Update with h2oGPT hash e195e9bfebca2b11ee3334c10df5997816cf7d6f

generate.py

@@ -276,8 +276,8 @@ def main(
 
     # allow set token directly
     use_auth_token = os.environ.get("HUGGINGFACE_API_TOKEN", use_auth_token)
-    allow_upload_to_user_data = bool(int(os.environ.get("allow_upload_to_user_data", allow_upload_to_user_data)))
-    allow_upload_to_my_data = bool(int(os.environ.get("allow_upload_to_my_data", allow_upload_to_my_data)))
+    allow_upload_to_user_data = bool(int(os.environ.get("allow_upload_to_user_data", str(int(allow_upload_to_user_data)))))
+    allow_upload_to_my_data = bool(int(os.environ.get("allow_upload_to_my_data", str(int(allow_upload_to_my_data)))))
     height = int(os.environ.get("HEIGHT", height))
     h2ocolors = bool(int(os.getenv('h2ocolors', h2ocolors)))
 
@@ -322,8 +322,8 @@ def main(
     if score_model == 'None' or score_model is None:
         score_model = ''
     concurrency_count = int(os.getenv('CONCURRENCY_COUNT', concurrency_count))
-    api_open = bool(int(os.getenv('API_OPEN', api_open)))
-    allow_api = bool(int(os.getenv('ALLOW_API', allow_api)))
+    api_open = bool(int(os.getenv('API_OPEN', str(int(api_open)))))
+    allow_api = bool(int(os.getenv('ALLOW_API', str(int(allow_api)))))
 
     n_gpus = torch.cuda.device_count() if torch.cuda.is_available else 0
     if n_gpus == 0:

gradio_runner.py

@@ -46,6 +46,17 @@ from generate import get_model, languages_covered, evaluate, eval_func_param_nam
 from apscheduler.schedulers.background import BackgroundScheduler
 
 
+def fix_newlines(text):
+    # ensure good visually, else markdown ignores multiple \n
+    # handle code blocks
+    ts = text.split('```')
+    for parti, part in enumerate(ts):
+        inside = parti % 2 == 1
+        if not inside:
+            ts[parti] = ts[parti].replace('\n', '<br>')
+    return '```'.join(ts)
+
+
 def go_gradio(**kwargs):
     allow_api = kwargs['allow_api']
     is_public = kwargs['is_public']
@@ -770,7 +781,7 @@ def go_gradio(**kwargs):
 
         def evaluate_gradio(*args1, **kwargs1):
             for res_dict in evaluate(*args1, **kwargs1):
-                yield '<br>' + res_dict['response'].replace("\n", "<br>")
+                yield '<br>' + fix_newlines(res_dict['response'])
 
         fun = partial(evaluate_gradio,
                       **kwargs_evaluate)
@@ -926,8 +937,7 @@ def go_gradio(**kwargs):
                 # e.g. when user just hits enter in textbox,
                 # else will have <human>: <bot>: on single line, which seems to be "ok" for LLM but not usual
                 user_message1 = '\n'
-            # ensure good visually, else markdown ignores multiple \n
-            user_message1 = user_message1.replace('\n', '<br>')
+            user_message1 = fix_newlines(user_message1)
 
             history = args_list[-1]
             if undo and history:
@@ -1052,7 +1062,7 @@ def go_gradio(**kwargs):
                     output = output_fun['response']
                     extra = output_fun['sources']  # FIXME: can show sources in separate text box etc.
                     # ensure good visually, else markdown ignores multiple \n
-                    bot_message = output.replace('\n', '<br>')
+                    bot_message = fix_newlines(output)
                     history[-1][1] = bot_message
                     yield history, ''
             except StopIteration:

prompter.py

@@ -36,8 +36,12 @@ prompt_type_to_model_name = {
         'h2oai/h2ogpt-gm-oasst1-en-2048-open-llama-7b-preview-300bt',
         'h2oai/h2ogpt-gm-oasst1-en-2048-open-llama-7b-preview-300bt-v2',
         'h2oai/h2ogpt-gm-oasst1-en-2048-open-llama-7b-preview-700bt',
+        'h2oai/h2ogpt-gm-oasst1-en-2048-open-llama-7b',
         'h2oai/h2ogpt-gm-oasst1-multilang-2048-falcon-7b',
         'h2oai/h2ogpt-gm-oasst1-multilang-2048-falcon-7b-v2',
+        'h2oai/h2ogpt-gm-oasst1-en-2048-falcon-7b',
+        'h2oai/h2ogpt-gm-oasst1-en-2048-falcon-7b-v2',
+        'h2oai/h2ogpt-gm-oasst1-en-2048-falcon-40b-v1',
     ],
     'instruct': [],
     'instruct_with_end': ['databricks/dolly-v2-12b'],

requirements.txt

@@ -1,7 +1,7 @@
 # for generate (gradio server) and finetune
 datasets==2.12.0
 sentencepiece==0.1.97
-gradio==3.31.0
+gradio==3.34.0
 huggingface_hub==0.14.1
 appdirs==1.4.4
 fire==0.5.0
@@ -35,7 +35,7 @@ tensorboard==2.12.1
 neptune==1.1.1
 
 # for gradio client
-gradio_client==0.2.5
+gradio_client==0.2.6
 beautifulsoup4==4.12.2
 markdown==3.4.1
 
@@ -101,7 +101,8 @@ tabulate==0.9.0
 pip-licenses==4.3.0
 
 # weaviate vector db
-weaviate-client==3.19.2# optional for chat with PDF
+weaviate-client==3.19.2
+# optional for chat with PDF
 langchain==0.0.193
 pypdf==3.8.1
 tiktoken==0.3.3
@@ -145,7 +146,8 @@ tabulate==0.9.0
 pip-licenses==4.3.0
 
 # weaviate vector db
-weaviate-client==3.19.2faiss-gpu==1.7.2
+weaviate-client==3.19.2
+faiss-gpu==1.7.2
 gpt4all==0.2.3
 llama-cpp-python==0.1.55
 arxiv==1.4.7