prepared dataset caching, other misc fixes (#665)

* prepared dataset caching, other misc fixes

* also don't load from disk cache unless explicit

examples/cerebras/qlora.yml

@@ -7,7 +7,7 @@ push_dataset_to_hub:
 datasets:
   - path: teknium/GPT4-LLM-Cleaned
     type: alpaca
-dataset_prepared_path: last_run_prepared
+dataset_prepared_path:
 val_set_size: 0.01
 adapter: qlora
 lora_model_dir:

examples/code-llama/13b/lora.yml

@@ -11,7 +11,7 @@ strict: false
 datasets:
   - path: mhenrichsen/alpaca_2k_test
     type: alpaca
-dataset_prepared_path: last_run_prepared
+dataset_prepared_path:
 val_set_size: 0.01
 output_dir: ./lora-out
 

examples/code-llama/13b/qlora.yml

@@ -11,7 +11,7 @@ strict: false
 datasets:
   - path: mhenrichsen/alpaca_2k_test
     type: alpaca
-dataset_prepared_path: last_run_prepared
+dataset_prepared_path:
 val_set_size: 0.01
 output_dir: ./qlora-out
 

examples/code-llama/34b/lora.yml

@@ -11,7 +11,7 @@ strict: false
 datasets:
   - path: mhenrichsen/alpaca_2k_test
     type: alpaca
-dataset_prepared_path: last_run_prepared
+dataset_prepared_path:
 val_set_size: 0.01
 output_dir: ./lora-out
 

examples/code-llama/34b/qlora.yml

@@ -11,7 +11,7 @@ strict: false
 datasets:
   - path: mhenrichsen/alpaca_2k_test
     type: alpaca
-dataset_prepared_path: last_run_prepared
+dataset_prepared_path:
 val_set_size: 0.01
 output_dir: ./qlora-out
 

examples/code-llama/7b/lora.yml

@@ -11,7 +11,7 @@ strict: false
 datasets:
   - path: mhenrichsen/alpaca_2k_test
     type: alpaca
-dataset_prepared_path: last_run_prepared
+dataset_prepared_path:
 val_set_size: 0.01
 output_dir: ./lora-out
 

examples/code-llama/7b/qlora.yml

@@ -11,7 +11,7 @@ strict: false
 datasets:
   - path: mhenrichsen/alpaca_2k_test
     type: alpaca
-dataset_prepared_path: last_run_prepared
+dataset_prepared_path:
 val_set_size: 0.01
 output_dir: ./qlora-out
 

examples/falcon/config-7b-lora.yml

@@ -12,7 +12,7 @@ push_dataset_to_hub:
 datasets:
   - path: teknium/GPT4-LLM-Cleaned
     type: alpaca:chat
-dataset_prepared_path: last_run_prepared
+dataset_prepared_path:
 val_set_size: 0.01
 adapter: lora
 lora_model_dir:

examples/falcon/config-7b-qlora.yml

@@ -18,7 +18,7 @@ datasets:
     data_files:
       - Chain-of-Thought/formatted_cot_data/gsm8k_train.json
     type: "alpaca:chat"
-dataset_prepared_path: last_run_prepared
+dataset_prepared_path:
 val_set_size: 0.01
 # enable QLoRA
 adapter: qlora

examples/falcon/config-7b.yml

@@ -12,7 +12,7 @@ push_dataset_to_hub:
 datasets:
   - path: teknium/GPT4-LLM-Cleaned
     type: alpaca:chat
-dataset_prepared_path: last_run_prepared
+dataset_prepared_path:
 val_set_size: 0.01
 adapter:
 lora_model_dir:

examples/gptj/qlora.yml

@@ -7,7 +7,7 @@ push_dataset_to_hub:
 datasets:
   - path: teknium/GPT4-LLM-Cleaned
     type: alpaca
-dataset_prepared_path: last_run_prepared
+dataset_prepared_path:
 val_set_size: 0.01
 adapter: qlora
 lora_model_dir:

examples/jeopardy-bot/config.yml

@@ -6,7 +6,7 @@ load_in_8bit: false
 datasets:
   - path: openaccess-ai-collective/jeopardy
     type: jeopardy
-dataset_prepared_path: last_run_prepared
+dataset_prepared_path:
 val_set_size: 0.02
 adapter:
 lora_model_dir:

examples/llama-2/gptq-lora.yml

@@ -15,7 +15,7 @@ hf_use_auth_token: true
 datasets:
   - path: mhenrichsen/alpaca_2k_test
     type: alpaca
-dataset_prepared_path: last_run_prepared
+dataset_prepared_path:
 val_set_size: 0.01
 adapter: lora
 lora_model_dir:

examples/llama-2/lora.yml

@@ -11,7 +11,7 @@ strict: false
 datasets:
   - path: mhenrichsen/alpaca_2k_test
     type: alpaca
-dataset_prepared_path: last_run_prepared
+dataset_prepared_path:
 val_set_size: 0.01
 output_dir: ./lora-out
 

examples/llama-2/qlora.yml

@@ -11,7 +11,7 @@ strict: false
 datasets:
   - path: mhenrichsen/alpaca_2k_test
     type: alpaca
-dataset_prepared_path: last_run_prepared
+dataset_prepared_path:
 val_set_size: 0.01
 output_dir: ./qlora-out
 

examples/llama-2/relora.yml

@@ -11,7 +11,7 @@ strict: false
 datasets:
   - path: teknium/GPT4-LLM-Cleaned
     type: alpaca
-dataset_prepared_path: last_run_prepared
+dataset_prepared_path:
 val_set_size: 0.01
 output_dir: ./relora-out
 

examples/llama-2/tiny-llama.yml

@@ -12,7 +12,7 @@ strict: false
 datasets:
   - path: mhenrichsen/alpaca_2k_test
     type: alpaca
-dataset_prepared_path: last_run_prepared
+dataset_prepared_path:
 val_set_size: 0.01
 output_dir: ./lora-out
 

examples/mistral/config.yml

@@ -11,7 +11,7 @@ strict: false
 datasets:
   - path: mhenrichsen/alpaca_2k_test
     type: alpaca
-dataset_prepared_path: last_run_prepared
+dataset_prepared_path:
 val_set_size: 0.01
 output_dir: ./out
 

examples/mpt-7b/config.yml

@@ -6,7 +6,7 @@ load_in_8bit: false
 datasets:
   - path: vicgalle/alpaca-gpt4
     type: alpaca
-dataset_prepared_path: last_run_prepared
+dataset_prepared_path:
 val_set_size: 0.02
 adapter:
 lora_model_dir:

examples/openllama-3b/config.yml

@@ -9,7 +9,7 @@ push_dataset_to_hub:
 datasets:
   - path: teknium/GPT4-LLM-Cleaned
     type: alpaca
-dataset_prepared_path: last_run_prepared
+dataset_prepared_path:
 val_set_size: 0.02
 adapter:
 lora_model_dir:

examples/openllama-3b/lora.yml

@@ -9,7 +9,7 @@ push_dataset_to_hub:
 datasets:
   - path: teknium/GPT4-LLM-Cleaned
     type: alpaca
-dataset_prepared_path: last_run_prepared
+dataset_prepared_path:
 val_set_size: 0.02
 adapter: lora
 lora_model_dir:

examples/openllama-3b/qlora.yml

@@ -9,7 +9,7 @@ push_dataset_to_hub:
 datasets:
   - path: teknium/GPT4-LLM-Cleaned
     type: alpaca
-dataset_prepared_path: last_run_prepared
+dataset_prepared_path:
 val_set_size: 0.01
 adapter: qlora
 lora_model_dir:

examples/phi/phi-ft.yml

@@ -13,7 +13,7 @@ datasets:
   - path: garage-bAInd/Open-Platypus
     type: alpaca
 
-dataset_prepared_path: last_run_prepared
+dataset_prepared_path:
 val_set_size: 0.05
 output_dir: ./phi-sft-out
 

examples/phi/phi-qlora.yml

@@ -13,7 +13,7 @@ datasets:
   - path: garage-bAInd/Open-Platypus
     type: alpaca
 
-dataset_prepared_path: last_run_prepared
+dataset_prepared_path:
 val_set_size: 0.05
 output_dir: ./phi-sft-out
 

examples/pythia-12b/config.yml

@@ -10,7 +10,7 @@ device_map: auto
 datasets:
   - path: vicgalle/alpaca-gpt4
     type: alpaca
-dataset_prepared_path: last_run_prepared
+dataset_prepared_path:
 val_set_size: 0.05
 adapter:
 lora_model_dir:

examples/pythia/lora.yml

@@ -4,7 +4,7 @@ load_in_8bit: true
 datasets:
   - path: teknium/GPT4-LLM-Cleaned
     type: alpaca
-dataset_prepared_path: last_run_prepared
+dataset_prepared_path:
 val_set_size: 0.05
 adapter: lora
 lora_model_dir:

examples/redpajama/config-3b.yml

@@ -7,7 +7,7 @@ load_in_8bit: false
 datasets:
   - path: vicgalle/alpaca-gpt4
     type: alpaca
-dataset_prepared_path: last_run_prepared
+dataset_prepared_path:
 val_set_size: 0.02
 adapter:
 lora_model_dir:

examples/replit-3b/config-lora.yml

@@ -5,7 +5,7 @@ load_in_8bit: false
 datasets:
   - path: vicgalle/alpaca-gpt4
     type: alpaca
-dataset_prepared_path: last_run_prepared
+dataset_prepared_path:
 val_set_size: 0.05
 adapter: lora
 lora_model_dir:

examples/xgen-7b/xgen-7b-8k-qlora.yml

@@ -16,7 +16,7 @@ datasets:
     data_files:
       - openassistant_best_replies_train.jsonl
     type: "completion"
-dataset_prepared_path: last_run_prepared
+dataset_prepared_path:
 val_set_size: 0.01
 # enable QLoRA
 adapter: qlora

src/axolotl/cli/__init__.py

@@ -51,7 +51,7 @@ def print_axolotl_text_art(suffix=None):
 
 
 def get_multi_line_input() -> Optional[str]:
-    print("Give me an instruction (Ctrl + D to finish): ")
+    print("Give me an instruction (Ctrl + D to submit): ")
     instruction = ""
     for line in sys.stdin:
         instruction += line  # pylint: disable=consider-using-join

src/axolotl/utils/data.py

@@ -122,7 +122,7 @@ def load_tokenized_prepared_datasets(
 
     if dataset:
         ...
-    elif any(prepared_ds_path.glob("*")):
+    elif cfg.dataset_prepared_path and any(prepared_ds_path.glob("*")):
         LOG.info(f"Loading prepared dataset from disk at {prepared_ds_path}...")
         dataset = load_from_disk(str(prepared_ds_path))
         LOG.info("Prepared dataset loaded from disk...")
@@ -357,7 +357,7 @@ def load_tokenized_prepared_datasets(
         if len(datasets) > 1:
             LOG.info("shuffle merged datasets")
             dataset = dataset.shuffle(seed=seed)
-        if cfg.local_rank == 0:
+        if cfg.local_rank == 0 and cfg.dataset_prepared_path:
             LOG.info(f"Saving merged prepared dataset to disk... {prepared_ds_path}")
             dataset.save_to_disk(prepared_ds_path)
             if cfg.push_dataset_to_hub:
@@ -425,7 +425,7 @@ def load_prepare_datasets(
 
         if dataset:
             ...
-        elif any(prepared_ds_path.glob("*")):
+        elif cfg.dataset_prepared_path and any(prepared_ds_path.glob("*")):
             LOG.info(
                 f"Loading prepared packed dataset from disk at {prepared_ds_path}..."
             )

src/axolotl/utils/tokenization.py

@@ -31,7 +31,8 @@ def check_example_labels(example, tokenizer, text_only=False):
         )
         colored_tokens.append(colored_token)
 
-    LOG.info(" ".join(colored_tokens))
+    delimiter = "" if text_only else " "
+    LOG.info(delimiter.join(colored_tokens))
     LOG.info("\n\n\n")
     print(" ".join(colored_tokens))