isaacmg

""" Use torchMoji to predict emojis from a single text input
"""

from __future__ import print_function, division, unicode_literals
import examples.example_helper
import json
import csv
import argparse
import numpy as np

isaacmg

Serving a TTS model with Seldon

In this walkthrough you will learn to serve a multi-part PyTorch model with Seldon. +* +*

s2i build . seldonio/seldon-core-s2i-python3:0.10 kubeflow/tts_encoder

docker run --name "EncoderServe" --rm -p 5000:5000 kubeflow/tts_encoder

isaacmg

!allennlp train babi_train_meta.jsonnet -s /tmp/serialization_dir --include-package allennlp.training.metatrainer

isaacmg

class AttendDiagnose(nn.Module):
  def __init__(self, number_measurements, filter_number):
    super().__init__()
    self.d_model = filter_number*number_measurements
    self.embedding_conv = nn.Conv1d(number_measurements, filter_number*number_measurements, 1) 
    self.pe = PositionalEncoding(filter_number*number_measurements)
    # embed_dim and attention_heads  
    self.masked_attn = nn.modules.activation.MultiheadAttention(filter_number*number_measurements, 8)
    self.norm = nn.modules.normalization.LayerNorm(self.d_model)
    self.final_layer = nn.Linear(self.d_model, 1)

isaacmg

class SimpleTransformer(torch.nn.Module):
    def __init__(self, n_time_series, seq_len, d_model=128):
        super().__init__()
        self.dense_shape = torch.nn.Linear(n_time_series, d_model)
        self.pe = SimplePositionalEncoding(d_model)
        self.transformer = Transformer(d_model, nhead=8)
        self.final_layer = torch.nn.Linear(d_model, 1)
        self.sequence_size = seq_len
    def forward(self, x, t, tgt_mask, src_mask=None):
        if src_mask:

isaacmg

# ML Core 
scikit-learn==0.22.1
pandas==1.01
torch==1.40
tb-nightly==2.2.0
# Graphs and param management
seaborn==0.10.0
future==0.18.2
wandb==0.8.25
# GCP dependencies

isaacmg

@TrainerBase.register('metatrainer')
class MetaTrainer(Trainer):
    def __init__(self,
                 model: Model,
                 meta_model: MetaModel,
                 optimizer: torch.optim.Optimizer,
                 iterator: DataIterator,
                 train_datasets: List[Iterable[Instance]],
                 validation_datasets: Optional[Iterable[Instance]] = None,

isaacmg

FROM pytorch/pytorch:1.4-cuda10.1-cudnn7-devel
COPY requirements.txt /tmp/
RUN pip install -r /tmp/requirements.txt
ARG url
RUN git clone -n https://github.com/example/example_repo
RUN git checkout 543231

isaacmg

def train_epoch_loop(data_loader:DataLoader, opt:torch.optim, model:PyTorchForecast, takes_target:bool, forward_params={})
      i = 0
      running_loss = 0.0
      for src, trg in data_loader:
          opt.zero_grad()
          # Convert to CPU/GPU/TPU 
          src = src.to(model.device)
          trg = trg.to(model.device)
          # TODO figure how to avoid
          if takes_target:

isaacmg

label = ade_eval
model_type = spert
model_path = data/models/ade
tokenizer_path = data/models/ade
dataset_path = data/datasets/conll04/conll04_test.json
types_path = data/datasets/conll04/conll04_types.json
eval_batch_size = 1
rel_filter_threshold = 0.4
size_embedding = 25
prop_drop = 0.1