Example of using Hugging Face streaming dataset
Based on//arxiv.org/pdf/2408.06537
Recipes
Choose a task below. Each recipe documents its working directory, model/data
Use Claude Code with local Qwen3.8 served by Eole
This experimental recipe connects the Claude Code CLI to Eole's
Use Codex with local Qwen served by Eole
This experimental recipe targets the Codex CLI and Linux desktop app through
DeepSeek-OCR
Run from the repository root in a CUDA-enabled Eole environment. The examples
EuroLLM model and legacy Gradio example
For the current bundled translation UI, start with the
Train a language model on FineWeb 10BT
Run from recipes/fineweb10B in your Eole environment. This is a GPT-2-sized
OpenAI GPT2
https://huggingface.co/openai-community/gpt2
Run a supported Hugging Face model directly
Run from the repository root in a CUDA-enabled PyTorch environment:
HunyuanOCR
Set environment variables
Coding correctness with LiveCodeBench
Generate one greedy Python solution per problem through Eole's
Llama2
---
Llama3
---
Mistral inference
Run from the repository root in your CUDA-enabled Eole environment. This example
Mixtral
---
Evaluation with MMLU dataset
How to run
Model conversion smoke tests
This runner converts selected Hugging Face models, generates a short continuation,
NLLB conversion and translation
Run from this recipe directory: cd recipes/nllb. This is an English-to-German
Pixtral
Only the mistral-community/pixtral-12b version is supported right now. The official mistralai/Pixtral-12B-2409 config is rather incomplete.
Qwen3.5: text, vision, streaming, and serving
Run from the repository root. Follow the installation guide
Qwen3.8-27B and native MTP speculative decoding
4 items
Local Knowledge (RAG) workspace
1 item
RL fine-tuning recipes
Fine-tune an existing checkpoint with on-policy REINFORCE and a registered
Eole on RunPod
1 item
Scoring Recipes
3 items
Serve a model with Eole
Run all commands from the repository root. The example serves
Translate in the Eole workspace
A dedicated translation view shares Eole's local workspace, model server,
Eole local workspace
eole serve now includes a browser interface at /. No Gradio, Node runtime,
Whisper
NOTE: To make your life easier, run these commands from the recipe directory (here recipes/whisper).
Train a language model on WikiText-103
Run from recipes/wiki_103 in your Eole environment. Preparation requires
Translation WMT17 en-de
---
TowerInstruct (Mistral)
---
TowerInstruct-(Llama2)
---
Llama3
---