eb-gen : An EasyBuild Easyconfig Generator, based on a fine tuned Qwen2.5-Coder-7B-Instruct .

ollama run kuikuisven/eb-gen:7b

curl http://localhost:11434/api/chat \
  -d '{
    "model": "kuikuisven/eb-gen:7b",
    "messages": [{"role": "user", "content": "Hello!"}]
  }'

from ollama import chat

response = chat(
    model='kuikuisven/eb-gen:7b',
    messages=[{'role': 'user', 'content': 'Hello!'}],
)
print(response.message.content)

import ollama from 'ollama'

const response = await ollama.chat({
  model: 'kuikuisven/eb-gen:7b',
  messages: [{role: 'user', content: 'Hello!'}],
})
console.log(response.message.content)

Details

Updated 1 week ago

1 week ago

1c277f357c15 · 15GB ·

model

archqwen2

parameters7.62B

quantizationF16

15GB

template

<|im_start|>system {{ .System }}<|im_end|> <|im_start|>user {{ .Prompt }}<|im_end|> <|im_start|>assi

116B

system

You are an EasyBuild easyconfig generator. Given software metadata, produce a complete .eb easyconfi

233B

params

{ "stop": [ "<|im_end|>" ], "temperature": 0.1, "top_p": 0.9 }

64B

Readme

Fine-tuning an EasyBuild Easyconfig Generator

QLoRA fine-tuning of Qwen2.5-Coder-7B-Instruct on 11K EasyBuild easyconfig files, producing a model that generates valid .eb recipes from software metadata.

Overview


Base model	Qwen2.5-Coder-7B-Instruct
Method	QLoRA (4-bit NF4, LoRA rank 64)
Training data	10,453 examples from 11K `.eb` files
Trainable params	~161M / 4.5B (3.6%)
Hardware	NVIDIA V100 16GB
Training time	~2-3 hours (3 epochs)