pseudotensor commited on
Commit
b64f5c9
1 Parent(s): edf6dca

Update with h2oGPT hash 69842ebb9decb32f33bd06c4e6e3bd9c5fb4a7ab

Browse files
Files changed (3) hide show
  1. gen.py +9 -7
  2. gpt_langchain.py +2 -1
  3. gradio_runner.py +5 -3
gen.py CHANGED
@@ -403,13 +403,13 @@ def main(
403
 
404
  # auto-set langchain_mode
405
  if have_langchain and langchain_mode is None:
 
 
406
  if allow_upload_to_user_data and not is_public and user_path:
407
- langchain_mode = 'UserData'
408
- print("Auto set langchain_mode=%s" % langchain_mode, flush=True)
409
  elif allow_upload_to_my_data:
410
- langchain_mode = 'MyData'
411
- print("Auto set langchain_mode=%s."
412
- " To use UserData to pull files from disk,"
413
  " set user_path and ensure allow_upload_to_user_data=True" % langchain_mode, flush=True)
414
  else:
415
  raise RuntimeError("Please pass --langchain_mode=<chosen mode> out of %s" % langchain_modes)
@@ -1555,7 +1555,8 @@ def evaluate(
1555
  where_from = "openai_client"
1556
 
1557
  openai.api_key = os.getenv("OPENAI_API_KEY")
1558
- stop_sequences = list(set(prompter.terminate_response + [prompter.PreResponse]))
 
1559
  stop_sequences = [x for x in stop_sequences if x]
1560
  # OpenAI will complain if ask for too many new tokens, takes it as min in some sense, wrongly so.
1561
  max_new_tokens_openai = min(max_new_tokens, model_max_length - num_prompt_tokens)
@@ -1764,7 +1765,8 @@ def evaluate(
1764
 
1765
  # prompt must include all human-bot like tokens, already added by prompt
1766
  # https://github.com/huggingface/text-generation-inference/tree/main/clients/python#types
1767
- stop_sequences = list(set(prompter.terminate_response + [prompter.PreResponse]))
 
1768
  stop_sequences = [x for x in stop_sequences if x]
1769
  gen_server_kwargs = dict(do_sample=do_sample,
1770
  max_new_tokens=max_new_tokens,
 
403
 
404
  # auto-set langchain_mode
405
  if have_langchain and langchain_mode is None:
406
+ # start in chat mode, in case just want to chat and don't want to get "No documents to query" by default.
407
+ langchain_mode = LangChainMode.CHAT_LLM.value
408
  if allow_upload_to_user_data and not is_public and user_path:
409
+ print("Auto set langchain_mode=%s. Could use UserData instead." % langchain_mode, flush=True)
 
410
  elif allow_upload_to_my_data:
411
+ print("Auto set langchain_mode=%s. Could use MyData instead."
412
+ " To allow UserData to pull files from disk,"
 
413
  " set user_path and ensure allow_upload_to_user_data=True" % langchain_mode, flush=True)
414
  else:
415
  raise RuntimeError("Please pass --langchain_mode=<chosen mode> out of %s" % langchain_modes)
 
1555
  where_from = "openai_client"
1556
 
1557
  openai.api_key = os.getenv("OPENAI_API_KEY")
1558
+ terminate_response = prompter.terminate_response or []
1559
+ stop_sequences = list(set(terminate_response + [prompter.PreResponse]))
1560
  stop_sequences = [x for x in stop_sequences if x]
1561
  # OpenAI will complain if ask for too many new tokens, takes it as min in some sense, wrongly so.
1562
  max_new_tokens_openai = min(max_new_tokens, model_max_length - num_prompt_tokens)
 
1765
 
1766
  # prompt must include all human-bot like tokens, already added by prompt
1767
  # https://github.com/huggingface/text-generation-inference/tree/main/clients/python#types
1768
+ terminate_response = prompter.terminate_response or []
1769
+ stop_sequences = list(set(terminate_response + [prompter.PreResponse]))
1770
  stop_sequences = [x for x in stop_sequences if x]
1771
  gen_server_kwargs = dict(do_sample=do_sample,
1772
  max_new_tokens=max_new_tokens,
gpt_langchain.py CHANGED
@@ -643,7 +643,8 @@ def get_llm(use_openai_model=False,
643
 
644
  callbacks = [StreamingGradioCallbackHandler()]
645
  assert prompter is not None
646
- stop_sequences = list(set(prompter.terminate_response + [prompter.PreResponse]))
 
647
  stop_sequences = [x for x in stop_sequences if x]
648
 
649
  if gr_client:
 
643
 
644
  callbacks = [StreamingGradioCallbackHandler()]
645
  assert prompter is not None
646
+ terminate_response = prompter.terminate_response or []
647
+ stop_sequences = list(set(terminate_response + [prompter.PreResponse]))
648
  stop_sequences = [x for x in stop_sequences if x]
649
 
650
  if gr_client:
gradio_runner.py CHANGED
@@ -434,6 +434,7 @@ def go_gradio(**kwargs):
434
  value='All',
435
  interactive=True,
436
  multiselect=True,
 
437
  )
438
  sources_visible = kwargs['langchain_mode'] != 'Disabled' and enable_sources_list
439
  with gr.Row():
@@ -454,10 +455,11 @@ def go_gradio(**kwargs):
454
  with gr.Column(scale=2):
455
  sources_text = gr.HTML(label='Sources Added', interactive=False)
456
 
457
- doc_exception_text = gr.Textbox(value="", visible=True, label='Document Exceptions',
458
- interactive=False)
 
459
  with gr.TabItem("Document Viewer"):
460
- with gr.Row():
461
  with gr.Column(scale=2):
462
  get_viewable_sources_btn = gr.Button(value="Update UI with Document(s) from DB", scale=0,
463
  size='sm',
 
434
  value='All',
435
  interactive=True,
436
  multiselect=True,
437
+ visible=kwargs['langchain_mode'] != 'Disabled',
438
  )
439
  sources_visible = kwargs['langchain_mode'] != 'Disabled' and enable_sources_list
440
  with gr.Row():
 
455
  with gr.Column(scale=2):
456
  sources_text = gr.HTML(label='Sources Added', interactive=False)
457
 
458
+ doc_exception_text = gr.Textbox(value="", label='Document Exceptions',
459
+ interactive=False,
460
+ visible=kwargs['langchain_mode'] != 'Disabled')
461
  with gr.TabItem("Document Viewer"):
462
+ with gr.Row(visible=kwargs['langchain_mode'] != 'Disabled'):
463
  with gr.Column(scale=2):
464
  get_viewable_sources_btn = gr.Button(value="Update UI with Document(s) from DB", scale=0,
465
  size='sm',