Implement stopping string search in string space (#2847)

README.md

@@ -350,4 +350,4 @@ The presets that are included by default are the result of a contest that receiv
 
 - Gradio dropdown menu refresh button, code for reloading the interface: https://github.com/AUTOMATIC1111/stable-diffusion-webui
 - Godlike preset: https://github.com/KoboldAI/KoboldAI-Client/wiki/Settings-Presets
-- Code for early stopping in chat mode, code for some of the sliders: https://github.com/PygmalionAI/gradio-ui/
+- Code for some of the sliders: https://github.com/PygmalionAI/gradio-ui/

modules/callbacks.py

@@ -9,33 +9,6 @@ import transformers
 import modules.shared as shared
 
 
-class _SentinelTokenStoppingCriteria(transformers.StoppingCriteria):
-
-    def __init__(self, sentinel_token_ids: list, starting_idx: int):
-        transformers.StoppingCriteria.__init__(self)
-        self.sentinel_token_ids = sentinel_token_ids
-        self.starting_idx = starting_idx
-        self.shortest = min([x.shape[-1] for x in sentinel_token_ids])
-
-    def __call__(self, input_ids: torch.LongTensor, _scores: torch.FloatTensor) -> bool:
-        for sample in input_ids:
-            trimmed_sample = sample[self.starting_idx:]
-            trimmed_len = trimmed_sample.shape[-1]
-            if trimmed_len < self.shortest:
-                continue
-
-            for sentinel in self.sentinel_token_ids:
-                sentinel_len = sentinel.shape[-1]
-                if trimmed_len < sentinel_len:
-                    continue
-
-                window = trimmed_sample[-sentinel_len:]
-                if torch.all(torch.eq(sentinel, window)):
-                    return True
-
-        return False
-
-
 class Stream(transformers.StoppingCriteria):
     def __init__(self, callback_func=None):
         self.callback_func = callback_func

modules/chat.py

@@ -1,4 +1,3 @@
-import ast
 import base64
 import copy
 import functools
@@ -144,40 +143,10 @@ def get_stopping_strings(state):
             f"\n{state['name2']}:"
         ]
 
-    stopping_strings += ast.literal_eval(f"[{state['custom_stopping_strings']}]")
-    return stopping_strings
-
-
-def extract_message_from_reply(reply, state):
-    next_character_found = False
-    stopping_strings = get_stopping_strings(state)
-
     if state['stop_at_newline']:
-        lines = reply.split('\n')
+        stopping_strings.append("\n")
-        reply = lines[0].strip()
-        if len(lines) > 1:
-            next_character_found = True
-    else:
-        for string in stopping_strings:
-            idx = reply.find(string)
-            if idx != -1:
-                reply = reply[:idx]
-                next_character_found = True
 
-        # If something like "\nYo" is generated just before "\nYou:"
+    return stopping_strings
-        # is completed, trim it
-        if not next_character_found:
-            for string in stopping_strings:
-                for j in range(len(string) - 1, 0, -1):
-                    if reply[-j:] == string[:j]:
-                        reply = reply[:-j]
-                        break
-                else:
-                    continue
-
-                break
-
-    return reply, next_character_found
 
 
 def chatbot_wrapper(text, history, state, regenerate=False, _continue=False, loading_message=True):
@@ -191,7 +160,6 @@ def chatbot_wrapper(text, history, state, regenerate=False, _continue=False, loa
     # Defining some variables
     just_started = True
     visible_text = None
-    eos_token = '\n' if state['stop_at_newline'] else None
     stopping_strings = get_stopping_strings(state)
 
     # Preparing the input
@@ -231,11 +199,10 @@ def chatbot_wrapper(text, history, state, regenerate=False, _continue=False, loa
     cumulative_reply = ''
     for i in range(state['chat_generation_attempts']):
         reply = None
-        for j, reply in enumerate(generate_reply(prompt + cumulative_reply, state, eos_token=eos_token, stopping_strings=stopping_strings, is_chat=True)):
+        for j, reply in enumerate(generate_reply(prompt + cumulative_reply, state, stopping_strings=stopping_strings, is_chat=True)):
             reply = cumulative_reply + reply
 
             # Extract the reply
-            reply, next_character_found = extract_message_from_reply(reply, state)
             visible_reply = re.sub("(<USER>|<user>|{{user}})", state['name1'], reply)
             visible_reply = apply_extensions("output", visible_reply)
 
@@ -262,9 +229,6 @@ def chatbot_wrapper(text, history, state, regenerate=False, _continue=False, loa
                 if state['stream']:
                     yield output
 
-            if next_character_found:
-                break
-
         if reply in [None, cumulative_reply]:
             break
         else:
@@ -281,7 +245,6 @@ def impersonate_wrapper(text, start_with, state):
 
     # Defining some variables
     cumulative_reply = ''
-    eos_token = '\n' if state['stop_at_newline'] else None
     prompt = generate_chat_prompt('', state, impersonate=True)
     stopping_strings = get_stopping_strings(state)
 
@@ -289,16 +252,12 @@ def impersonate_wrapper(text, start_with, state):
     cumulative_reply = text
     for i in range(state['chat_generation_attempts']):
         reply = None
-        for reply in generate_reply(prompt + cumulative_reply, state, eos_token=eos_token, stopping_strings=stopping_strings, is_chat=True):
+        for reply in generate_reply(prompt + cumulative_reply, state, stopping_strings=stopping_strings, is_chat=True):
             reply = cumulative_reply + reply
-            reply, next_character_found = extract_message_from_reply(reply, state)
             yield reply.lstrip(' ')
             if shared.stop_everything:
                 return
 
-            if next_character_found:
-                break
-
         if reply in [None, cumulative_reply]:
             break
         else:

modules/text_generation.py

@@ -9,8 +9,7 @@ import torch
 import transformers
 
 import modules.shared as shared
-from modules.callbacks import (Iteratorize, Stream,
+from modules.callbacks import Iteratorize, Stream
-                               _SentinelTokenStoppingCriteria)
 from modules.extensions import apply_extensions
 from modules.html_generator import generate_4chan_html, generate_basic_html
 from modules.logging_colors import logger
@@ -42,11 +41,6 @@ def encode(prompt, add_special_tokens=True, add_bos_token=True, truncation_lengt
         if not add_bos_token and input_ids[0][0] == shared.tokenizer.bos_token_id:
             input_ids = input_ids[:, 1:]
 
-        # Llama adds this extra token when the first character is '\n', and this
-        # compromises the stopping criteria, so we just remove it
-        if type(shared.tokenizer) is transformers.LlamaTokenizer and input_ids[0][0] == 29871:
-            input_ids = input_ids[:, 1:]
-
     # Handling truncation
     if truncation_length is not None:
         input_ids = input_ids[:, -truncation_length:]
@@ -139,15 +133,43 @@ def stop_everything_event():
     shared.stop_everything = True
 
 
-def generate_reply_wrapper(question, state, eos_token=None, stopping_strings=None):
+def generate_reply_wrapper(question, state, stopping_strings=None):
-    for reply in generate_reply(question, state, eos_token, stopping_strings, is_chat=False):
+    reply = question if not shared.is_seq2seq else ''
+    yield formatted_outputs(reply, shared.model_name)
+
+    for reply in generate_reply(question, state, stopping_strings, is_chat=False):
         if not shared.is_seq2seq:
             reply = question + reply
 
         yield formatted_outputs(reply, shared.model_name)
 
 
-def _generate_reply(question, state, eos_token=None, stopping_strings=None, is_chat=False):
+def apply_stopping_strings(reply, all_stop_strings):
+    stop_found = False
+    for string in all_stop_strings:
+        idx = reply.find(string)
+        if idx != -1:
+            reply = reply[:idx]
+            stop_found = True
+            break
+
+    if not stop_found:
+        # If something like "\nYo" is generated just before "\nYou:"
+        # is completed, trim it
+        for string in all_stop_strings:
+            for j in range(len(string) - 1, 0, -1):
+                if reply[-j:] == string[:j]:
+                    reply = reply[:-j]
+                    break
+            else:
+                continue
+
+            break
+
+    return reply, stop_found
+
+
+def _generate_reply(question, state, stopping_strings=None, is_chat=False):
     state = apply_extensions('state', state)
     generate_func = apply_extensions('custom_generate_reply')
     if generate_func is None:
@@ -168,29 +190,39 @@ def _generate_reply(question, state, eos_token=None, stopping_strings=None, is_c
     if not is_chat:
         question = apply_extensions('input', question)
 
+    # Finding the stopping strings
+    all_stop_strings = []
+    for st in (stopping_strings, ast.literal_eval(f"[{state['custom_stopping_strings']}]")):
+        if type(st) is list and len(st) > 0:
+            all_stop_strings += st
+
     if shared.args.verbose:
         print(f'\n\n{question}\n--------------------\n')
 
     shared.stop_everything = False
     clear_torch_cache()
     seed = set_manual_seed(state['seed'])
-    is_stream = state['stream']
     last_update = -1
     reply = ''
-    for reply in generate_func(question, original_question, seed, state, eos_token, stopping_strings, is_chat=is_chat):
+    is_stream = state['stream']
+    if len(all_stop_strings) > 0 and not state['stream']:
+        state['stream'] = True
+
+    for reply in generate_func(question, original_question, seed, state, stopping_strings, is_chat=is_chat):
+        reply, stop_found = apply_stopping_strings(reply, all_stop_strings)
         if is_stream:
             cur_time = time.time()
             if cur_time - last_update > 0.041666666666666664:  # Limit streaming to 24 fps
                 last_update = cur_time
                 yield reply
-        else:
-            yield reply
 
-    if is_stream:
+        if stop_found:
+            break
+
     yield reply
 
 
-def generate_reply_HF(question, original_question, seed, state, eos_token=None, stopping_strings=None, is_chat=False):
+def generate_reply_HF(question, original_question, seed, state, stopping_strings=None, is_chat=False):
     generate_params = {}
     for k in ['max_new_tokens', 'do_sample', 'temperature', 'top_p', 'typical_p', 'repetition_penalty', 'encoder_repetition_penalty', 'top_k', 'min_length', 'no_repeat_ngram_size', 'num_beams', 'penalty_alpha', 'length_penalty', 'early_stopping', 'tfs', 'top_a', 'mirostat_mode', 'mirostat_tau', 'mirostat_eta']:
         generate_params[k] = state[k]
@@ -213,11 +245,6 @@ def generate_reply_HF(question, original_question, seed, state, eos_token=None,
     output = input_ids[0]
     cuda = not any((shared.args.cpu, shared.args.deepspeed))
 
-    # Find the eos tokens
-    eos_token_ids = [shared.tokenizer.eos_token_id] if shared.tokenizer.eos_token_id is not None else []
-    if eos_token is not None:
-        eos_token_ids.append(int(encode(eos_token)[0][-1]))
-
     # Add the encoded tokens to generate_params
     question, input_ids, inputs_embeds = apply_extensions('tokenizer', state, question, input_ids, None)
     original_input_ids = input_ids
@@ -225,17 +252,10 @@ def generate_reply_HF(question, original_question, seed, state, eos_token=None,
     if inputs_embeds is not None:
         generate_params.update({'inputs_embeds': inputs_embeds})
 
-    # Create the StoppingCriteriaList with the stopping strings (needs to be done after tokenizer extensions)
+    # Find the eos tokens
-    stopping_criteria_list = transformers.StoppingCriteriaList()
+    eos_token_ids = [shared.tokenizer.eos_token_id] if shared.tokenizer.eos_token_id is not None else []
-    for st in (stopping_strings, ast.literal_eval(f"[{state['custom_stopping_strings']}]")):
-        if type(st) is list and len(st) > 0:
-            sentinel_token_ids = [encode(string, add_special_tokens=False) for string in st]
-            stopping_criteria_list.append(_SentinelTokenStoppingCriteria(sentinel_token_ids=sentinel_token_ids, starting_idx=len(input_ids[0])))
-            break
-
-    # Update generate_params with the eos token and the stopping strings
     generate_params['eos_token_id'] = eos_token_ids
-    generate_params['stopping_criteria'] = stopping_criteria_list
+    generate_params['stopping_criteria'] = transformers.StoppingCriteriaList()
 
     t0 = time.time()
     try:
@@ -280,7 +300,7 @@ def generate_reply_HF(question, original_question, seed, state, eos_token=None,
         return
 
 
-def generate_reply_custom(question, original_question, seed, state, eos_token=None, stopping_strings=None, is_chat=False):
+def generate_reply_custom(question, original_question, seed, state, stopping_strings=None, is_chat=False):
     seed = set_manual_seed(state['seed'])
 
     t0 = time.time()
@@ -312,7 +332,7 @@ def generate_reply_custom(question, original_question, seed, state, eos_token=No
         return
 
 
-def generate_reply_flexgen(question, original_question, seed, state, eos_token=None, stopping_strings=None, is_chat=False):
+def generate_reply_flexgen(question, original_question, seed, state, stopping_strings=None, is_chat=False):
     generate_params = {}
     for k in ['max_new_tokens', 'do_sample', 'temperature']:
         generate_params[k] = state[k]
@@ -326,8 +346,8 @@ def generate_reply_flexgen(question, original_question, seed, state, eos_token=N
 
     # Find the eos tokens
     eos_token_ids = [shared.tokenizer.eos_token_id] if shared.tokenizer.eos_token_id is not None else []
-    if eos_token is not None:
+    if not state['ban_eos_token']:
-        eos_token_ids.append(int(encode(eos_token)[0][-1]))
+        generate_params['stop'] = eos_token_ids[-1]
 
     # Add the encoded tokens to generate_params
     question, input_ids, inputs_embeds = apply_extensions('tokenizer', state, question, input_ids, None)
@@ -336,9 +356,6 @@ def generate_reply_flexgen(question, original_question, seed, state, eos_token=N
     if inputs_embeds is not None:
         generate_params.update({'inputs_embeds': inputs_embeds})
 
-    # Update generate_params with the eos token and the stopping strings
-    generate_params['stop'] = eos_token_ids[-1]
-
     t0 = time.time()
     try:
         if not is_chat: