Instructions to use brozonoyer/relay-fastdllm-v2-c40m60-relay-sg-step200 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use brozonoyer/relay-fastdllm-v2-c40m60-relay-sg-step200 with Transformers:

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

pipe = pipeline("text-generation", model="brozonoyer/relay-fastdllm-v2-c40m60-relay-sg-step200", trust_remote_code=True)
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoModelForCausalLM
model = AutoModelForCausalLM.from_pretrained("brozonoyer/relay-fastdllm-v2-c40m60-relay-sg-step200", trust_remote_code=True, device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use brozonoyer/relay-fastdllm-v2-c40m60-relay-sg-step200 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "brozonoyer/relay-fastdllm-v2-c40m60-relay-sg-step200"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "brozonoyer/relay-fastdllm-v2-c40m60-relay-sg-step200",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/brozonoyer/relay-fastdllm-v2-c40m60-relay-sg-step200

SGLang

How to use brozonoyer/relay-fastdllm-v2-c40m60-relay-sg-step200 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 "brozonoyer/relay-fastdllm-v2-c40m60-relay-sg-step200" \
    --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": "brozonoyer/relay-fastdllm-v2-c40m60-relay-sg-step200",
		"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 "brozonoyer/relay-fastdllm-v2-c40m60-relay-sg-step200" \
        --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": "brozonoyer/relay-fastdllm-v2-c40m60-relay-sg-step200",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use brozonoyer/relay-fastdllm-v2-c40m60-relay-sg-step200 with Docker Model Runner:
```
docker model run hf.co/brozonoyer/relay-fastdllm-v2-c40m60-relay-sg-step200
```

RELAY (sg) adaptation of Fast-dLLM v2 1.5B (c40m60, step 200)

Released alongside the paper Learned Relay Representations for Forward-Thinking Discrete Diffusion Models. Reproduces the RELAY (sg) row of Table 2.

Quick start

from transformers import AutoModelForCausalLM, AutoTokenizer

repo = "brozonoyer/relay-fastdllm-v2-c40m60-relay-sg-step200"
tokenizer = AutoTokenizer.from_pretrained(repo, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(repo, trust_remote_code=True)

config.json ships with use_relay=True and relay_layer=-1; the safetensors shard contains a model.relay_layer_norm.{weight,bias} tensor that the bundled modeling.py instantiates and consumes inside the 2-step relay forward (paper Algorithm 1).

Reproduce the Table 2 numbers with EvalPlus

From the public release of the training code (e.g. relay/fast-dllm-v2/v2/):

mkdir -p evalplus_results

# HumanEval+
python scripts/generate_evalplus_jsonl.py \
  --model_path brozonoyer/relay-fastdllm-v2-c40m60-relay-sg-step200 \
  --dataset humaneval --use_carry --threshold 0.85 \
  --output_jsonl evalplus_results/relay-sg_humaneval.jsonl
evalplus.evaluate --dataset humaneval --samples evalplus_results/relay-sg_humaneval.jsonl

# MBPP+
python scripts/generate_evalplus_jsonl.py \
  --model_path brozonoyer/relay-fastdllm-v2-c40m60-relay-sg-step200 \
  --dataset mbpp --use_carry --threshold 0.85 \
  --output_jsonl evalplus_results/relay-sg_mbpp.jsonl
evalplus.evaluate --dataset mbpp --samples evalplus_results/relay-sg_mbpp.jsonl

--use_carry enables the 2-step relay-state carry at inference; both checkpoints were trained with relay on, so it is required at eval to match the reported numbers.

Training

200 optimizer steps starting from Efficient-Large-Model/Fast_dLLM_v2_1.5B on the 60k OpenCodeInstruct + OpenMathInstruct-2 c40m60 mixture (24 000 code + 36 000 math rows). Effective batch size 32, learning rate 5e-6, BD block 32 / sub-block 8, threshold 0.85. The RELAY (sg) variant uses bptt_use_relay=1, bptt_stop_grad_h_s=1.

Full training command (from relay/fast-dllm-v2/v2/):

USE_RELAY=1 BPTT_STOP_GRAD_H_S=1 \
  sbatch train_scripts/finetune_opencode_openmath_bptt.sbatch

See the public release of the training code for the full pipeline.

Downloads last month: 9

Safetensors

Model size

2B params

Tensor type

BF16

Model tree for brozonoyer/relay-fastdllm-v2-c40m60-relay-sg-step200

Base model

Qwen/Qwen2.5-1.5B

Finetuned

Qwen/Qwen2.5-1.5B-Instruct

Finetuned

Efficient-Large-Model/Fast_dLLM_v2_1.5B

Finetuned

(2)

this model