Updated run.py to support Huggingface backend
#2
by
Neel-Shah-29
- opened
- ChatAtomicFlow.yaml +1 -2
- demo.yaml +2 -0
- run.py +9 -0
ChatAtomicFlow.yaml
CHANGED
@@ -23,8 +23,7 @@ backend:
|
|
23 |
|
24 |
|
25 |
top_p: 0.2
|
26 |
-
|
27 |
-
presence_penalty: 0
|
28 |
stream: True
|
29 |
|
30 |
|
|
|
23 |
|
24 |
|
25 |
top_p: 0.2
|
26 |
+
|
|
|
27 |
stream: True
|
28 |
|
29 |
|
demo.yaml
CHANGED
@@ -15,6 +15,7 @@ backend:
|
|
15 |
model_name:
|
16 |
openai: "gpt-3.5-turbo"
|
17 |
azure: "azure/gpt-4"
|
|
|
18 |
|
19 |
# ~~~ generation_parameters ~~
|
20 |
n: 1
|
@@ -24,6 +25,7 @@ backend:
|
|
24 |
top_p: 0.2
|
25 |
frequency_penalty: 0
|
26 |
presence_penalty: 0
|
|
|
27 |
|
28 |
n_api_retries: 6
|
29 |
wait_time_between_retries: 20
|
|
|
15 |
model_name:
|
16 |
openai: "gpt-3.5-turbo"
|
17 |
azure: "azure/gpt-4"
|
18 |
+
huggingface: "huggingface/meta-llama/Meta-Llama-3-70B-Instruct"
|
19 |
|
20 |
# ~~~ generation_parameters ~~
|
21 |
n: 1
|
|
|
25 |
top_p: 0.2
|
26 |
frequency_penalty: 0
|
27 |
presence_penalty: 0
|
28 |
+
stream: False
|
29 |
|
30 |
n_api_retries: 6
|
31 |
wait_time_between_retries: 20
|
run.py
CHANGED
@@ -20,6 +20,11 @@ CACHING_PARAMETERS.do_caching = False # Set to True in order to disable caching
|
|
20 |
# clear_cache() # Uncomment this line to clear the cache
|
21 |
|
22 |
logging.set_verbosity_debug()
|
|
|
|
|
|
|
|
|
|
|
23 |
|
24 |
dependencies = [
|
25 |
{"url": "aiflows/ChatFlowModule", "revision": os.getcwd()},
|
@@ -44,6 +49,10 @@ if __name__ == "__main__":
|
|
44 |
# OpenAI backend
|
45 |
api_information = [ApiInfo(backend_used="openai",
|
46 |
api_key = os.getenv("OPENAI_API_KEY"))]
|
|
|
|
|
|
|
|
|
47 |
# # Azure backend
|
48 |
# api_information = ApiInfo(backend_used = "azure",
|
49 |
# api_base = os.getenv("AZURE_API_BASE"),
|
|
|
20 |
# clear_cache() # Uncomment this line to clear the cache
|
21 |
|
22 |
logging.set_verbosity_debug()
|
23 |
+
# Comment out if using huggingface backend
|
24 |
+
# The set_verbose() method is used for debugging and logging the outputs and drop_params() method automatically drops the parameters which are not set my litellm (since the response format of OpenAI and huggingface is different we require this to translate Huggingface to OpenAI response format).
|
25 |
+
# import litellm
|
26 |
+
# litellm.set_verbose=True
|
27 |
+
# litellm.drop_params=True
|
28 |
|
29 |
dependencies = [
|
30 |
{"url": "aiflows/ChatFlowModule", "revision": os.getcwd()},
|
|
|
49 |
# OpenAI backend
|
50 |
api_information = [ApiInfo(backend_used="openai",
|
51 |
api_key = os.getenv("OPENAI_API_KEY"))]
|
52 |
+
#Huggingface backend
|
53 |
+
# Here the API_BASE is the api link to your hosted server. By this way, you can run any flows on your own server.
|
54 |
+
# api_information = [ApiInfo(backend_used="huggingface",
|
55 |
+
# api_key = os.getenv("HUGGINGFACE_API_KEY"), api_base="http://0.0.0.0:5000/v1/completions")]
|
56 |
# # Azure backend
|
57 |
# api_information = ApiInfo(backend_used = "azure",
|
58 |
# api_base = os.getenv("AZURE_API_BASE"),
|