Mirrors/text-generation-webui
1
0
Fork 0
mirror of https://github.com/oobabooga/text-generation-webui.git synced 2024-11-23 00:18:20 +01:00
Code Issues Packages Projects Releases Wiki Activity

Various fixes in chat mode

Browse Source
This commit is contained in:
oobabooga 2023-03-12 02:31:45 -03:00
parent 0bd5430988
commit b0e8cb8c88
2 changed files with 62 additions and 56 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

16
modules/chat.py
View File

@ -115,14 +115,18 @@ def chatbot_wrapper(text, max_new_tokens, do_sample, temperature, top_p, typical
visible_text = visible_text.replace('\n', '<br>') visible_text = visible_text.replace('\n', '<br>')
text = apply_extensions(text, "input") text = apply_extensions(text, "input")
if custom_generate_chat_prompt is None:
prompt = generate_chat_prompt(text, max_new_tokens, name1, name2, context, chat_prompt_size)
else:
prompt = custom_generate_chat_prompt(text, max_new_tokens, name1, name2, context, chat_prompt_size)
# Generate # Generate
reply = '' reply = ''
for i in range(chat_generation_attempts): for i in range(chat_generation_attempts):
# The prompt needs to be generated here because, as the reply
# grows, it may become necessary to remove more old messages to
# fit into the 2048 tokens window.
if custom_generate_chat_prompt is None:
prompt = generate_chat_prompt(text, max_new_tokens, name1, name2, context, chat_prompt_size-len(encode(' '+reply)[0]))
else:
prompt = custom_generate_chat_prompt(text, max_new_tokens, name1, name2, context, chat_prompt_size-len(encode(' '+reply)[0]))
for reply in generate_reply(f"{prompt}{' ' if len(reply) > 0 else ''}{reply}", max_new_tokens, do_sample, temperature, top_p, typical_p, repetition_penalty, top_k, min_length, no_repeat_ngram_size, num_beams, penalty_alpha, length_penalty, early_stopping, eos_token=eos_token, stopping_string=f"\n{name1}:"): for reply in generate_reply(f"{prompt}{' ' if len(reply) > 0 else ''}{reply}", max_new_tokens, do_sample, temperature, top_p, typical_p, repetition_penalty, top_k, min_length, no_repeat_ngram_size, num_beams, penalty_alpha, length_penalty, early_stopping, eos_token=eos_token, stopping_string=f"\n{name1}:"):
# Extracting the reply # Extracting the reply
@ -156,10 +160,10 @@ def impersonate_wrapper(text, max_new_tokens, do_sample, temperature, top_p, typ
if 'pygmalion' in shared.model_name.lower(): if 'pygmalion' in shared.model_name.lower():
name1 = "You" name1 = "You"
prompt = generate_chat_prompt(text, max_new_tokens, name1, name2, context, chat_prompt_size, impersonate=True)
reply = '' reply = ''
for i in range(chat_generation_attempts): for i in range(chat_generation_attempts):
prompt = generate_chat_prompt(text, max_new_tokens, name1, name2, context, chat_prompt_size-len(encode(' '+reply)[0]), impersonate=True)
for reply in generate_reply(prompt+reply, max_new_tokens, do_sample, temperature, top_p, typical_p, repetition_penalty, top_k, min_length, no_repeat_ngram_size, num_beams, penalty_alpha, length_penalty, early_stopping, eos_token=eos_token, stopping_string=f"\n{name2}:"): for reply in generate_reply(prompt+reply, max_new_tokens, do_sample, temperature, top_p, typical_p, repetition_penalty, top_k, min_length, no_repeat_ngram_size, num_beams, penalty_alpha, length_penalty, early_stopping, eos_token=eos_token, stopping_string=f"\n{name2}:"):
reply, next_character_found, substring_found = extract_message_from_reply(prompt, reply, name1, name2, check, impersonate=True) reply, next_character_found, substring_found = extract_message_from_reply(prompt, reply, name1, name2, check, impersonate=True)
if not substring_found: if not substring_found:

2
modules/text_generation.py
View File

@ -159,6 +159,7 @@ def generate_reply(question, max_new_tokens, do_sample, temperature, top_p, typi
else: else:
generate_params.insert(0, "inputs=input_ids") generate_params.insert(0, "inputs=input_ids")
try:
# Generate the entire reply at once. # Generate the entire reply at once.
if shared.args.no_stream: if shared.args.no_stream:
with torch.no_grad(): with torch.no_grad():
@ -220,6 +221,7 @@ def generate_reply(question, max_new_tokens, do_sample, temperature, top_p, typi
if shared.soft_prompt: if shared.soft_prompt:
inputs_embeds, filler_input_ids = generate_softprompt_input_tensors(input_ids) inputs_embeds, filler_input_ids = generate_softprompt_input_tensors(input_ids)
finally:
t1 = time.time() t1 = time.time()
print(f"Output generated in {(t1-t0):.2f} seconds ({(len(output)-len(original_input_ids[0]))/(t1-t0):.2f} tokens/s, {len(output)-len(original_input_ids[0])} tokens)") print(f"Output generated in {(t1-t0):.2f} seconds ({(len(output)-len(original_input_ids[0]))/(t1-t0):.2f} tokens/s, {len(output)-len(original_input_ids[0])} tokens)")
return return