoding behavior and preserves the same deterministic output under Transformers 4.57.6.

lavrenko changed pull request status to open 1 day ago

RaushanTurganbay

Transformers Community org about 15 hours ago

Nice

RaushanTurganbay changed pull request status to merged about 15 hours ago

Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.

Tap or paste here to upload images

· Sign up or log in to comment