Instructions to use openeurollm/Prelude-SFT-Excellent-Instructions with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use openeurollm/Prelude-SFT-Excellent-Instructions with Transformers:

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

pipe = pipeline("text-generation", model="openeurollm/Prelude-SFT-Excellent-Instructions")

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoModel
model = AutoModel.from_pretrained("openeurollm/Prelude-SFT-Excellent-Instructions", device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use openeurollm/Prelude-SFT-Excellent-Instructions with vLLM:

Install from pip and serve model

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

Use Docker

docker model run hf.co/openeurollm/Prelude-SFT-Excellent-Instructions

SGLang

How to use openeurollm/Prelude-SFT-Excellent-Instructions 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 "openeurollm/Prelude-SFT-Excellent-Instructions" \
    --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": "openeurollm/Prelude-SFT-Excellent-Instructions",
		"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 "openeurollm/Prelude-SFT-Excellent-Instructions" \
        --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": "openeurollm/Prelude-SFT-Excellent-Instructions",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use openeurollm/Prelude-SFT-Excellent-Instructions with Docker Model Runner:
```
docker model run hf.co/openeurollm/Prelude-SFT-Excellent-Instructions
```

Prelude SFT — Excellent Instructions

Prelude fine-tuned on Excellent-SFT for two epochs with explicit thinking-mode conditioning. Reasoning examples receive /think or no instruction in equal proportions before filtering; direct-answer examples receive /no_think. The implementation is described in post-training PR #68.

Reproduction details

Item	Value
Training repository	OpenEuroLLM/post-training
Training code reference	PR #68; use `main` after merge
Thinking-conditioning implementation	PR #68: condition Qwen training targets on thinking mode
Experiment and evaluation discussion	Taskboard #442
Base model	openeurollm/prelude-anneal300b-256k-rope20m-cont
Base revision	`47a42eb598c0417691b8054e03dac172241bca6b`
Training data	openeurollm/OpenEuroLLM-Excellent-SFT
Data revision	`c4ee55804f256c5f8cc4a6ef3fdecbe4a4ea95e0`
Tokenizer	openeurollm/tokenizer-256k
Tokenizer revision	`94bae05cf351c4feb435d6a73c2aa8edfd88ae04`
Chat template	Thinking-conditioned Qwen-style template bundled with this checkpoint
Thinking instructions	`thinking_instruction_probability: 0.5`; half of reasoning examples receive `/think`, the remainder none; direct-answer examples receive `/no_think`
Training duration	2 epochs; 55,892 steps
Maximum sequence length	32,768 tokens; packing enabled
Supervision	Assistant-only loss
Partially truncated supervised responses	Dropped (`truncated_span_action: drop`)
Optimizer	AdamW; β₁=0.9, β₂=0.95, ε=1e-8; weight decay=0
Learning rate	8e-5; linear decay; 3% warmup
Effective batch size	32 packed sequences; per-device batch size 1
Gradient clipping	1.0
Precision and kernels	BF16; FlashAttention 3; Liger enabled
Distributed training	DeepSpeed ZeRO-2; communication overlap disabled
Hardware	32 NVIDIA H100 GPUs across 8 nodes on MareNostrum 5 (MN5)
Gradient checkpointing	Enabled, non-reentrant
Seeds	Training and data: 42
Processed input tokens	58,365,718,851, across both epochs
Processed supervised targets	54,730,454,404, across both epochs

Use the bundled tokenizer and chat template for inference. enable_thinking controls the generation prefix; the separate, optional thinking_instruction argument accepts "think", "no_think", or None.

This repository contains inference weights and tokenizer files, not optimizer or resumable training state. Final token counters are included in token_counts.json. Consult the base-model and dataset cards for applicable licenses; no separate checkpoint release license has been assigned.

Downloads last month: -; Downloads are not tracked for this model. How to track

Model tree for openeurollm/Prelude-SFT-Excellent-Instructions

Base model

openeurollm/prelude-anneal300b-256k-rope20m-cont

Finetuned

(1)

this model

openeurollm
/

Prelude-SFT-Excellent-Instructions

Prelude SFT — Excellent Instructions

Reproduction details

Model tree for openeurollm/Prelude-SFT-Excellent-Instructions

Dataset used to train openeurollm/Prelude-SFT-Excellent-Instructions