XLsim_en-de

{AnanyaCoder/XLsim_en-de}

XLSim: MT Evaluation Metric based on Siamese Architecture

XLsim is a supervised reference-based metric that regresses on human scores provided by WMT (2017-2022). Using a cross-lingual language model XLM-RoBERTa-base [ https://huggingface.co/xlm-roberta-base ] , we train a supervised model using a Siamese network architecture with CosineSimilarityLoss.

<!--- Describe your model here -->

Usage (Sentence-Transformers)

Using this model becomes easy when you have sentence-transformers installed:

pip install -U sentence-transformers

Then you can use the model like this:

from sentence_transformers import SentenceTransformer,losses, models, util

metric_model = SentenceTransformer('{MODEL_NAME}')

#Compute embedding for both lists
mt_samples = ['This is a mt sentence1','This is a mt sentence2']
ref_samples = ['This is a ref sentence1','This is a ref sentence2']

mtembeddings = metric_model.encode(mt_samples, convert_to_tensor=True)
refembeddings = metric_model.encode(ref_samples, convert_to_tensor=True)

#Compute cosine-similarities
cosine_scores_refmt = util.cos_sim(mtembeddings, refembeddings)
#cosine_scores_srcmt = util.cos_sim(mtembeddings, srcembeddings) #qe
metric_model_scores = []
for i in range(len(mt_samples)):
    metric_model_scores.append(cosine_scores_refmt[i][i].tolist())

scores = metric_model_scores

Evaluation Results

<!--- Describe how your model was evaluated -->

For an automated evaluation of this model, see: WMT23 Metrics Shared Task findings

Training

The model was trained with the parameters:

DataLoader:

torch.utils.data.dataloader.DataLoader of length 6625 with parameters:

{'batch_size': 16, 'sampler': 'torch.utils.data.sampler.RandomSampler', 'batch_sampler': 'torch.utils.data.sampler.BatchSampler'}

Loss:

sentence_transformers.losses.CosineSimilarityLoss.CosineSimilarityLoss

Parameters of the fit()-Method:

{
    "epochs": 4,
    "evaluation_steps": 1000,
    "evaluator": "sentence_transformers.evaluation.EmbeddingSimilarityEvaluator.EmbeddingSimilarityEvaluator",
    "max_grad_norm": 1,
    "optimizer_class": "<class 'torch.optim.adamw.AdamW'>",
    "optimizer_params": {
        "lr": 2e-05
    },
    "scheduler": "WarmupLinear",
    "steps_per_epoch": null,
    "warmup_steps": 2650,
    "weight_decay": 0.01
}

Full Model Architecture

SentenceTransformer(
  (0): Transformer({'max_seq_length': 512, 'do_lower_case': False}) with Transformer model: XLMRobertaModel 
  (1): Pooling({'word_embedding_dimension': 768, 'pooling_mode_cls_token': False, 'pooling_mode_mean_tokens': True, 'pooling_mode_max_tokens': False, 'pooling_mode_mean_sqrt_len_tokens': False})
)

Citing & Authors

<!--- Describe where people can find more information -->

MEE4 and XLsim : IIIT HYD’s Submissions’ for WMT23 Metrics Shared Task (Mukherjee & Shrivastava, WMT 2023)