testtsettset / app.py
wuhp's picture
Update app.py
ce8cbe4 verified
raw
history blame
77.2 kB
import os
import re
import time
import json
import io
import requests
import gradio as gr
import google.generativeai as genai
from google.generativeai import types # Import types for configuration and tools
from huggingface_hub import create_repo, list_models, upload_file, constants
from huggingface_hub.utils import build_hf_headers, get_session, hf_raise_for_status
# Removed the debugging print that attempts to read GOOGLE_API_KEY from environment
# --- Define Gemini Model Information ---
# Dictionary mapping internal model name to (Display Name, Description)
GEMINI_MODELS = {
"gemini-1.5-flash": ("Gemini 1.5 Flash", "Fast and versatile performance across a diverse variety of tasks."),
"gemini-1.5-pro": ("Gemini 1.5 Pro", "Complex reasoning tasks requiring more intelligence."),
"gemini-1.5-flash-8b": ("Gemini 1.5 Flash 8B", "High volume and lower intelligence tasks."),
"gemini-2.0-flash": ("Gemini 2.0 Flash", "Next generation features, speed, thinking, realtime streaming, and multimodal generation."),
"gemini-2.0-flash-lite": ("Gemini 2.0 Flash-Lite", "Cost efficiency and low latency."),
# Note: Preview models might have shorter lifespans or different capabilities. Uncomment if you want to include them.
# "gemini-2.5-flash-preview-04-17": ("Gemini 2.5 Flash Preview (04-17)", "Adaptive thinking, cost efficiency."),
# "gemini-2.5-pro-preview-03-25": ("Gemini 2.5 Pro Preview (03-25)", "Enhanced thinking and reasoning, multimodal understanding, advanced coding, and more."),
}
# Create the list of choices for the Gradio Radio component
# Format is (Display Name, Internal Name)
GEMINI_MODEL_CHOICES = [(display_name, internal_name) for internal_name, (display_name, description) in GEMINI_MODELS.items()]
# Define the default model to be selected
DEFAULT_GEMINI_MODEL = "gemini-1.5-flash" # Ensure this key exists in GEMINI_MODELS
# --- Helper functions for Hugging Face integration ---
def show_profile(profile: gr.OAuthProfile | None) -> str:
"""Displays the logged-in Hugging Face profile username."""
if profile is None:
return "*Not logged in.*"
return f"✅ Logged in as **{profile.username}**"
def list_private_models(
profile: gr.OAuthProfile | None,
oauth_token: gr.OAuthToken | None
) -> str:
"""Lists private models for the logged-in user (not used in the main workflow, but kept)."""
if profile is None or oauth_token is None:
return "Please log in to see your models."
try:
models = [
f"{m.id} ({'private' if m.private else 'public'})"
for m in list_models(author=profile.username, token=oauth_token.token)
]
return "No models found." if not models else "Models:\n\n" + "\n - ".join(models)
except Exception as e:
# Catching generic exception is acceptable for helper functions
return f"Error listing models: {e}"
def create_space_action(repo_name: str, sdk: str, profile: gr.OAuthProfile, token: gr.OAuthToken):
"""Creates a new Hugging Face Space repository."""
if not profile or not token:
raise ValueError("Hugging Face profile or token is missing.")
repo_id = f"{profile.username}/{repo_name}"
try:
create_repo(
repo_id=repo_id,
token=token.token,
exist_ok=True, # Allow creating if it already exists
repo_type="space",
space_sdk=sdk
)
url = f"https://huggingface.co/spaces/{repo_id}"
iframe = f'<iframe src="{url}" width="100%" height="500px"></iframe>'
return repo_id, iframe
except Exception as e:
raise RuntimeError(f"Failed to create Space `{repo_id}`: {e}")
def upload_file_to_space_action(
file_obj: io.StringIO, # Specify type hint for clarity
path_in_repo: str,
repo_id: str,
profile: gr.OAuthProfile,
token: gr.OAuthToken
) -> None:
"""Uploads a file to a Huging Face Space repository."""
if not (profile and token and repo_id):
raise ValueError("Hugging Face profile, token, or repo_id is missing.")
try:
upload_file(
path_or_fileobj=file_obj,
path_in_repo=path_in_repo,
repo_id=repo_id,
token=token.token,
repo_type="space"
)
except Exception as e:
raise RuntimeError(f"Failed to upload `{path_in_repo}` to `{repo_id}`: {e}")
def _fetch_space_logs_level(repo_id: str, level: str, token: str) -> str:
"""Fetches build or run logs for a Space."""
if not repo_id or not token:
return f"Cannot fetch {level} logs: repo_id or token missing."
jwt_url = f"{constants.ENDPOINT}/api/spaces/{repo_id}/jwt"
try:
r = get_session().get(jwt_url, headers=build_hf_headers(token=token))
hf_raise_for_status(r) # Raise HTTPError for bad responses (4xx or 5xx)
jwt = r.json()["token"]
logs_url = f"https://api.hf.space/v1/{repo_id}/logs/{level}"
lines, count = [], 0
# Using stream=True is good for potentially large logs
with get_session().get(logs_url, headers=build_hf_headers(token=jwt), stream=True, timeout=30) as resp:
hf_raise_for_status(resp)
for raw in resp.iter_lines():
if count >= 200: # Limit output lines to prevent UI overload
lines.append("... truncated ...")
break
if not raw.startswith(b"data: "): # EventStream protocol expected from HF logs API
continue
payload = raw[len(b"data: "):]
try:
event = json.loads(payload.decode())
ts = event.get("timestamp", "")
txt = event.get("data", "").strip()
if txt:
lines.append(f"[{ts}] {txt}")
count += 1
except json.JSONDecodeError:
# Skip lines that aren't valid JSON events
continue
return "\n".join(lines) if lines else f"No {level} logs found."
except Exception as e:
# Catching generic exception is acceptable for helper functions
return f"Error fetching {level} logs for `{repo_id}`: {e}"
def get_build_logs_action(repo_id, profile, token):
"""Action to fetch build logs with a small delay."""
if not (repo_id and profile and token):
return "⚠️ Cannot fetch build logs: log in and create a Space first."
# Small delay to allow build process to potentially start on HF side
time.sleep(5)
return _fetch_space_logs_level(repo_id, "build", token.token)
def get_container_logs_action(repo_id, profile, token):
"""Action to fetch container logs with a delay."""
if not (repo_id and profile and token):
return "⚠️ Cannot fetch container logs: log in and create a Space first."
# Longer delay to allow container to start after build completes
time.sleep(10)
return _fetch_space_logs_level(repo_id, "run", token.token)
# --- Google Gemini integration with model selection and grounding ---
def configure_gemini(api_key: str | None, model_name: str | None) -> str:
"""Configures the Gemini API and checks if the model is accessible."""
# Check for empty string "" as well as None
if not api_key:
return "⚠️ Gemini API key is not set."
# Check if model_name is None or not a valid key in GEMINI_MODELS
if not model_name or model_name not in GEMINI_MODELS:
return "⚠️ Please select a valid Gemini model."
try:
genai.configure(api_key=api_key)
# Attempt a simple call to verify credentials and model availability
# This will raise an exception if the key is invalid or model not found
genai.GenerativeModel(model_name).generate_content("ping", stream=False)
# This message indicates the API call *for configuration check* was successful
return f"✅ Gemini configured successfully with **{GEMINI_MODELS[model_name][0]}**."
except Exception as e:
# This message indicates the API call *for configuration check* failed
return f"❌ Error configuring Gemini: {e}"
def get_model_description(model_name: str | None) -> str:
"""Retrieves the description for a given model name."""
if model_name is None or model_name not in GEMINI_MODELS:
return "Select a model to see its description."
# Use .get with a default value to handle cases where the key might not be found
return GEMINI_MODELS.get(model_name, (model_name, "No description available."))[1]
def call_gemini(prompt: str, api_key: str, model_name: str, use_grounding: bool = False) -> str:
"""Calls the Gemini API with a given prompt, optionally using grounding."""
# This check is crucial - it will raise an error *before* the API call if prereqs aren't met
# Check for empty string "" as well as None
if not isinstance(api_key, str) or api_key == "" or not model_name:
# This error indicates a failure in the workflow logic or state propagation
# because this function should only be called when prereqs are met.
raise ValueError(f"Gemini API call prerequisites not met: api_key={api_key}, model_name={model_name}")
try:
genai.configure(api_key=api_key)
model = genai.GenerativeModel(model_name)
# Define tools for grounding if requested.
# Using genai.types.GoogleSearch() is recommended for Gemini 2.0+
# and is backwards compatible with 1.5 for retrieval.
tools_config = [types.Tool(google_search=types.GoogleSearch())] if use_grounding else None
# Using generate_content and stream=False for simplicity here
response = model.generate_content(
prompt,
stream=False,
tools=tools_config # Pass the tools configuration
)
# Check if response is blocked
if response.prompt_feedback and response.prompt_feedback.block_reason:
raise RuntimeError(f"Gemini API call blocked: {response.prompt_feedback.block_reason}")
if not response.candidates:
# Check for safety ratings if no candidates are returned but not blocked
if response.prompt_feedback and response.prompt_feedback.safety_ratings:
ratings = "; ".join([f"{r.category}: {r.probability}" for r in response.prompt_feedback.safety_ratings])
raise RuntimeError(f"Gemini API call returned no candidates. Safety ratings: {ratings}")
else:
raise RuntimeError("Gemini API call returned no candidates.")
# If response.candidates is not empty, get the text
# Using response.text is a convenient way to get text from the first candidate part
return response.text or "" # Return empty string if no text
except Exception as e:
# Re-raising as RuntimeError for the workflow to catch and manage
raise RuntimeError(f"Gemini API call failed: {e}")
# --- AI workflow logic (State Machine) ---
# Define States for the workflow
STATE_IDLE = "idle"
STATE_AWAITING_REPO_NAME = "awaiting_repo_name"
STATE_CREATING_SPACE = "creating_space"
STATE_GENERATING_CODE = "generating_code"
STATE_UPLOADING_APP_PY = "uploading_app_py"
STATE_GENERATING_REQUIREMENTS = "generating_requirements"
STATE_UPLOADING_REQUIREMENTS = "uploading_requirements"
STATE_GENERATING_README = "generating_readme"
STATE_UPLOADING_README = "uploading_readme"
STATE_CHECKING_LOGS_BUILD = "checking_logs_build"
STATE_CHECKING_LOGS_RUN = "checking_logs_run"
STATE_DEBUGGING_CODE = "debugging_code"
STATE_UPLOADING_FIXED_APP_PY = "uploading_fixed_app_py"
STATE_COMPLETE = "complete"
MAX_DEBUG_ATTEMPTS = 3 # Limit the number of automatic debug attempts
def add_bot_message(history: list[dict], bot_message: str) -> list[dict]:
"""Helper to add a new assistant message to the chatbot history."""
history.append({"role": "assistant", "content": bot_message})
return history
# Add an initial welcome message to the chatbot (defined outside Blocks to be called by load chain)
def greet():
# Updated welcome message to reflect the change in API key handling
return [{"role": "assistant", "content": "Welcome! Please log in to Hugging Face and provide your Google AI Studio API key to start building Spaces. Once ready, type 'generate me a gradio app called myapp' or 'create' to begin."}]
# Helper function to update send button interactivity based on prereqs
# This function has the clean signature it expects. Wrappers handle Gradio's argument passing.
def check_send_button_ready(hf_profile: gr.OAuthProfile | None, hf_token: gr.OAuthToken | None, gemini_key: str | None, gemini_model: str | None) -> gr.update:
"""Checks if HF login and Gemini configuration are complete and returns update for button interactivity."""
# --- START ENHANCED DEBUGGING LOGS ---
print("\n--- check_send_button_ready START ---")
print(f" Received hf_profile: {hf_profile is not None}")
print(f" Received hf_token: {hf_token is not None}")
# For api_key, print part of the key if not None/empty for verification
api_key_display = gemini_key[:5] if isinstance(gemini_key, str) and gemini_key else ('Empty String' if isinstance(gemini_key, str) and gemini_key == "" else 'None')
print(f" Received gemini_key: Value starts with '{api_key_display}'")
print(f" Received gemini_model: {gemini_model}")
# --- END ENHANCED DEBUGGING LOGS ---
is_logged_in = hf_profile is not None and hf_token is not None
# Use bool() check for simplicity - handles None and "" correctly
is_gemini_ready = bool(gemini_key) and bool(gemini_model)
is_ready = is_logged_in and is_gemini_ready
print(f"check_send_button_ready - HF Ready: {is_logged_in}, Gemini Ready: {is_gemini_ready}, Button Ready: {is_ready}")
print("--- check_send_button_ready END ---\n")
return gr.update(interactive=is_ready)
# --- Wrappers to handle Gradio's argument passing in event chains ---
# These wrappers accept whatever Gradio passes (*args, **kwargs) and call the target function
# with the specific arguments it expects, extracted from *args based on the expected call signature.
# Wrapper for functions called in .then() chains with specific inputs list: expects (prev_output, *input_values)
# e.g., .then(wrapper, inputs=[s1, s2]) -> wrapper receives (prev_out, s1_val, s2_val)
def wrapper_from_then_inputs(func, num_inputs):
def wrapped(*args, **kwargs):
# We expect num_inputs values at the end of *args, after prev_output (index 0)
if len(args) > num_inputs:
required_args = args[-num_inputs:]
try:
return func(*required_args)
except Exception as e:
print(f"Error calling wrapped function {func.__name__} with args {required_args}: {e}")
# Provide a fallback or re-raise depending on context
if func == configure_gemini: return f"❌ Error configuring Gemini: {e}"
if func == get_model_description: return f"Error getting description: {e}"
if func == check_send_button_ready: return gr.update(interactive=False)
raise # Re-raise if no specific fallback
else:
print(f"Warning: wrapper_from_then_inputs for {func.__name__} received unexpected args (expecting at least {num_inputs+1}): {args}")
# Provide a fallback or re-raise
if func == configure_gemini: return "❌ Error configuring Gemini: unexpected arguments received."
if func == get_model_description: return "No description available (unexpected arguments received)."
if func == check_send_button_ready: return gr.update(interactive=False)
raise ValueError(f"Unexpected args received for {func.__name__}: {args}")
return wrapped
# Wrapper for functions called by .change() trigger with specific inputs list: expects (changed_value, *input_values)
# e.g., component.change(wrapper, inputs=[s1, s2]) -> wrapper receives (changed_val, s1_val, s2_val)
def wrapper_from_change_inputs(func, num_inputs):
def wrapped(*args, **kwargs):
# We expect num_inputs values at the end of *args, after the changed_value (index 0)
if len(args) > num_inputs:
required_args = args[-num_inputs:]
try:
return func(*required_args)
except Exception as e:
print(f"Error calling wrapped function {func.__name__} with args {required_args}: {e}")
if func == check_send_button_ready: return gr.update(interactive=False)
raise # Re-raise if no specific fallback
else:
print(f"Warning: wrapper_from_change_inputs for {func.__name__} received unexpected args (expecting at least {num_inputs+1}): {args}")
if func == check_send_button_ready: return gr.update(interactive=False)
raise ValueError(f"Unexpected args received for {func.__name__}: {args}")
return wrapped
# Wrapper for functions called in .then() chains with inputs=None: expects (prev_output,)
# e.g., .then(wrapper, inputs=None) -> wrapper receives (prev_out,)
def wrapper_from_prev_output(func):
def wrapped(*args, **kwargs):
# We expect only prev_output, or sometimes nothing if the chain starts
if len(args) >= 0: # Just accept anything here
try:
# The target function expects 0 args, so call it with no args
return func()
except Exception as e:
print(f"Error calling wrapped function {func.__name__} with args {args}: {e}")
# Provide a fallback or re-raise
if func == greet: return [{"role": "assistant", "content": f"❌ Error loading initial message: {e}"}]
raise # Re-raise if no specific fallback
else:
print(f"Warning: wrapper_from_prev_output for {func.__name__} received unexpected args: {args}")
if func == greet: return [{"role": "assistant", "content": "❌ Error loading initial message: unexpected arguments received."}]
raise ValueError(f"Unexpected args received for {func.__name__}: {args}")
return wrapped
# Instantiate specific wrappers using the generic ones
wrapper_check_button_change = wrapper_from_change_inputs(check_send_button_ready, 4) # Expects (changed, s1, s2, s3, s4)
wrapper_check_button_then = wrapper_from_then_inputs(check_send_button_ready, 4) # Expects (prev_out, s1, s2, s3, s4)
wrapper_configure_gemini_then = wrapper_from_then_inputs(configure_gemini, 2) # Expects (prev_out, s1, s2) -> api_key, model_name
wrapper_get_model_description_then = wrapper_from_then_inputs(get_model_description, 1) # Expects (prev_out, s1) -> model_name
wrapper_greet_then = wrapper_from_prev_output(greet) # Expects (prev_out,), needs 0 args
# This is the main generator function for the workflow, triggered by the 'Send' button
# Inputs and Outputs list must match exactly. The generator receives values from the inputs list.
def ai_workflow_chat(
message: str,
history: list[dict],
hf_profile: gr.OAuthProfile | None,
hf_token: gr.OAuthToken | None,
# Pass gemini_api_key and gemini_model as inputs - these come from the State variables
gemini_api_key_state: str | None,
gemini_model_state: str | None,
repo_id_state: str | None,
workflow_state: str,
space_sdk: str,
# NOTE: UI component values are passed *by value* to the generator
preview_html: str, # Value from iframe HTML
container_logs: str, # Value from run_txt Textbox
build_logs: str, # Value from build_txt Textbox
debug_attempts_state: int,
app_description_state: str | None,
repo_name_state: str | None,
generated_code_state: str | None,
use_grounding_state: bool, # Value from use_grounding_checkbox
# Absorb potential extra args passed by Gradio event listeners (important for generators)
*args, # Generators might receive extra args, need to accept them but don't need to yield unless they are state
**kwargs # Generators might receive extra kwargs
) -> tuple[
list[dict], # 0: Updated chat history (for chatbot)
str | None, # 1: Updated repo_id (for repo_id state)
str, # 2: Updated workflow state (for workflow state)
str, # 3: Updated iframe HTML (for iframe UI component)
str, # 4: Updated container logs (for run_txt UI component)
str, # 5: Updated build logs (for build_txt UI component)
int, # 6: Updated debug attempts count (for debug_attempts state)
str | None, # 7: Updated app description (for app_description state)
str | None, # 8: Updated repo name (for repo_name_state state)
str | None, # 9: Updated generated code (for generated_code_state state)
bool, # 10: Updated use_grounding_state (for use_grounding_state state)
str | None, # 11: Explicitly yield gemini_api_key_state
str | None, # 12: Explicitly yield gemini_model_state
]:
"""
Generator function to handle the AI workflow state machine.
Each 'yield' pauses execution and sends values to update Gradio outputs/state.
"""
# Unpack state variables from Gradio State components passed as inputs
repo_id = repo_id_state
state = workflow_state
attempts = debug_attempts_state
app_desc = app_description_state
repo_name = repo_name_state
generated_code = generated_code_state
use_grounding = use_grounding_state # Unpack grounding state
# Use the input parameters for Gemini key/model directly in the generator
current_gemini_key = gemini_api_key_state
current_gemini_model = gemini_model_state
# --- START DEBUGGING ai_workflow_chat inputs ---
print("\n--- ai_workflow_chat START (Inputs received) ---")
print(f" message: {message}")
print(f" history len: {len(history)}")
print(f" hf_profile: {hf_profile is not None}")
print(f" hf_token: {hf_token is not None}")
api_key_display = current_gemini_key[:5] if isinstance(current_gemini_key, str) and current_gemini_key else ('Empty String' if isinstance(current_gemini_key, str) and current_gemini_key == "" else 'None')
print(f" gemini_api_key_state: Value starts with '{api_key_display}'")
print(f" gemini_model_state: {current_gemini_model}")
print(f" repo_id_state: {repo_id_state}") # Check value here
print(f" workflow_state: {workflow_state}")
print(f" space_sdk: {space_sdk}")
print(f" use_grounding_state: {use_grounding_state}")
print(f" debug_attempts_state: {debug_attempts_state}")
print(f" app_description_state: {app_description_state}")
print(f" repo_name_state: {repo_name_state}")
print(f" generated_code_state: {'Present' if generated_code_state is not None else 'None'}")
print(f" *args (unexpected by generator): {args}") # Added debug for unexpected args
print(f" **kwargs (unexpected by generator): {kwargs}") # Added debug for unexpected kwargs
print("--- END DEBUGGING ai_workflow_chat inputs ---\n")
# Keep copies of potentially updated UI elements passed as inputs to update them later
# These are the *current values* of the UI components as of the button click
updated_preview = preview_html
updated_build = build_logs
updated_run = container_logs
# Add the user's message to the chat history immediately
user_message_entry = {"role": "user", "content": message}
# Add username if logged in (optional, but nice)
if hf_profile and hf_profile.username:
user_message_entry["name"] = hf_profile.username
history.append(user_message_entry)
# Yield immediately to update the chat UI with the user's message
# This provides immediate feedback to the user while the AI processes
# Ensure all state variables and UI outputs are yielded back in the correct order
# Include gemini_api_key_state and gemini_model_state in the yield tuple
# The yielded tuple must match the send_btn.click outputs list exactly.
yield (history, repo_id, state, updated_preview, updated_run, updated_build,
attempts, app_desc, repo_name, generated_code, use_grounding,
current_gemini_key, current_gemini_model) # Explicitly pass back current state values
try:
# --- State Machine Logic based on the current 'state' variable ---
# Although button interactivity prevents reaching here without key/model,
# the checks remain as a safeguard for the workflow logic itself.
# Use the local variables derived from state inputs (current_gemini_key, current_gemini_model)
if not (hf_profile and hf_token):
history = add_bot_message(history, "Workflow paused: Please log in to Hugging Face first.")
# Re-yield state to update chat and keep current state values
yield (history, repo_id, state, updated_preview, updated_run, updated_build,
attempts, app_desc, repo_name, generated_code, use_grounding,
current_gemini_key, current_gemini_model)
return # Stop workflow execution for this click
if not (isinstance(current_gemini_key, str) and current_gemini_key != "" and current_gemini_model):
history = add_bot_message(history, "Workflow cannot start: Please ensure your Gemini API key is entered and a model is selected.")
# Re-yield state to update chat and keep current state values
yield (history, repo_id, state, updated_preview, updated_run, updated_build,
attempts, app_desc, repo_name, generated_code, use_grounding,
current_gemini_key, current_gemini_model) # Correctly yielding back the received (likely empty/None) Gemini states
return # Stop workflow execution for this click
if state == STATE_IDLE:
# Look for specific commands in the user's message
reset_match = "reset" in message.lower()
# Capture app description AND repo name using regex
generate_match = re.search(r'generate (?:me )?(?:a|an) (.+) app called (\w+)', message, re.I)
# Capture repo name for a simple 'create space' command
create_match = re.search(r'create (?:a|an)? space called (\w+)', message, re.I)
if reset_match:
history = add_bot_message(history, "Workflow reset.")
# Reset relevant states and UI outputs, passing through current Gemini state
yield (history, None, STATE_IDLE, "<p>No Space created yet.</p>", "", "", 0,
None, None, None, False, current_gemini_key, current_gemini_model) # Correctly yielding back the received Gemini states
elif generate_match:
# User requested generation with description and name
new_app_desc = generate_match.group(1).strip() # Capture description part
new_repo_name = generate_match.group(2).strip() # Capture name part
history = add_bot_message(history, f"Acknowledged: '{message}'. Starting workflow to create Space `{hf_profile.username}/{new_repo_name}` for a '{new_app_desc}' app.")
# Update state variables for the next step (creation)
state = STATE_CREATING_SPACE
repo_name = new_repo_name
app_desc = new_app_desc
# Yield updated state variables, passing others through
yield (history, repo_id, state, updated_preview, updated_run, updated_build,
attempts, app_desc, repo_name, generated_code, use_grounding,
current_gemini_key, current_gemini_model) # Correctly yielding back the received Gemini states
elif create_match:
# User requested simple space creation with a name
new_repo_name = create_match.group(1).strip()
history = add_bot_message(history, f"Acknowledged: '{message}'. Starting workflow to create Space `{hf_profile.username}/{new_repo_name}`.")
state = STATE_CREATING_SPACE # Transition state to creation
repo_name = new_repo_name # Store the validated repo name
# Yield updated state variables, passing others through
yield (history, repo_id, state, updated_preview, updated_run, updated_build,
attempts, app_desc, repo_name, generated_code, use_grounding,
current_gemini_key, current_gemini_model) # Correctly yielding back the received Gemini states
elif "create" in message.lower() and not repo_id:
# User wants to create but didn't specify a name yet
history = add_bot_message(history, "Okay, what should the Space be called? (e.g., `my-awesome-app`)")
state = STATE_AWAITING_REPO_NAME # Transition to the state where we wait for the name
# Yield updated state, passing others through
yield (history, repo_id, state, updated_preview, updated_run, updated_build,
attempts, app_desc, repo_name, generated_code, use_grounding,
current_gemini_key, current_gemini_model) # Correctly yielding back the received Gemini states
else:
# Command not recognized in IDLE state
history = add_bot_message(history, "Command not recognized. Try 'generate me a gradio app called myapp', or 'reset'.")
# Yield current state, passing others through
yield (history, repo_id, state, updated_preview, updated_run, updated_build,
attempts, app_desc, repo_name, generated_code, use_grounding,
current_gemini_key, current_gemini_model) # Correctly yielding back the received Gemini states
elif state == STATE_AWAITING_REPO_NAME:
# User's message is expected to be the repo name
new_repo_name = message.strip()
# Basic validation for Hugging Face repo name format
# Allow letters, numbers, hyphens, underscores, max 100 chars (HF limit check)
if not new_repo_name or re.search(r'[^a-zA-Z0-9_-]', new_repo_name) or len(new_repo_name) > 100:
history = add_bot_message(history, "Invalid name. Please provide a single word/slug for the Space name (letters, numbers, underscores, hyphens only, max 100 chars).")
# Stay in AWAITING_REPO_NAME state and yield message (pass UI outputs through)
yield (history, repo_id, state, updated_preview, updated_run, updated_build,
attempts, app_desc, repo_name, generated_code, use_grounding,
current_gemini_key, current_gemini_model) # Correctly yielding back the received Gemini states
else:
history = add_bot_message(history, f"Using Space name `{new_repo_name}`. Creating Space `{hf_profile.username}/{new_repo_name}`...")
state = STATE_CREATING_SPACE # Transition state to creation
repo_name = new_repo_name # Store the validated repo name
# Yield updated state variables, passing others through
yield (history, repo_id, state, updated_preview, updated_run, updated_build,
attempts, app_desc, repo_name, generated_code, use_grounding,
current_gemini_key, current_gemini_model) # Correctly yielding back the received Gemini states
elif state == STATE_CREATING_SPACE:
# Ensure repo_name is available (it should have been set in the previous step)
if not repo_name:
history = add_bot_message(history, "Internal error: Repo name missing for creation. Resetting.")
# Reset state on error, passing through current Gemini state
yield (history, None, STATE_IDLE, "<p>Error creating space.</p>", "", "", 0,
None, None, None, use_grounding, current_gemini_key, current_gemini_model) # Correctly yielding back the received Gemini states
else:
try:
new_repo_id, iframe_html = create_space_action(repo_name, space_sdk, hf_profile, hf_token)
updated_preview = iframe_html
repo_id = new_repo_id # Update repo_id state variable
history = add_bot_message(history, f"✅ Space `{repo_id}` created. Click 'Send' to generate and upload code.")
state = STATE_GENERATING_CODE
# Yield updated state variables, passing others through
yield (history, repo_id, state, updated_preview, updated_run, updated_build,
attempts, app_desc, repo_name, generated_code, use_grounding,
current_gemini_key, current_gemini_model) # Correctly yielding back the received Gemini states
except Exception as e:
history = add_bot_message(history, f"❌ Error creating space: {e}. Click 'reset'.")
# Reset state on error, passing through current Gemini state
yield (history, None, STATE_IDLE, "<p>Error creating space.</p>", "", "", 0,
None, None, None, use_grounding, current_gemini_key, current_gemini_model) # Correctly yielding back the received Gemini states
elif state == STATE_GENERATING_CODE:
# Define the prompt for Gemini based on the app description or a default
prompt_desc = app_desc if app_desc else f'a simple {space_sdk} app'
prompt = f"""
You are an AI assistant specializing in Hugging Face Spaces using the {space_sdk} SDK.
Generate a full, single-file Python app based on:
'{prompt_desc}'
Ensure the code is runnable as `app.py` in a Hugging Face Space using the `{space_sdk}` SDK. Include necessary imports and setup.
Return **only** the python code block for `app.py`. Do not include any extra text, explanations, or markdown outside the code block.
"""
try:
history = add_bot_message(history, f"🧠 Generating `{prompt_desc}` `{space_sdk}` app (`app.py`) code with Gemini...")
if use_grounding:
history = add_bot_message(history, "(Using Grounding with Google Search)")
# Yield message before API call
yield (history, repo_id, state, updated_preview, updated_run, updated_build,
attempts, app_desc, repo_name, generated_code, use_grounding,
current_gemini_key, current_gemini_model) # Correctly yielding back the received Gemini states
code = call_gemini(prompt, current_gemini_key, current_gemini_model, use_grounding=use_grounding)
code = code.strip()
# Clean up markdown
if code.startswith("```python"): code = code[len("```python"):].strip()
if code.startswith("```"): code = code[len("```"):].strip()
if code.endswith("```"): code = code[:-len("```")].strip()
if not code: raise ValueError("Gemini returned empty code.")
history = add_bot_message(history, "✅ `app.py` code generated. Click 'Send' to upload.")
state = STATE_UPLOADING_APP_PY
generated_code = code
# Yield updated state variables
yield (history, repo_id, state, updated_preview, updated_run, updated_build,
attempts, app_desc, repo_name, generated_code, use_grounding,
current_gemini_key, current_gemini_model) # Correctly yielding back the received Gemini states
except Exception as e:
history = add_bot_message(history, f"❌ Error generating code: {e}. Click 'reset'.")
# Reset state on error, passing through current Gemini state
yield (history, None, STATE_IDLE, updated_preview, updated_run, updated_build, 0,
None, None, None, use_grounding, current_gemini_key, current_gemini_model) # Correctly yielding back the received Gemini states
elif state == STATE_UPLOADING_APP_PY:
# Retrieve the generated code from the state variable
code_to_upload = generated_code
if not code_to_upload:
history = add_bot_message(history, "Internal error: No code to upload. Resetting.")
yield (history, None, STATE_IDLE, updated_preview, updated_run, updated_build, 0,
None, None, None, use_grounding, current_gemini_key, current_gemini_model) # Correctly yielding back the received Gemini states
else:
history = add_bot_message(history, "☁️ Uploading `app.py`...")
# Yield message before upload
yield (history, repo_id, state, updated_preview, updated_run, updated_build,
attempts, app_desc, repo_name, generated_code, use_grounding,
current_gemini_key, current_gemini_model) # Correctly yielding back the received Gemini states
try:
upload_file_to_space_action(io.StringIO(code_to_upload), "app.py", repo_id, hf_profile, hf_token)
history = add_bot_message(history, "✅ Uploaded `app.py`. Click 'Send' to generate requirements.")
state = STATE_GENERATING_REQUIREMENTS
generated_code = None
# Yield updated state variables
yield (history, repo_id, state, updated_preview, updated_run, updated_build,
attempts, app_desc, repo_name, generated_code, use_grounding,
current_gemini_key, current_gemini_model) # Correctly yielding back the received Gemini states
except Exception as e:
history = add_bot_message(history, f"❌ Error uploading `app.py`: {e}. Click 'reset'.")
# Reset state on error, passing through current Gemini state
yield (history, None, STATE_IDLE, updated_preview, updated_run, updated_build, 0,
None, None, None, use_grounding, current_gemini_key, current_gemini_model) # Correctly yielding back the received Gemini states
elif state == STATE_GENERATING_REQUIREMENTS:
history = add_bot_message(history, "📄 Generating `requirements.txt`...")
# Yield message before generating requirements
yield (history, repo_id, state, updated_preview, updated_run, updated_build,
attempts, app_desc, repo_name, generated_code, use_grounding,
current_gemini_key, current_gemini_model) # Correctly yielding back the received Gemini states
# Logic to determine required packages based on SDK and keywords in the app description
reqs_list = ["gradio"] if space_sdk == "gradio" else ["streamlit"]
# Add essential libraries regardless of description keywords or grounding
essential_libs = ["google-generativeai", "huggingface_hub"]
# Only add if Gemini is actually needed for the app (determined by description or if key is present)
# If we are here, key and model are available based on STATE_IDLE checks
reqs_list.extend(essential_libs)
# Add common libraries if description suggests they might be needed
if app_desc:
app_desc_lower = app_desc.lower()
if "requests" in app_desc_lower or "api" in app_desc_lower:
reqs_list.append("requests")
# Image processing libraries
if "image" in app_desc_lower or "upload" in app_desc_lower or "blur" in app_desc_lower or "vision" in app_desc_lower or "photo" in app_desc_lower:
reqs_list.append("Pillow")
if "numpy" in app_desc_lower: reqs_list.append("numpy")
if "pandas" in app_desc_lower or "dataframe" in app_desc_lower: reqs_list.append("pandas")
# Add scikit-image and opencv if image processing is heavily implied
if any(lib in app_desc_lower for lib in ["scikit-image", "skimage", "cv2", "opencv"]):
reqs_list.extend(["scikit-image", "opencv-python"]) # Note: opencv-python for pip
# Add transformers if large models are implied
if any(lib in app_desc_lower for lib in ["transformer", "llama", "mistral", "bert", "gpt2"]):
reqs_list.append("transformers")
# Add torch or tensorflow if deep learning frameworks are implied
if any(lib in app_desc_lower for lib in ["torch", "pytorch", "tensorflow", "keras"]):
reqs_list.extend(["torch", "tensorflow"]) # Users might need specific versions, but this is a start
# Use dict.fromkeys to get unique items while preserving insertion order (Python 3.7+)
reqs_list = list(dict.fromkeys(reqs_list))
# Sort alphabetically for cleaner requirements.txt
reqs_list.sort()
reqs_content = "\n".join(reqs_list) + "\n"
history = add_bot_message(history, "✅ `requirements.txt` generated. Click 'Send' to upload.")
state = STATE_UPLOADING_REQUIREMENTS # Transition state
generated_code = reqs_content # Store requirements content
# Yield updated state variables and history (pass UI outputs and other states through)
yield (history, repo_id, state, updated_preview, updated_run, updated_build,
attempts, app_desc, repo_name, generated_code, use_grounding,
current_gemini_key, current_gemini_model) # Correctly yielding back the received Gemini states
elif state == STATE_UPLOADING_REQUIREMENTS:
# Retrieve requirements content from state variable
reqs_content_to_upload = generated_code
if not reqs_content_to_upload:
history = add_bot_message(history, "Internal error: No requirements content to upload. Resetting.")
yield (history, None, STATE_IDLE, updated_preview, updated_run, updated_build, 0,
None, None, None, use_grounding, current_gemini_key, current_gemini_model) # Correctly yielding back the received Gemini states
else:
history = add_bot_message(history, "☁️ Uploading `requirements.txt`...")
# Yield message before upload (pass UI outputs and states through)
yield (history, repo_id, state, updated_preview, updated_run, updated_build,
attempts, app_desc, repo_name, generated_code, use_grounding,
current_gemini_key, current_gemini_model) # Correctly yielding back the received Gemini states
try:
# Perform requirements file upload
upload_file_to_space_action(io.StringIO(reqs_content_to_upload), "requirements.txt", repo_id, hf_profile, hf_token)
history = add_bot_message(history, "✅ Uploaded `requirements.txt`. Click 'Send' to generate README.")
state = STATE_GENERATING_README # Transition state
generated_code = None # Clear content after use
# Yield updated state variables and history (pass UI outputs and other states through)
yield (history, repo_id, state, updated_preview, updated_run, updated_build,
attempts, app_desc, repo_name, generated_code, use_grounding,
current_gemini_key, current_gemini_model) # Correctly yielding back the received Gemini states
except Exception as e:
history = add_bot_message(history, f"❌ Error uploading `requirements.txt`: {e}. Click 'reset'.")
# Yield error message and reset state on failure
yield (history, None, STATE_IDLE, updated_preview, updated_run, updated_build, 0,
None, None, None, use_grounding, current_gemini_key, current_gemini_model) # Correctly yielding back the received Gemini states
elif state == STATE_GENERATING_README:
history = add_bot_message(history, "📝 Generating `README.md`...")
# Yield message before generating README (pass UI outputs and states through)
yield (history, repo_id, state, updated_preview, updated_run, updated_build,
attempts, app_desc, repo_name, generated_code, use_grounding,
current_gemini_key, current_gemini_model) # Correctly yielding back the received Gemini states
# Generate simple README content with Space metadata header
readme_title = repo_name if repo_name else "My Awesome Space"
readme_description = app_desc if app_desc else f"This Hugging Face Space hosts an AI-generated {space_sdk} application."
readme_content = f"""---
title: {readme_title}
emoji: 🚀
colorFrom: blue
colorTo: yellow
sdk: {space_sdk}
app_file: app.py
pinned: false
---
# {readme_title}
{readme_description}
This Space was automatically generated by an AI workflow using Google Gemini and Gradio.
""" # Added Space metadata header and slightly improved content
history = add_bot_message(history, "✅ `README.md` generated. Click 'Send' to upload.")
state = STATE_UPLOADING_README # Transition state
generated_code = readme_content # Store README content
# Yield updated state variables and history (pass UI outputs and other states through)
yield (history, repo_id, state, updated_preview, updated_run, updated_build,
attempts, app_desc, repo_name, generated_code, use_grounding,
current_gemini_key, current_gemini_model) # Correctly yielding back the received Gemini states
elif state == STATE_UPLOADING_README:
# Retrieve README content from state variable
readme_content_to_upload = generated_code
if not readme_content_to_upload:
history = add_bot_message(history, "Internal error: No README content to upload. Resetting.")
yield (history, None, STATE_IDLE, updated_preview, updated_run, updated_build, 0,
None, None, None, use_grounding, current_gemini_key, current_gemini_model) # Correctly yielding back the received Gemini states
else:
history = add_bot_message(history, "☁️ Uploading `README.md`...")
# Yield message before upload (pass UI outputs and states through)
yield (history, repo_id, state, updated_preview, updated_run, updated_build,
attempts, app_desc, repo_name, generated_code, use_grounding,
current_gemini_key, current_gemini_model) # Correctly yielding back the received Gemini states
try:
# Perform README file upload
upload_file_to_space_action(io.StringIO(readme_content_to_upload), "README.md", repo_id, hf_profile, hf_token)
history = add_bot_message(history, "✅ Uploaded `README.md`. All files uploaded. Space is now building. Click 'Send' to check build logs.")
state = STATE_CHECKING_LOGS_BUILD # Transition to checking build logs
generated_code = None # Clear content after use
# Yield updated state variables and history (pass UI outputs and other states through)
yield (history, repo_id, state, updated_preview, updated_run, updated_build,
attempts, app_desc, repo_name, generated_code, use_grounding,
current_gemini_key, current_gemini_model) # Correctly yielding back the received Gemini states
except Exception as e:
history = add_bot_message(history, f"❌ Error uploading `README.md`: {e}. Click 'reset'.")
# Yield error message and reset state on failure
yield (history, None, STATE_IDLE, updated_preview, updated_run, updated_build, 0,
None, None, None, use_grounding, current_gemini_key, current_gemini_model) # Correctly yielding back the received Gemini states
elif state == STATE_CHECKING_LOGS_BUILD:
history = add_bot_message(history, "🔍 Fetching build logs...")
# Yield message before fetching logs (which includes a delay) (pass UI outputs and states through)
yield (history, repo_id, state, updated_preview, updated_run, updated_build,
attempts, app_desc, repo_name, generated_code, use_grounding,
current_gemini_key, current_gemini_model) # Correctly yielding back the received Gemini states
# Fetch build logs from HF Space
build_logs_text = get_build_logs_action(repo_id, hf_profile, hf_token)
updated_build = build_logs_text # Update the logs display variable
# Simple check for common error indicators in logs (case-insensitive)
if "error" in updated_build.lower() or "exception" in updated_build.lower() or "build failed" in updated_build.lower():
history = add_bot_message(history, "⚠️ Build logs indicate potential issues. Please inspect above. Click 'Send' to check container logs (app might still start despite build warnings).")
state = STATE_CHECKING_LOGS_RUN # Transition even on build error, to see if container starts
# Yield updated state, logs, and variables
yield (history, repo_id, state, updated_preview, updated_run, updated_build,
attempts, app_desc, repo_name, generated_code, use_grounding,
current_gemini_key, current_gemini_model) # Correctly yielding back the received Gemini states
else:
history = add_bot_message(history, "✅ Build logs fetched. Click 'Send' to check container logs.")
state = STATE_CHECKING_LOGS_RUN # Transition to next log check
# Yield updated state, logs, and variables
yield (history, repo_id, state, updated_preview, updated_run, updated_build,
attempts, app_desc, repo_name, generated_code, use_grounding,
current_gemini_key, current_gemini_model) # Correctly yielding back the received Gemini states
elif state == STATE_CHECKING_LOGS_RUN:
history = add_bot_message(history, "🔍 Fetching container logs...")
# Yield message before fetching logs (includes a delay) (pass UI outputs and states through)
yield (history, repo_id, state, updated_preview, updated_run, updated_build,
attempts, app_desc, repo_name, generated_code, use_grounding,
current_gemini_key, current_gemini_model) # Correctly yielding back the received Gemini states
# Fetch container logs from HF Space
container_logs_text = get_container_logs_action(repo_id, hf_profile, hf_token)
updated_run = container_logs_text # Update the logs display variable
# Check for errors in run logs and if we have debug attempts left
if ("error" in updated_run.lower() or "exception" in updated_run.lower()) and attempts < MAX_DEBUG_ATTEMPTS:
attempts += 1 # Increment debug attempts counter
history = add_bot_message(history, f"❌ Errors detected in container logs. Attempting debug fix #{attempts}/{MAX_DEBUG_ATTEMPTS}. Click 'Send' to proceed.")
state = STATE_DEBUGGING_CODE # Transition to the debugging state
# Yield updated state, logs, attempts, and variables
yield (history, repo_id, state, updated_preview, updated_run, updated_build,
attempts, app_desc, repo_name, generated_code, use_grounding,
current_gemini_key, current_gemini_model) # Correctly yielding back the received Gemini states
elif ("error" in updated_run.lower() or "exception" in updated_run.lower()) and attempts >= MAX_DEBUG_ATTEMPTS:
# Max debug attempts reached
history = add_bot_message(history, f"❌ Errors detected in container logs. Max debug attempts ({MAX_DEBUG_ATTEMPTS}) reached. Please inspect logs manually or click 'reset'.")
state = STATE_COMPLETE # Workflow ends on failure after attempts
# Yield updated state, logs, attempts, and variables
yield (history, repo_id, state, updated_preview, updated_run, updated_build,
attempts, app_desc, repo_name, generated_code, use_grounding,
current_gemini_key, current_gemini_model) # Correctly yielding back the received Gemini states
else:
# No significant errors found in logs, assume success
history = add_bot_message(history, "✅ App appears to be running successfully! Check the iframe above. Click 'reset' to start a new project.")
state = STATE_COMPLETE # Workflow ends on success
# Yield updated state, logs, attempts, and variables
yield (history, repo_id, state, updated_preview, updated_run, updated_build,
attempts, app_desc, repo_name, generated_code, use_grounding,
current_gemini_key, current_gemini_model) # Correctly yielding back the received Gemini states
elif state == STATE_DEBUGGING_CODE:
history = add_bot_message(history, f"🧠 Calling Gemini to generate fix based on logs...")
if use_grounding:
history = add_bot_message(history, "(Using Grounding with Google Search)")
# Yield message before Gemini API call (pass UI outputs and states through)
yield (history, repo_id, state, updated_preview, updated_run, updated_build,
attempts, app_desc, repo_name, generated_code, use_grounding,
current_gemini_key, current_gemini_model) # Correctly yielding back the received Gemini states
# Construct prompt for Gemini including the container logs
debug_prompt = f"""
You are debugging a {space_sdk} Space. The goal is to fix the code in `app.py` based on the container logs provided.
Here are the container logs:
Use code with caution.
Python
{updated_run}
Generate the *complete, fixed* content for `app.py` based on these logs.
Return **only** the python code block for app.py. Do not include any extra text, explanations, or markdown outside the code block.
"""
try:
# Call Gemini to generate the corrected code, optionally using grounding
# Note: Grounding might be less effective for debugging based *only* on logs,
# but we include the option as requested.
# Use the current_gemini_key and current_gemini_model derived from state inputs
fix_code = call_gemini(debug_prompt, current_gemini_key, current_gemini_model, use_grounding=use_grounding)
fix_code = fix_code.strip()
# Clean up potential markdown formatting
if fix_code.startswith("```python"):
fix_code = fix_code[len("```python"):].strip()
if fix_code.startswith("```"):
fix_code = fix_code[len("```"):].strip()
if fix_code.endswith("```"):
fix_code = fix_code[:-len("```")].strip()
if not fix_code:
raise ValueError("Gemini returned empty fix code.")
history = add_bot_message(history, "✅ Fix code generated. Click 'Send' to upload.")
state = STATE_UPLOADING_FIXED_APP_PY # Transition to the upload state for the fix
generated_code = fix_code # Store the generated fix code
# Yield updated state, code, and variables (pass UI outputs and states through)
yield (history, repo_id, state, updated_preview, updated_run, updated_build,
attempts, app_desc, repo_name, generated_code, use_grounding,
current_gemini_key, current_gemini_model) # Correctly yielding back the received Gemini states
except Exception as e:
history = add_bot_message(history, f"❌ Error generating debug code: {e}. Click 'reset'.")
# Yield error message and reset state on failure
yield (history, None, STATE_IDLE, updated_preview, updated_run, updated_build, 0,
None, None, None, use_grounding,
current_gemini_key, current_gemini_model) # Correctly yielding back the received Gemini states
elif state == STATE_UPLOADING_FIXED_APP_PY:
# Retrieve the fixed code from the state variable
fixed_code_to_upload = generated_code
if not fixed_code_to_upload:
history = add_bot_message(history, "Internal error: No fixed code available to upload. Resetting.")
yield (history, None, STATE_IDLE, updated_preview, updated_run, updated_build, 0,
None, None, None, use_grounding, current_gemini_key, current_gemini_model) # Correctly yielding back the received Gemini states
else:
history = add_bot_message(history, "☁️ Uploading fixed `app.py`...")
# Yield message before upload (pass UI outputs and states through)
yield (history, repo_id, state, updated_preview, updated_run, updated_build,
attempts, app_desc, repo_name, generated_code, use_grounding,
current_gemini_key, current_gemini_model) # Correctly yielding back the received Gemini states
try:
# Perform the upload of the fixed app.py
upload_file_to_space_action(io.StringIO(fixed_code_to_upload), "app.py", repo_id, hf_profile, hf_token)
history = add_bot_message(history, "✅ Fixed `app.py` uploaded. Space will rebuild. Click 'Send' to check logs again.")
state = STATE_CHECKING_LOGS_RUN # Go back to checking run logs after uploading the fix
generated_code = None # Clear code after use
# Yield updated state, code, and variables (pass UI outputs and states through)
yield (history, repo_id, state, updated_preview, updated_run, updated_build,
attempts, app_desc, repo_name, generated_code, use_grounding,
current_gemini_key, current_gemini_model) # Correctly yielding back the received Gemini states
except Exception as e:
history = add_bot_message(history, f"❌ Error uploading fixed `app.py`: {e}. Click 'reset'.")
# Yield error message and reset state on failure
yield (history, None, STATE_IDLE, updated_preview, updated_run, updated_build, 0,
None, None, None, use_grounding, current_gemini_key, current_gemini_model) # Correctly yielding back the received Gemini states
elif state == STATE_COMPLETE:
# If in the complete state, the workflow is finished for this project.
# Subsequent clicks just add user messages; we simply yield the current state.
yield (history, repo_id, state, updated_preview, updated_run, updated_build,
attempts, app_desc, repo_name, generated_code, use_grounding,
current_gemini_key, current_gemini_model) # Correctly yielding back the received Gemini states
except Exception as e:
# This catches any unexpected errors that occur within any state's logic
error_message = f"Workflow step failed unexpectedly ({state}): {e}. Click 'Send' to re-attempt this step or 'reset'."
history = add_bot_message(history, error_message)
print(f"Critical Error in state {state}: {e}") # Log the error for debugging purposes
# On unexpected error, reset to IDLE, but pass through the current Gemini state
yield (history, None, STATE_IDLE, updated_preview, updated_run, updated_build, 0,
None, None, None, use_grounding,
current_gemini_key, current_gemini_model) # Correctly yielding back the received Gemini states
# --- Build the Gradio UI ---
with gr.Blocks(title="AI-Powered HF Space App Builder") as ai_builder_tab:
# Gradio State variables - these persist their values across user interactions (clicks)
# Define these first as they might be used in default values for components
hf_profile = gr.State(None)
hf_token = gr.State(None)
# Initialize gemini_api_key_state to empty string
gemini_api_key_state = gr.State("") # start with no key
# Initialize gemini_model_state with the default model key
gemini_model_state = gr.State(DEFAULT_GEMINI_MODEL) # Default selected model
repo_id = gr.State(None) # Stores the ID of the created Space
workflow = gr.State(STATE_IDLE) # Stores the current state of the AI workflow
sdk_state = gr.State("gradio") # Stores the selected Space SDK (Gradio or Streamlit)
debug_attempts = gr.State(0) # Counter for how many debugging attempts have been made
app_description = gr.State(None) # Stores the user's initial description of the desired app
repo_name_state = gr.State(None) # Stores the chosen repository name for the Space
generated_code_state = gr.State(None) # Temporary storage for generated file content (app.py, reqs, README)
# New State variable for grounding checkbox
use_grounding_state = gr.State(False)
with gr.Row():
# Sidebar column for inputs and status displays
with gr.Column(scale=1, min_width=300):
gr.Markdown("## Hugging Face Login")
# Define login_status before it's used in login_btn.click outputs
login_status = gr.Markdown("*Not logged in.*")
# Hugging Face Login Button
login_btn = gr.LoginButton(variant="huggingface")
gr.Markdown("## Google AI Studio / Gemini")
# Define gemini_input and gemini_status before they are used in change handlers
# Blank out textbox on load and update info text
gemini_input = gr.Textbox(
label="Your Google AI Studio API Key", # Changed label
type="password", # Hides input for security
interactive=True,
value="", # Don't pre-fill from env var
info="Enter your own key here" # Updated info text
)
gemini_status = gr.Markdown("") # Display Gemini configuration status
# Define model_selector before it's used in its change handler
model_selector = gr.Radio(
# Use the list of choices generated from the GEMINI_MODELS dictionary
choices=GEMINI_MODEL_CHOICES,
value=DEFAULT_GEMINI_MODEL, # Default selection using the key
label="Select model",
interactive=True
)
# Add a markdown field to display the model description
# Initialize with the description of the default model
model_description_text = gr.Markdown(get_model_description(DEFAULT_GEMINI_MODEL))
# Define grounding checkbox before its change handler
use_grounding_checkbox = gr.Checkbox(
label="Enable Grounding with Google Search",
value=False, # Default to off
interactive=True,
info="Use Google Search results to inform Gemini's response (may improve factuality)."
)
gr.Markdown("## Space SDK")
# Define sdk_selector before its change handler
sdk_selector = gr.Radio(choices=["gradio","streamlit"], value="gradio", label="Template SDK", interactive=True)
gr.Markdown("## Workflow Status")
# Define status_text and repo_id_text before they are used in change handlers
status_text = gr.Textbox(label="Current State", value=STATE_IDLE, interactive=False)
repo_id_text = gr.Textbox(label="Current Space ID", value="None", interactive=False)
# Main content area column
with gr.Column(scale=3):
# Define chatbot, user_input, send_btn before send_btn.click
chatbot = gr.Chatbot(type='messages', label="AI Workflow Chat")
user_input = gr.Textbox(placeholder="Type your message…", interactive=True)
# Define send_btn here, BEFORE it's used in send_button_update_output
send_btn = gr.Button("Send", interactive=False) # Start disabled by default
# Define iframe, build_txt, run_txt after send_btn
# These are UI components, NOT State variables
iframe = gr.HTML("<p>No Space created yet.</p>") # HTML element for the Space iframe
build_txt = gr.Textbox(label="Build Logs", lines=10, interactive=False, value="", max_lines=20) # Set max_lines for scrollability
run_txt = gr.Textbox(label="Container Logs", lines=10, interactive=False, value="", max_lines=20) # Set max_lines for scrollability
# --- Define Event Handlers and Chains AFTER all components and required lists are defined ---
# Define the inputs used for checking prerequisites (These are State components)
send_button_interactive_binding_inputs = [
hf_profile,
hf_token,
gemini_api_key_state,
gemini_model_state
]
# Define the output for updating the send button interactivity
send_button_update_output = [send_btn]
# Trigger check_send_button_ready whenever any prerequisite state changes
# Use the specific change wrapper which expects (changed_value, *input_values)
hf_profile.change(
wrapper_check_button_change,
inputs=send_button_interactive_binding_inputs, # Pass all 4 prerequisite states
outputs=send_button_update_output, # Update only the send button
)
hf_token.change(
wrapper_check_button_change,
inputs=send_button_interactive_binding_inputs,
outputs=send_button_update_output,
)
# When gemini_api_key_state changes (updated by gemini_input.change), check button readiness
gemini_api_key_state.change(
wrapper_check_button_change,
inputs=send_button_interactive_binding_inputs,
outputs=send_button_update_output,
)
# When gemini_model_state changes (updated by model_selector.change), check button readiness
gemini_model_state.change(
wrapper_check_button_change,
inputs=send_button_interactive_binding_inputs,
outputs=send_button_update_output,
)
# Handle login button click: Update profile/token state -> Their .change handlers trigger check_send_button_ready
login_btn.click(
lambda x: (x[0], x[1]), # Lambda takes the LoginButton output (profile, token tuple) and returns it
inputs=[login_btn], # Pass the LoginButton itself to get its output
outputs=[hf_profile, hf_token] # Update state variables
)
# Handle Gemini Key Input change: Update key state -> Configure Gemini status -> Update send button state
gemini_input.change(
# Lambda receives the new value of gemini_input (1 arg) because inputs=None (implied)
lambda new_key_value: new_key_value,
inputs=None, # Only need the new value of the changed component
outputs=[gemini_api_key_state] # This output becomes the implicit first arg for the next .then() in this chain
).then(
# Configure Gemini using the updated state variables
# Use the then_inputs wrapper which expects (prev_output, api_key_val, model_name_val)
wrapper_configure_gemini_then,
inputs=[gemini_api_key_state, gemini_model_state], # Explicitly pass the required states
outputs=[gemini_status] # Update Gemini status display. This output becomes the implicit first arg for the next .then()
)
# The .then chain continues from the outputs of the configure_gemini call, handled by gemini_api_key_state.change handler above
# Handle Gemini Model Selector change: Update model state -> Update description -> Configure Gemini status -> Update send button state
model_selector.change(
# Lambda receives the new value of model_selector (1 arg)
lambda new_model_name: new_model_name,
inputs=None, # Only need the new value of the changed component
outputs=[gemini_model_state] # This output becomes the implicit first arg for the next .then()
).then(
# Update the model description display
# Use the then_inputs wrapper which expects (prev_output, model_name_val)
wrapper_get_model_description_then,
inputs=[gemini_model_state], # Get the new state value
outputs=[model_description_text] # Update description UI. This output becomes implicit first arg for next .then()
).then(
# Configure Gemini using the updated state variables
# Use the then_inputs wrapper which expects (prev_output, api_key_val, model_name_val)
wrapper_configure_gemini_then,
inputs=[gemini_api_key_state, gemini_model_state], # Explicitly pass the required states
outputs=[gemini_status] # Update Gemini status display. This output becomes the implicit first arg for the next .then()
)
# The .then chain continues from the outputs of the configure_gemini call, handled by gemini_model_state.change handler above
# Handle Grounding checkbox change: update grounding state
use_grounding_checkbox.change(
lambda v: v, inputs=use_grounding_checkbox, outputs=use_grounding_state
)
# Handle SDK selector change: update sdk state
sdk_selector.change(
lambda s: s, inputs=sdk_selector, outputs=sdk_state
)
# Link Workflow State variable change to UI status display
# Lambda receives the new state value (1 arg) because inputs=None
workflow.change(lambda new_state_value: new_state_value, inputs=None, outputs=status_text)
# Link Repo ID State variable change to UI status display
# Lambda receives the new state value (1 arg) because inputs=None
# The warning about receiving 0 args persists, likely ignorable for this lambda
repo_id.change(lambda new_repo_id_value: new_repo_id_value if new_repo_id_value else "None", inputs=None, outputs=repo_id_text)
# The main event handler for the Send button (generator)
# This .click() event triggers the ai_workflow_chat generator function
# Inputs are read from UI components AND State variables
# Outputs are updated by the values yielded from the generator
# Ensure inputs and outputs match the ai_workflow_chat signature and yield tuple EXACTLY.
# This call is direct, not in a .then() chain, so it does NOT receive a prev_output arg.
# It receives args only from the inputs list.
send_btn.click(
ai_workflow_chat, # The generator function to run (signature handles potential extra args, just in case)
inputs=[
user_input, chatbot, # UI component inputs (message, current chat history)
hf_profile, hf_token, # HF State variables
gemini_api_key_state, gemini_model_state, # Gemini State variables
repo_id, workflow, sdk_state, # Workflow State variables
iframe, run_txt, build_txt, # UI component inputs (current values)
debug_attempts, app_description, repo_name_state, generated_code_state, # Other State variables
use_grounding_state # Grounding state input
],
outputs=[
chatbot, # Updates Chatbot
repo_id, workflow, # Updates State variables (repo_id, workflow)
iframe, run_txt, build_txt, # Updates UI components (iframe, logs)
debug_attempts, app_description, repo_name_state, generated_code_state, # Updates other State variables
use_grounding_state, # Updates Grounding state
gemini_api_key_state, gemini_model_state # Updates Gemini State variables
]
).success( # Chain a .success() event to run *after* the .click() handler completes without error
# Clear the user input textbox after the message is sent and processed
lambda: gr.update(value=""),
inputs=None,
outputs=user_input # Update the user input textbox
)
# --- Initial Load Event Chain (Defined INSIDE gr.Blocks, AFTER components and required lists are defined) ---
# This chain runs once when the app loads
ai_builder_tab.load(
# Action 1: Show profile (loads cached login if available)
# show_profile expects 1 arg (profile) or None. It receives 1 from load. Correct.
show_profile,
inputs=None,
outputs=login_status # Updates UI. This output becomes the implicit first arg for the next .then()
).then(
# Action 2: Configure Gemini using initial state
# Use the then_inputs wrapper which expects (prev_output, api_key_val, model_name_val)
wrapper_configure_gemini_then,
inputs=[gemini_api_key_state, gemini_model_state], # Explicitly pass the required states
outputs=[gemini_status] # Update Gemini status display. This output becomes the implicit first arg for the next .then()
).then(
# Action 3: After initial load checks, update the button state based on initial states
# Use the then_inputs wrapper which expects (prev_output, *prereq_state_values)
wrapper_check_button_then,
inputs=send_button_interactive_binding_inputs, # Pass all 4 prerequisite states
outputs=send_button_update_output, # Update the send button. This output becomes implicit first arg for next .then()
).then(
# Action 4: Update the model description text based on the default selected model
# Use the then_inputs wrapper which expects (prev_output, model_name_val)
wrapper_get_model_description_then,
inputs=[gemini_model_state], # Get the default model name from state
outputs=[model_description_text] # Update description UI. This output becomes implicit first arg for next .then()
).then(
# Action 5: Add the initial welcome message to the chat history
# Use the prev_output wrapper which expects (prev_output,)
wrapper_greet_then,
inputs=None, # Greet takes no explicit inputs
outputs=chatbot # Updates the chatbot display
)
# The main workflow function and other helper functions are correctly defined OUTSIDE the gr.Blocks context
# because they operate on the *values* passed to them by Gradio event triggers, not the UI component objects themselves.
if __name__ == "__main__":
# Optional: Configure retries for huggingface_hub requests to make them more robust
# from requests.adapters import HTTPAdapter
# from urllib3.util.retry import Retry
# retry_strategy = Retry(total=5, backoff_factor=1, status_forcelist=[429, 500, 502, 503, 504]) # Define retry strategy for specific HTTP codes
# adapter = HTTPAdapter(max_retries=retry_strategy)
# session = get_session() # Get the session object used internally by huggingface_hub
# session.mount("http://", adapter)
# session.mount("https://", adapter)
# Optional: Configure Gradio settings using environment variables
# Set max upload size (e.g., 100MB) for files like app.py
os.environ["GRADIO_MAX_FILE_SIZE"] = "100MB"
# Optional: Set a local temporary directory for Gradio uploads
os.environ["GRADIO_TEMP_DIR"] = "./tmp"
os.makedirs(os.environ["GRADIO_TEMP_DIR"], exist_ok=True) # Ensure the directory exists
# Launch the Gradio UI
# The Gradio launch call blocks execution.
ai_builder_tab.launch()