✅ Update test fixtures for no empty train streams

Signed-off-by: Evaline Ju <[email protected]>

tests/fixtures/__init__.py

@@ -115,7 +115,11 @@ def causal_lm_train_kwargs():
  "base_model": HFAutoCausalLM.bootstrap(
  model_name=CAUSAL_LM_MODEL, tokenizer_name=CAUSAL_LM_MODEL
  ),
- "train_stream": caikit.core.data_model.DataStream.from_iterable([]),
+ "train_stream": caikit.core.data_model.DataStream.from_iterable([
+ caikit_nlp.data_model.GenerationTrainRecord(
+ input="@foo what a cute dog!", output="no complaint"
+ ),
+ ]),
  "num_epochs": 0,
  "tuning_config": caikit_nlp.data_model.TuningConfig(
  num_virtual_tokens=8, prompt_tuning_init_text="hello world"
@@ -149,7 +153,11 @@ def seq2seq_lm_train_kwargs():
  "base_model": HFAutoSeq2SeqLM.bootstrap(
  model_name=SEQ2SEQ_LM_MODEL, tokenizer_name=SEQ2SEQ_LM_MODEL
  ),
- "train_stream": caikit.core.data_model.DataStream.from_iterable([]),
+ "train_stream": caikit.core.data_model.DataStream.from_iterable([
+ caikit_nlp.data_model.GenerationTrainRecord(
+ input="@foo what a cute dog!", output="no complaint"
+ ),
+ ]),
  "num_epochs": 0,
  "tuning_config": caikit_nlp.data_model.TuningConfig(
  num_virtual_tokens=16, prompt_tuning_init_text="hello world"

tests/modules/text_generation/test_peft_prompt_tuning.py

@@ -258,6 +258,19 @@ def test_prompt_output_types(causal_lm_train_kwargs):
  assert model
 
 
+def test_error_empty_stream(causal_lm_train_kwargs):
+ patch_kwargs = {
+ "num_epochs": 1,
+ "verbalizer": "Tweet text : {{input}} Label : ",
+ "train_stream": caikit.core.data_model.DataStream.from_iterable([]),
+ }
+ causal_lm_train_kwargs.update(patch_kwargs)
+ with pytest.raises(ValueError):
+ caikit_nlp.modules.text_generation.PeftPromptTuning.train(
+ **causal_lm_train_kwargs
+ )
+
+
 ### Implementation details
 # These tests can probably be removed and tested directly through .save() once
 # full seq2seq support is completed and verified.