Spaces:
Sleeping
Sleeping
cleanup and harmonization
Browse files- .gitignore +1 -0
- Dockerfile +2 -1
- README.md +8 -19
- app/main.py +2 -88
- app/prompt.py +0 -7
- app/utils.py +178 -0
- params.cfg +35 -0
- requirements.txt +19 -5
.gitignore
ADDED
@@ -0,0 +1 @@
|
|
|
|
|
1 |
+
.DS_Store
|
Dockerfile
CHANGED
@@ -13,7 +13,8 @@ RUN pip install --no-cache-dir -r requirements.txt
|
|
13 |
|
14 |
# -------- copy source --------
|
15 |
COPY app ./app
|
16 |
-
COPY
|
|
|
17 |
|
18 |
# Ports:
|
19 |
# • 7860 → Gradio UI (HF Spaces standard)
|
|
|
13 |
|
14 |
# -------- copy source --------
|
15 |
COPY app ./app
|
16 |
+
COPY params.cfg .
|
17 |
+
COPY .env* ./
|
18 |
|
19 |
# Ports:
|
20 |
# • 7860 → Gradio UI (HF Spaces standard)
|
README.md
CHANGED
@@ -8,27 +8,16 @@ pinned: false
|
|
8 |
license: mit
|
9 |
---
|
10 |
|
11 |
-
#
|
12 |
|
13 |
-
This is
|
14 |
-
|
15 |
-
## How to use
|
16 |
-
|
17 |
-
1. Enter your question in the "Query" field
|
18 |
-
2. Paste relevant documents or context in the "Context" field
|
19 |
-
3. Click submit to get an AI-generated answer based on your context
|
20 |
-
|
21 |
-
## Features
|
22 |
-
|
23 |
-
- Uses state-of-the-art language models via Hugging Face Inference API
|
24 |
-
- Supports multiple model providers
|
25 |
-
- Clean, intuitive interface
|
26 |
-
- Example queries to get started
|
27 |
|
28 |
## Configuration
|
29 |
|
30 |
-
|
31 |
-
|
32 |
-
|
|
|
|
|
33 |
|
34 |
-
|
|
|
8 |
license: mit
|
9 |
---
|
10 |
|
11 |
+
# Generation Module
|
12 |
|
13 |
+
This is an LLM-based generation service designed to be deployed as a modular component of a broader RAG system. The service runs on a docker container and exposes a gradio UI on port 7860 as well as an MCP endpoint.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
14 |
|
15 |
## Configuration
|
16 |
|
17 |
+
1. The module requires an API key (set as an environment variable) for an inference provider to run. Multiple inference providers are supported. Make sure to set the appropriate environment variables:
|
18 |
+
- OpenAI: `OPENAI_API_KEY`
|
19 |
+
- Anthropic: `ANTHROPIC_API_KEY`
|
20 |
+
- Cohere: `COHERE_API_KEY`
|
21 |
+
- HuggingFace: `HF_TOKEN`
|
22 |
|
23 |
+
2. Inference provider and model settings are accessible via params.cfg
|
app/main.py
CHANGED
@@ -1,81 +1,5 @@
|
|
1 |
-
import os, asyncio, logging
|
2 |
import gradio as gr
|
3 |
-
from
|
4 |
-
from .prompt import build_prompt
|
5 |
-
|
6 |
-
# ---------------------------------------------------------------------
|
7 |
-
# model / client initialisation
|
8 |
-
# ---------------------------------------------------------------------
|
9 |
-
HF_TOKEN = os.getenv("HF_TOKEN")
|
10 |
-
MODEL_ID = os.getenv("MODEL_ID", "meta-llama/Meta-Llama-3-8B-Instruct")
|
11 |
-
MAX_TOKENS = int(os.getenv("MAX_NEW_TOKENS", "512"))
|
12 |
-
TEMPERATURE = float(os.getenv("TEMPERATURE", "0.2"))
|
13 |
-
|
14 |
-
if not HF_TOKEN:
|
15 |
-
raise RuntimeError(
|
16 |
-
"HF_TOKEN env-var missing. "
|
17 |
-
)
|
18 |
-
|
19 |
-
client = InferenceClient(model=MODEL_ID, token=HF_TOKEN)
|
20 |
-
|
21 |
-
# ---------------------------------------------------------------------
|
22 |
-
# Core generation function for both Gradio UI and MCP
|
23 |
-
# ---------------------------------------------------------------------
|
24 |
-
async def _call_llm(prompt: str) -> str:
|
25 |
-
"""
|
26 |
-
Try text_generation first (for models/providers that still support it);
|
27 |
-
fall back to chat_completion when the provider is chat-only (Novita, etc.).
|
28 |
-
"""
|
29 |
-
try:
|
30 |
-
# hf-inference
|
31 |
-
return await asyncio.to_thread(
|
32 |
-
client.text_generation,
|
33 |
-
prompt,
|
34 |
-
max_new_tokens=MAX_TOKENS,
|
35 |
-
temperature=TEMPERATURE,
|
36 |
-
)
|
37 |
-
except ValueError as e:
|
38 |
-
if "Supported task: conversational" not in str(e):
|
39 |
-
raise # genuine error → bubble up
|
40 |
-
|
41 |
-
# fallback for Novita
|
42 |
-
messages = [{"role": "user", "content": prompt}]
|
43 |
-
completion = await asyncio.to_thread(
|
44 |
-
client.chat_completion,
|
45 |
-
messages=messages,
|
46 |
-
model=MODEL_ID,
|
47 |
-
max_tokens=MAX_TOKENS,
|
48 |
-
temperature=TEMPERATURE,
|
49 |
-
)
|
50 |
-
return completion.choices[0].message.content.strip()
|
51 |
-
|
52 |
-
async def rag_generate(query: str, context: str) -> str:
|
53 |
-
"""
|
54 |
-
Generate an answer to a query using provided context through RAG.
|
55 |
-
|
56 |
-
This function takes a user query and relevant context, then uses a language model
|
57 |
-
to generate a comprehensive answer based on the provided information.
|
58 |
-
|
59 |
-
Args:
|
60 |
-
query (str): The user's question or query
|
61 |
-
context (str): The relevant context/documents to use for answering
|
62 |
-
|
63 |
-
Returns:
|
64 |
-
str: The generated answer based on the query and context
|
65 |
-
"""
|
66 |
-
if not query.strip():
|
67 |
-
return "Error: Query cannot be empty"
|
68 |
-
|
69 |
-
if not context.strip():
|
70 |
-
return "Error: Context cannot be empty"
|
71 |
-
|
72 |
-
prompt = build_prompt(query, context)
|
73 |
-
try:
|
74 |
-
answer = await _call_llm(prompt)
|
75 |
-
return answer
|
76 |
-
except Exception as e:
|
77 |
-
logging.exception("Generation failed")
|
78 |
-
return f"Error: {str(e)}"
|
79 |
|
80 |
# ---------------------------------------------------------------------
|
81 |
# Gradio Interface with MCP support
|
@@ -102,17 +26,7 @@ ui = gr.Interface(
|
|
102 |
show_copy_button=True
|
103 |
),
|
104 |
title="RAG Generation Service",
|
105 |
-
description="Ask questions
|
106 |
-
examples=[
|
107 |
-
[
|
108 |
-
"What is the main benefit mentioned?",
|
109 |
-
"Machine learning has revolutionized many industries. The main benefit is increased efficiency and accuracy in data processing."
|
110 |
-
],
|
111 |
-
[
|
112 |
-
"Who is the CEO?",
|
113 |
-
"Company ABC was founded in 2020. The current CEO is Jane Smith, who has led the company to significant growth."
|
114 |
-
]
|
115 |
-
]
|
116 |
)
|
117 |
|
118 |
# Launch with MCP server enabled
|
|
|
|
|
1 |
import gradio as gr
|
2 |
+
from .utils import rag_generate
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
3 |
|
4 |
# ---------------------------------------------------------------------
|
5 |
# Gradio Interface with MCP support
|
|
|
26 |
show_copy_button=True
|
27 |
),
|
28 |
title="RAG Generation Service",
|
29 |
+
description="Ask questions based on provided context. Intended for use in RAG pipelines (i.e. context supplied by semantic retriever service) as an MCP server.",
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
30 |
)
|
31 |
|
32 |
# Launch with MCP server enabled
|
app/prompt.py
DELETED
@@ -1,7 +0,0 @@
|
|
1 |
-
def build_prompt(question: str, context: str) -> str:
|
2 |
-
return (
|
3 |
-
"You are an expert assistant. Answer the USER question using only the "
|
4 |
-
"CONTEXT provided. If the context is insufficient say 'I don't know.'.\n\n"
|
5 |
-
f"### CONTEXT\n{context}\n\n"
|
6 |
-
f"### USER QUESTION\n{question}\n\n### ASSISTANT ANSWER\n"
|
7 |
-
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
app/utils.py
ADDED
@@ -0,0 +1,178 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import os, asyncio, logging
|
2 |
+
import configparser
|
3 |
+
import logging
|
4 |
+
from dotenv import load_dotenv
|
5 |
+
|
6 |
+
# LangChain imports
|
7 |
+
from langchain_openai import ChatOpenAI
|
8 |
+
from langchain_anthropic import ChatAnthropic
|
9 |
+
from langchain_cohere import ChatCohere
|
10 |
+
from langchain_huggingface import ChatHuggingFace, HuggingFaceEndpoint
|
11 |
+
from langchain_core.messages import SystemMessage, HumanMessage
|
12 |
+
|
13 |
+
# Local .env file
|
14 |
+
load_dotenv()
|
15 |
+
|
16 |
+
def getconfig(configfile_path: str):
|
17 |
+
"""
|
18 |
+
Read the config file
|
19 |
+
Params
|
20 |
+
----------------
|
21 |
+
configfile_path: file path of .cfg file
|
22 |
+
"""
|
23 |
+
config = configparser.ConfigParser()
|
24 |
+
try:
|
25 |
+
config.read_file(open(configfile_path))
|
26 |
+
return config
|
27 |
+
except:
|
28 |
+
logging.warning("config file not found")
|
29 |
+
|
30 |
+
# ---------------------------------------------------------------------
|
31 |
+
# Provider-agnostic authentication and configuration
|
32 |
+
# ---------------------------------------------------------------------
|
33 |
+
def get_auth_config(provider: str) -> dict:
|
34 |
+
"""Get authentication configuration for different providers"""
|
35 |
+
auth_configs = {
|
36 |
+
"openai": {"api_key": os.getenv("OPENAI_API_KEY")},
|
37 |
+
"huggingface": {"api_key": os.getenv("HF_TOKEN")},
|
38 |
+
"anthropic": {"api_key": os.getenv("ANTHROPIC_API_KEY")},
|
39 |
+
"cohere": {"api_key": os.getenv("COHERE_API_KEY")},
|
40 |
+
}
|
41 |
+
|
42 |
+
if provider not in auth_configs:
|
43 |
+
raise ValueError(f"Unsupported provider: {provider}")
|
44 |
+
|
45 |
+
auth_config = auth_configs[provider]
|
46 |
+
api_key = auth_config.get("api_key")
|
47 |
+
|
48 |
+
if not api_key:
|
49 |
+
raise RuntimeError(f"Missing API key for provider '{provider}'. Please set the appropriate environment variable.")
|
50 |
+
|
51 |
+
return auth_config
|
52 |
+
|
53 |
+
# ---------------------------------------------------------------------
|
54 |
+
# Model / client initialization
|
55 |
+
# ---------------------------------------------------------------------
|
56 |
+
config = getconfig("params.cfg")
|
57 |
+
|
58 |
+
PROVIDER = config.get("generator", "PROVIDER")
|
59 |
+
MODEL = config.get("generator", "MODEL")
|
60 |
+
MAX_TOKENS = int(config.get("generator", "MAX_TOKENS"))
|
61 |
+
TEMPERATURE = float(config.get("generator", "TEMPERATURE"))
|
62 |
+
|
63 |
+
# Set up authentication for the selected provider
|
64 |
+
auth_config = get_auth_config(PROVIDER)
|
65 |
+
|
66 |
+
def get_chat_model():
|
67 |
+
"""Initialize the appropriate LangChain chat model based on provider"""
|
68 |
+
common_params = {
|
69 |
+
"temperature": TEMPERATURE,
|
70 |
+
"max_tokens": MAX_TOKENS,
|
71 |
+
}
|
72 |
+
|
73 |
+
if PROVIDER == "openai":
|
74 |
+
return ChatOpenAI(
|
75 |
+
model=MODEL,
|
76 |
+
openai_api_key=auth_config["api_key"],
|
77 |
+
**common_params
|
78 |
+
)
|
79 |
+
elif PROVIDER == "anthropic":
|
80 |
+
return ChatAnthropic(
|
81 |
+
model=MODEL,
|
82 |
+
anthropic_api_key=auth_config["api_key"],
|
83 |
+
**common_params
|
84 |
+
)
|
85 |
+
elif PROVIDER == "cohere":
|
86 |
+
return ChatCohere(
|
87 |
+
model=MODEL,
|
88 |
+
cohere_api_key=auth_config["api_key"],
|
89 |
+
**common_params
|
90 |
+
)
|
91 |
+
elif PROVIDER == "huggingface":
|
92 |
+
# Initialize HuggingFaceEndpoint with explicit parameters
|
93 |
+
llm = HuggingFaceEndpoint(
|
94 |
+
repo_id=MODEL,
|
95 |
+
huggingfacehub_api_token=auth_config["api_key"],
|
96 |
+
task="text-generation",
|
97 |
+
temperature=TEMPERATURE,
|
98 |
+
max_new_tokens=MAX_TOKENS
|
99 |
+
)
|
100 |
+
return ChatHuggingFace(llm=llm)
|
101 |
+
else:
|
102 |
+
raise ValueError(f"Unsupported provider: {PROVIDER}")
|
103 |
+
|
104 |
+
# Initialize provider-agnostic chat model
|
105 |
+
chat_model = get_chat_model()
|
106 |
+
|
107 |
+
# ---------------------------------------------------------------------
|
108 |
+
# Core generation function for both Gradio UI and MCP
|
109 |
+
# ---------------------------------------------------------------------
|
110 |
+
async def _call_llm(messages: list) -> str:
|
111 |
+
"""
|
112 |
+
Provider-agnostic LLM call using LangChain.
|
113 |
+
|
114 |
+
Args:
|
115 |
+
messages: List of LangChain message objects
|
116 |
+
|
117 |
+
Returns:
|
118 |
+
Generated response content as string
|
119 |
+
"""
|
120 |
+
try:
|
121 |
+
# Use async invoke for better performance
|
122 |
+
response = await chat_model.ainvoke(messages)
|
123 |
+
return response.content.strip()
|
124 |
+
except Exception as e:
|
125 |
+
logging.exception(f"LLM generation failed with provider '{PROVIDER}' and model '{MODEL}': {e}")
|
126 |
+
raise
|
127 |
+
|
128 |
+
def build_messages(question: str, context: str) -> list:
|
129 |
+
"""
|
130 |
+
Build messages in LangChain format.
|
131 |
+
|
132 |
+
Args:
|
133 |
+
question: The user's question
|
134 |
+
context: The relevant context for answering
|
135 |
+
|
136 |
+
Returns:
|
137 |
+
List of LangChain message objects
|
138 |
+
"""
|
139 |
+
system_content = (
|
140 |
+
"You are an expert assistant. Answer the USER question using only the "
|
141 |
+
"CONTEXT provided. If the context is insufficient say 'I don't know.'"
|
142 |
+
)
|
143 |
+
|
144 |
+
user_content = f"### CONTEXT\n{context}\n\n### USER QUESTION\n{question}"
|
145 |
+
|
146 |
+
return [
|
147 |
+
SystemMessage(content=system_content),
|
148 |
+
HumanMessage(content=user_content)
|
149 |
+
]
|
150 |
+
|
151 |
+
|
152 |
+
async def rag_generate(query: str, context: str) -> str:
|
153 |
+
"""
|
154 |
+
Generate an answer to a query using provided context through RAG.
|
155 |
+
|
156 |
+
This function takes a user query and relevant context, then uses a language model
|
157 |
+
to generate a comprehensive answer based on the provided information.
|
158 |
+
|
159 |
+
Args:
|
160 |
+
query (str): The user's question or query
|
161 |
+
context (str): The relevant context/documents to use for answering
|
162 |
+
|
163 |
+
Returns:
|
164 |
+
str: The generated answer based on the query and context
|
165 |
+
"""
|
166 |
+
if not query.strip():
|
167 |
+
return "Error: Query cannot be empty"
|
168 |
+
|
169 |
+
if not context.strip():
|
170 |
+
return "Error: Context cannot be empty"
|
171 |
+
|
172 |
+
try:
|
173 |
+
messages = build_messages(query, context)
|
174 |
+
answer = await _call_llm(messages)
|
175 |
+
return answer
|
176 |
+
except Exception as e:
|
177 |
+
logging.exception("Generation failed")
|
178 |
+
return f"Error: {str(e)}"
|
params.cfg
ADDED
@@ -0,0 +1,35 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
[generator]
|
2 |
+
PROVIDER = huggingface
|
3 |
+
MODEL = meta-llama/Meta-Llama-3-8B-Instruct
|
4 |
+
MAX_TOKENS = 512
|
5 |
+
TEMPERATURE = 0.2
|
6 |
+
|
7 |
+
## OpenAI
|
8 |
+
# [generator]
|
9 |
+
# PROVIDER = openai
|
10 |
+
# MODEL = gpt-4o
|
11 |
+
# MAX_TOKENS = 512
|
12 |
+
# TEMPERATURE = 0.2
|
13 |
+
|
14 |
+
## Anthropic
|
15 |
+
# [generator]
|
16 |
+
# PROVIDER = anthropic
|
17 |
+
# MODEL = claude-3-haiku-20240307
|
18 |
+
# MAX_TOKENS = 512
|
19 |
+
# TEMPERATURE = 0.2
|
20 |
+
|
21 |
+
## Cohere
|
22 |
+
# [generator]
|
23 |
+
# PROVIDER = cohere
|
24 |
+
# MODEL = command
|
25 |
+
# MAX_TOKENS = 512
|
26 |
+
# TEMPERATURE = 0.2
|
27 |
+
|
28 |
+
|
29 |
+
## Environment Variables Required
|
30 |
+
|
31 |
+
# Make sure to set the appropriate environment variables:
|
32 |
+
# - OpenAI: `OPENAI_API_KEY`
|
33 |
+
# - Anthropic: `ANTHROPIC_API_KEY`
|
34 |
+
# - Cohere: `COHERE_API_KEY`
|
35 |
+
# - HuggingFace: `HF_TOKEN`
|
requirements.txt
CHANGED
@@ -1,5 +1,19 @@
|
|
1 |
-
|
2 |
-
gradio
|
3 |
-
|
4 |
-
|
5 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# Core dependencies
|
2 |
+
gradio>=4.0.0
|
3 |
+
gradio[mcp]
|
4 |
+
python-dotenv>=1.0.0
|
5 |
+
|
6 |
+
# LangChain core
|
7 |
+
langchain-core>=0.1.0
|
8 |
+
langchain-community>=0.0.1
|
9 |
+
|
10 |
+
# Provider-specific LangChain packages
|
11 |
+
langchain-openai>=0.1.0
|
12 |
+
langchain-anthropic>=0.1.0
|
13 |
+
langchain-cohere>=0.1.0
|
14 |
+
langchain-together>=0.1.0
|
15 |
+
langchain-huggingface>=0.0.1
|
16 |
+
|
17 |
+
# Additional dependencies that might be needed
|
18 |
+
requests>=2.31.0
|
19 |
+
pydantic>=2.0.0
|