File size: 5,098 Bytes
db744c6 e3278c7 db744c6 e3278c7 db744c6 e3278c7 db744c6 e3278c7 db744c6 e3278c7 db744c6 e3278c7 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 |
---
library_name: transformers
tags: []
---
# Model Card for Model ID
<!-- Provide a quick summary of what the model is/does. -->
Classification model finetuned for prompt-model routing based on code prompt difficulty.
## Model Details
Base model: answerdotai/ModernBERT-base
### Model Description
<!-- Provide a longer summary of what this model is. -->
- **Developed by:** [Christian @ Prime Intellect]
- **Finetuned from model:** [answerdotai/ModernBERT-base]
## How to Get Started with the Model
<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
```
from transformers import pipeline
import torch
# Load the model
classifier = pipeline(
"text-classification",
model="cdreetz/modern-bert-router",
device=0 if torch.cuda.is_available() else -1
)
# Test easy problem
easy_problem = """
Write a function that returns the sum of two numbers.
Example:
Input: add(2, 3)
Output: 5
"""
# Test hard problem
hard_problem = """
Given a binary tree, find the maximum path sum. The path may start and end at any node in the tree.
A path is defined as any sequence of nodes from some starting node to any node in the tree along the parent-child connections.
The path must contain at least one node and does not need to go through the root.
Example:
Input: root = [1,2,3]
Output: 6 (2 -> 1 -> 3)
"""
# Run predictions
result_easy = classifier(easy_problem)[0]
result_hard = classifier(hard_problem)[0]
print("Easy Problem:")
print(f" Difficulty: {result_easy['label']}")
print(f" Confidence: {result_easy['score']:.2%}\n")
print("Hard Problem:")
print(f" Difficulty: {result_hard['label']}")
print(f" Confidence: {result_hard['score']:.2%}")
```
## Training Details
```
# /// script
# dependencies = [
# "chatan",
# "transformers",
# "datasets",
# "torch",
# "accelerate",
# "scikit-learn",
# "triton",
# "huggingface_hub"
# ]
# ///
import os
import asyncio
import chatan as ch
from datasets import Dataset as hf_ds
from transformers import AutoTokenizer, AutoModelForSequenceClassification, Trainer, TrainingArguments
import numpy as np
from sklearn.metrics import accuracy_score, f1_score
import torch
import triton
from huggingface_hub import login
#torch._dynamo.config.suppress_errors = True
login()
async def create_dataset():
gen = ch.async_generator("openai", os.getenv("OPENAI_API_KEY"))
ds = ch.async_dataset({
"difficulty": ch.sample.choice(["easy", "hard"]),
"text": gen("write a coding problem of difficulty {difficulty}")
})
df = await ds.generate(n=1000, max_concurrent_rows=500)
df['labels'] = df['difficulty'].map({"easy": 0, "hard": 1})
dataset = hf_ds.from_pandas(df[['text', 'labels']])
return dataset.train_test_split(test_size=0.2, seed=42)
def train(dataset):
tokenizer = AutoTokenizer.from_pretrained("answerdotai/ModernBERT-base")
model = AutoModelForSequenceClassification.from_pretrained(
"answerdotai/ModernBERT-base",
num_labels=2,
label2id={"easy": 0, "hard": 1},
id2label={0: "easy", 1: "hard"},
problem_type="single_label_classification"
)
def tokenize(examples):
return tokenizer(
examples['text'],
padding='max_length',
truncation=True,
max_length=512
)
tokenized = dataset.map(tokenize, batched=True, remove_columns=['text'])
tokenized.set_format('torch')
training_args = TrainingArguments(
output_dir="./modernbert-router",
eval_strategy="epoch",
num_train_epochs=3,
per_device_train_batch_size=16,
learning_rate=2e-5,
save_strategy="epoch",
load_best_model_at_end=True
)
trainer = Trainer(
model=model,
args=training_args,
train_dataset=tokenized["train"],
eval_dataset=tokenized["test"],
processing_class=tokenizer,
compute_metrics=lambda p: {
"accuracy": accuracy_score(p.label_ids, np.argmax(p.predictions, axis=1)),
"f1": f1_score(p.label_ids, np.argmax(p.predictions, axis=1))
}
)
print("starting training")
trainer.train()
model.save_pretrained("./modernbert-router")
tokenizer.save_pretrained("./modernbert-router")
print("donezo")
#model = AutoModelForSequenceClassification.from_pretrained("./modernbert-router")
#tokenizer = AutoTokenizer.from_pretrained("./modernbert-router")
#model.push_to_hub("cdreetz/modern-bert-router")
#tokenizer.push_to_hub("cdreetz/modern-bert-router")
if __name__ == "__main__":
dataset = asyncio.run(create_dataset())
train(dataset)
```
## Citation [optional]
<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
**BibTeX:**
```
@software{modern-bert-router,
author = {Reetz, Christian},
title = {ModernBERTRouter},
url = {https://huggingface.co/cdreetz/modern-bert-router/},
year = {2025}
}
```
|