Instructions to use cyttic/trocr-noise-bigram2 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use cyttic/trocr-noise-bigram2 with Transformers:

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

pipe = pipeline("image-text-to-text", model="cyttic/trocr-noise-bigram2")

# Load model directly
from transformers import AutoTokenizer, AutoModelForMultimodalLM

tokenizer = AutoTokenizer.from_pretrained("cyttic/trocr-noise-bigram2")
model = AutoModelForMultimodalLM.from_pretrained("cyttic/trocr-noise-bigram2", device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use cyttic/trocr-noise-bigram2 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "cyttic/trocr-noise-bigram2"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "cyttic/trocr-noise-bigram2",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/cyttic/trocr-noise-bigram2

SGLang

How to use cyttic/trocr-noise-bigram2 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 "cyttic/trocr-noise-bigram2" \
    --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": "cyttic/trocr-noise-bigram2",
		"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 "cyttic/trocr-noise-bigram2" \
        --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": "cyttic/trocr-noise-bigram2",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use cyttic/trocr-noise-bigram2 with Docker Model Runner:
```
docker model run hf.co/cyttic/trocr-noise-bigram2
```

trocr-noise-bigram2

This model is a fine-tuned version of cyttic/exp2-frozen-benyehuda-cont on an unknown dataset. It achieves the following results on the evaluation set:

Loss: 0.5353
Cer: 0.0311
Wer: 0.0865

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 2e-05
train_batch_size: 8
eval_batch_size: 8
seed: 42
gradient_accumulation_steps: 2
total_train_batch_size: 16
optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
lr_scheduler_type: linear
lr_scheduler_warmup_steps: 0.1
num_epochs: 2

Training results

Training Loss	Epoch	Step	Validation Loss	Cer	Wer
4.4701	0.1290	2000	2.0749	0.1892	0.3917
3.4282	0.2581	4000	1.5678	0.1288	0.2889
2.6796	0.3871	6000	1.2115	0.0937	0.2210
2.3398	0.5161	8000	1.0724	0.0778	0.1845
1.9402	0.6452	10000	0.9204	0.0661	0.1640
1.8243	0.7742	12000	0.8303	0.0570	0.1429
1.6875	0.9032	14000	0.7434	0.0509	0.1294
1.2807	1.0323	16000	0.6928	0.0457	0.1188
1.2310	1.1613	18000	0.6552	0.0419	0.1105
1.1186	1.2903	20000	0.6258	0.0380	0.1035
1.0514	1.4194	22000	0.6097	0.0380	0.1000
1.1760	1.5484	24000	0.5739	0.0346	0.0934
1.0901	1.6774	26000	0.5598	0.0339	0.0922
1.0063	1.8065	28000	0.5434	0.0328	0.0888
0.9992	1.9355	30000	0.5368	0.0309	0.0861
1.0274	2.0	31000	0.5353	0.0311	0.0865

Framework versions

Transformers 5.9.0
Pytorch 2.11.0+cu128
Datasets 5.0.1
Tokenizers 0.22.2

Downloads last month: 44

Safetensors

Model size

0.3B params

Tensor type

F32

Model tree for cyttic/trocr-noise-bigram2

Base model

cyttic/exp2-frozen-benyehuda-cont

Finetuned

(50)

this model

Finetunes

1 model