:white_check_mark: Fix imports

src/handlers.py

@@ -2,29 +2,12 @@ import logging
 
 from transformers import PreTrainedTokenizer
 
-from constants import MAX_ATTEMPTS
-from params import ReducerParams
-from shared import all_tokens
-from shared import text
-from shared import token_id_predictions
-from shared import tokenizer
-from utils import get_start_and_whitespace_tokens
+from src.params import ReducerParams
+from src.shared import all_tokens
 
 logger = logging.getLogger(__name__)
 
 
-def lm_is_correct(params: ReducerParams) -> bool:
-    # NOTE: out of range if remaining attempts is 0
-    if params.remaining_attempts < 1:
-        return False
-
-    idx = MAX_ATTEMPTS - params.remaining_attempts
-
-    current_guess = params.lm_guesses[]
-    current_target = all_tokens[params.current_word_index]
-    return current_guess == current_target
-
-
 def get_current_text(params: ReducerParams, tokenizer: PreTrainedTokenizer):
     return tokenizer.decode(all_tokens[: params.word_number])
 

src/interface.py

@@ -1,9 +1,9 @@
 import gradio as gr
 
-import reducer
+from src import reducer
+from src import shared
 from src.constants import MAX_ATTEMPTS
-from src.state import STATE
-from src.state import tokenizer
+from src.constants import STARTING_INDEX
 
 
 def build_demo():
@@ -23,7 +23,7 @@ def build_demo():
             )
         with gr.Row():
             prompt_text = gr.Textbox(
-                value=tokenizer.decode(all_tokens[: STATE.current_word_index]),
+                value=shared.all_tokens[:STARTING_INDEX],
                 label="Context",
                 interactive=False,
             )

src/predictions.py

@@ -6,11 +6,10 @@ predicted tokens.
 import torch
 from transformers import AutoModelForCausalLM
 from transformers import AutoTokenizer
-from transformers import PreTrainedModel
 from transformers import PreTrainedTokenizer
 
-from constants import MAX_ATTEMPTS
-from text import get_text
+from src.constants import MAX_ATTEMPTS
+from src.text import get_text
 
 tokenizer = AutoTokenizer.from_pretrained("gpt2")
 

src/reducer.py

@@ -1,16 +1,15 @@
 import logging
 
-from handlers import handle_lm_win
-from handlers import handle_next_attempt
-from handlers import handle_next_word
-from handlers import handle_no_input
-from handlers import handle_out_of_attempts
-from handlers import handle_player_win
-from handlers import handle_tie
-from handlers import lm_is_correct
-from params import ReducerParams
-from shared import tokenizer, token_id_predictions, token_predictions
-from utils import guess_is_correct
+from src.handlers import handle_lm_win
+from src.handlers import handle_next_attempt
+from src.handlers import handle_next_word
+from src.handlers import handle_no_input
+from src.handlers import handle_out_of_attempts
+from src.handlers import handle_player_win
+from src.handlers import handle_tie
+from src.params import ReducerParams
+from src.utils import guess_is_correct
+from src.utils import lm_is_correct
 
 logger = logging.getLogger(__name__)
 

src/shared.py

@@ -1,9 +1,9 @@
 from transformers import AutoTokenizer
 
-from predictions import make_predictions
-from text import get_text
+from src.predictions import make_predictions
+from src.text import get_text
 
 tokenizer = AutoTokenizer.from_pretrained("gpt2")
-token_id_predictions, token_predictions = make_predictions()
+token_id_predictions, token_predictions = make_predictions(tokenizer)
 text = get_text()
 all_tokens = tokenizer.encode(text)

src/utils.py

@@ -2,7 +2,9 @@ import logging
 
 from transformers import PreTrainedTokenizer
 
-from params import ReducerParams
+from src.constants import MAX_ATTEMPTS
+from src.params import ReducerParams
+from src.shared import token_id_predictions
 
 logger = logging.getLogger(__name__)
 
@@ -19,12 +21,26 @@ def get_start_and_whitespace_tokens(
     return predicted_token_start, predicted_token_whitespace
 
 
+def lm_is_correct(params: ReducerParams) -> bool:
+    # NOTE: out of range if remaining attempts is 0
+    if params.remaining_attempts < 1:
+        return False
+
+    idx = MAX_ATTEMPTS - params.remaining_attempts
+
+    current_guess = token_id_predictions[params.word_number][1][idx]
+    current_target = token_id_predictions[params.word_number][0]
+
+    return current_guess == current_target
+
+
 def guess_is_correct(params: ReducerParams, current_target: int, tokenizer: PreTrainedTokenizer) -> bool:
     """
     We check if the predicted token or a corresponding one with a leading whitespace
     matches that of the next token
     """
     logger.debug("Next token: '{}'".format(tokenizer.convert_ids_to_tokens([current_target])))
-    predicted_token_start, predicted_token_whitespace = get_start_and_whitespace_tokens(text)
+
+    predicted_token_start, predicted_token_whitespace = get_start_and_whitespace_tokens(params.guess_field)
     logger.debug(tokenizer.convert_ids_to_tokens([predicted_token_start, predicted_token_whitespace]))
     return current_target in (predicted_token_start, predicted_token_whitespace)