Update app.py
Browse files
app.py
CHANGED
@@ -1,28 +1,44 @@
|
|
1 |
# -*- coding: utf-8 -*-
|
2 |
"""
|
3 |
-
|
4 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
5 |
"""
|
6 |
|
7 |
-
import
|
|
|
|
|
8 |
import numpy as np
|
9 |
-
import
|
10 |
-
import torch
|
11 |
import librosa
|
12 |
-
|
13 |
from scipy.io.wavfile import write as write_wav
|
14 |
-
import os
|
15 |
-
import re
|
16 |
from huggingface_hub import login
|
17 |
-
import
|
18 |
|
19 |
-
|
20 |
-
|
21 |
-
|
22 |
-
|
23 |
-
|
|
|
|
|
|
|
|
|
24 |
|
25 |
-
#
|
26 |
STT_MODEL_ID = "EYEDOL/SALAMA_C3"
|
27 |
LLM_MODEL_ID = "EYEDOL/Llama-3.2-1B_ON_ALPACA5"
|
28 |
TTS_TOKENIZER_ID = "facebook/mms-tts-swh"
|
@@ -31,6 +47,18 @@ TTS_ONNX_MODEL_PATH = "swahili_tts.onnx"
|
|
31 |
TEMP_DIR = "temp"
|
32 |
os.makedirs(TEMP_DIR, exist_ok=True)
|
33 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
34 |
|
35 |
class WeeboAssistant:
|
36 |
def __init__(self):
|
@@ -48,125 +76,224 @@ class WeeboAssistant:
|
|
48 |
self.torch_dtype = torch.bfloat16 if self.device == "cuda" else torch.float32
|
49 |
print(f"Using device: {self.device}")
|
50 |
|
51 |
-
# STT
|
52 |
print(f"Loading STT model: {STT_MODEL_ID}")
|
53 |
self.stt_processor = AutoProcessor.from_pretrained(STT_MODEL_ID)
|
|
|
54 |
self.stt_model = AutoModelForSpeechSeq2Seq.from_pretrained(
|
55 |
-
STT_MODEL_ID,
|
56 |
-
torch_dtype=self.torch_dtype,
|
57 |
-
low_cpu_mem_usage=True,
|
58 |
-
use_safetensors=True
|
59 |
-
)
|
|
|
|
|
|
|
|
|
|
|
60 |
print("STT model loaded successfully.")
|
61 |
|
62 |
-
# LLM
|
63 |
print(f"Loading LLM: {LLM_MODEL_ID}")
|
64 |
-
self.llm_tokenizer = AutoTokenizer.from_pretrained(LLM_MODEL_ID)
|
65 |
-
|
66 |
-
|
67 |
-
|
68 |
-
|
69 |
-
|
70 |
-
|
71 |
-
|
72 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
73 |
|
74 |
-
# TTS
|
75 |
print(f"Loading TTS model: {TTS_ONNX_MODEL_PATH}")
|
76 |
-
|
77 |
-
|
78 |
-
|
79 |
-
|
|
|
80 |
self.tts_tokenizer = AutoTokenizer.from_pretrained(TTS_TOKENIZER_ID)
|
81 |
print("TTS model and tokenizer loaded successfully.")
|
82 |
|
83 |
print("-" * 30)
|
84 |
print("All models initialized successfully! ✅")
|
85 |
|
|
|
86 |
def transcribe_audio(self, audio_tuple):
|
|
|
87 |
if audio_tuple is None:
|
88 |
return ""
|
89 |
sample_rate, audio_data = audio_tuple
|
|
|
90 |
if audio_data.ndim > 1:
|
91 |
audio_data = audio_data.mean(axis=1)
|
|
|
92 |
if audio_data.dtype != np.float32:
|
93 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
94 |
if sample_rate != self.STT_SAMPLE_RATE:
|
95 |
audio_data = librosa.resample(y=audio_data, orig_sr=sample_rate, target_sr=self.STT_SAMPLE_RATE)
|
96 |
if len(audio_data) < 1000:
|
97 |
return "(Audio too short to transcribe)"
|
|
|
98 |
inputs = self.stt_processor(audio_data, sampling_rate=self.STT_SAMPLE_RATE, return_tensors="pt")
|
99 |
-
inputs = {k: v.to(self.device) for k, v in inputs.items()}
|
100 |
with torch.no_grad():
|
101 |
generated_ids = self.stt_model.generate(**inputs, max_new_tokens=128)
|
102 |
transcription = self.stt_processor.batch_decode(generated_ids, skip_special_tokens=True)[0]
|
103 |
return transcription.strip()
|
104 |
|
105 |
def generate_speech(self, text):
|
|
|
106 |
if not text:
|
107 |
return None
|
108 |
text = text.strip()
|
|
|
109 |
inputs = self.tts_tokenizer(text, return_tensors="np")
|
110 |
-
|
|
|
111 |
audio_waveform = self.tts_session.run(None, ort_inputs)[0].flatten()
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
112 |
output_path = os.path.join(TEMP_DIR, f"{os.urandom(8).hex()}.wav")
|
113 |
-
write_wav(output_path, self.TTS_SAMPLE_RATE,
|
114 |
return output_path
|
115 |
|
116 |
def get_llm_response(self, chat_history):
|
117 |
-
|
118 |
-
|
|
|
|
|
|
|
|
|
|
|
119 |
for user_msg, assistant_msg in chat_history:
|
120 |
-
|
|
|
|
|
121 |
if assistant_msg:
|
122 |
-
|
123 |
-
|
124 |
-
prompt =
|
125 |
-
|
126 |
-
|
127 |
-
|
128 |
-
|
129 |
-
self.
|
130 |
-
|
131 |
-
|
132 |
-
|
133 |
-
|
134 |
-
)
|
135 |
-
|
136 |
generation_kwargs = dict(
|
137 |
-
|
|
|
138 |
max_new_tokens=512,
|
139 |
-
eos_token_id=terminators,
|
140 |
do_sample=True,
|
141 |
temperature=0.6,
|
142 |
top_p=0.9,
|
|
|
|
|
143 |
)
|
144 |
-
|
145 |
-
thread
|
146 |
-
|
147 |
-
|
|
|
148 |
return streamer
|
149 |
|
|
|
|
|
150 |
assistant = WeeboAssistant()
|
151 |
|
152 |
|
|
|
|
|
153 |
def s2s_pipeline(audio_input, chat_history):
|
|
|
154 |
user_text = assistant.transcribe_audio(audio_input)
|
155 |
if not user_text or user_text.startswith("("):
|
156 |
chat_history.append((user_text or "(No valid speech detected)", None))
|
157 |
yield chat_history, None, "Please record your voice again."
|
158 |
return
|
159 |
-
|
160 |
chat_history.append((user_text, ""))
|
161 |
yield chat_history, None, "..."
|
162 |
-
|
163 |
response_stream = assistant.get_llm_response(chat_history)
|
164 |
llm_response_text = ""
|
165 |
for text_chunk in response_stream:
|
166 |
llm_response_text += text_chunk
|
|
|
167 |
chat_history[-1] = (user_text, llm_response_text)
|
168 |
yield chat_history, None, llm_response_text
|
169 |
-
|
|
|
170 |
final_audio_path = assistant.generate_speech(llm_response_text)
|
171 |
yield chat_history, final_audio_path, llm_response_text
|
172 |
|
@@ -174,7 +301,7 @@ def s2s_pipeline(audio_input, chat_history):
|
|
174 |
def t2t_pipeline(text_input, chat_history):
|
175 |
chat_history.append((text_input, ""))
|
176 |
yield chat_history
|
177 |
-
|
178 |
response_stream = assistant.get_llm_response(chat_history)
|
179 |
llm_response_text = ""
|
180 |
for text_chunk in response_stream:
|
@@ -187,10 +314,11 @@ def clear_textbox():
|
|
187 |
return gr.Textbox(value="")
|
188 |
|
189 |
|
|
|
190 |
with gr.Blocks(theme=gr.themes.Soft(), title="Msaidizi wa Kiswahili") as demo:
|
191 |
gr.Markdown("# 🤖 Msaidizi wa Sauti wa Kiswahili (Swahili Voice Assistant)")
|
192 |
gr.Markdown("Ongea na msaidizi kwa Kiswahili. Toa sauti, andika maandishi, na upate majibu kwa sauti au maandishi.")
|
193 |
-
|
194 |
with gr.Tabs():
|
195 |
with gr.TabItem("🎙️ Sauti-kwa-Sauti (Speech-to-Speech)"):
|
196 |
with gr.Row():
|
@@ -201,7 +329,7 @@ with gr.Blocks(theme=gr.themes.Soft(), title="Msaidizi wa Kiswahili") as demo:
|
|
201 |
s2s_chatbot = gr.Chatbot(label="Mazungumzo (Conversation)", bubble_full_width=False, height=400)
|
202 |
s2s_audio_out = gr.Audio(type="filepath", label="Jibu la Sauti (Audio Response)", autoplay=True)
|
203 |
s2s_text_out = gr.Textbox(label="Jibu la Maandishi (Text Response)", interactive=False)
|
204 |
-
|
205 |
with gr.TabItem("⌨️ Maandishi-kwa-Maandishi (Text-to-Text)"):
|
206 |
t2t_chatbot = gr.Chatbot(label="Mazungumzo (Conversation)", bubble_full_width=False, height=500)
|
207 |
with gr.Row():
|
@@ -225,46 +353,48 @@ with gr.Blocks(theme=gr.themes.Soft(), title="Msaidizi wa Kiswahili") as demo:
|
|
225 |
fn=s2s_pipeline,
|
226 |
inputs=[s2s_audio_in, s2s_chatbot],
|
227 |
outputs=[s2s_chatbot, s2s_audio_out, s2s_text_out],
|
228 |
-
queue=True
|
229 |
).then(
|
230 |
fn=lambda: gr.Audio(value=None),
|
231 |
inputs=None,
|
232 |
-
outputs=s2s_audio_in
|
233 |
)
|
234 |
|
235 |
t2t_submit_btn.click(
|
236 |
fn=t2t_pipeline,
|
237 |
inputs=[t2t_text_in, t2t_chatbot],
|
238 |
outputs=[t2t_chatbot],
|
239 |
-
queue=True
|
240 |
).then(
|
241 |
fn=clear_textbox,
|
242 |
inputs=None,
|
243 |
-
outputs=t2t_text_in
|
244 |
)
|
245 |
-
|
246 |
t2t_text_in.submit(
|
247 |
fn=t2t_pipeline,
|
248 |
inputs=[t2t_text_in, t2t_chatbot],
|
249 |
outputs=[t2t_chatbot],
|
250 |
-
queue=True
|
251 |
).then(
|
252 |
fn=clear_textbox,
|
253 |
inputs=None,
|
254 |
-
outputs=t2t_text_in
|
255 |
)
|
256 |
|
257 |
tool_s2t_btn.click(
|
258 |
fn=assistant.transcribe_audio,
|
259 |
inputs=tool_s2t_audio_in,
|
260 |
outputs=tool_s2t_text_out,
|
261 |
-
queue=True
|
262 |
)
|
|
|
263 |
tool_t2s_btn.click(
|
264 |
fn=assistant.generate_speech,
|
265 |
inputs=tool_t2s_text_in,
|
266 |
outputs=tool_t2s_audio_out,
|
267 |
-
queue=True
|
268 |
)
|
269 |
|
270 |
-
|
|
|
|
1 |
# -*- coding: utf-8 -*-
|
2 |
"""
|
3 |
+
Fixed and self-contained Swahili multimodal assistant for Hugging Face Spaces.
|
4 |
+
|
5 |
+
Key fixes / improvements over original:
|
6 |
+
- Robust loading of an LLM repo that may lack `model_type` in config.json by
|
7 |
+
loading the model object directly and using `trust_remote_code=True` as a
|
8 |
+
fallback. Avoids `pipeline(... )` raising ValueError on AutoConfig.
|
9 |
+
- Correct handling of `pipeline(..., device=...)` which expects an int GPU
|
10 |
+
index or -1 for CPU (previously passed a string like "cpu").
|
11 |
+
- Streaming generation implemented by calling `model.generate(..., streamer=TextIteratorStreamer(...))`
|
12 |
+
in a background thread so the main thread can iterate over the streamer.
|
13 |
+
- Use standard HF env var `HF_TOKEN` and graceful error message if not set.
|
14 |
+
- Minor robustness improvements (resampling audio, handling mono/stereo, temp
|
15 |
+
filenames, etc.).
|
16 |
+
|
17 |
+
Drop this file into your Space and replace the old app.py contents.
|
18 |
"""
|
19 |
|
20 |
+
import os
|
21 |
+
import re
|
22 |
+
import threading
|
23 |
import numpy as np
|
24 |
+
import gradio as gr
|
|
|
25 |
import librosa
|
26 |
+
import torch
|
27 |
from scipy.io.wavfile import write as write_wav
|
|
|
|
|
28 |
from huggingface_hub import login
|
29 |
+
import onnxruntime
|
30 |
|
31 |
+
from transformers import (
|
32 |
+
AutoProcessor,
|
33 |
+
AutoModelForSpeechSeq2Seq,
|
34 |
+
AutoTokenizer,
|
35 |
+
AutoConfig,
|
36 |
+
AutoModelForCausalLM,
|
37 |
+
pipeline,
|
38 |
+
TextIteratorStreamer,
|
39 |
+
)
|
40 |
|
41 |
+
# -------------------- Configuration --------------------
|
42 |
STT_MODEL_ID = "EYEDOL/SALAMA_C3"
|
43 |
LLM_MODEL_ID = "EYEDOL/Llama-3.2-1B_ON_ALPACA5"
|
44 |
TTS_TOKENIZER_ID = "facebook/mms-tts-swh"
|
|
|
47 |
TEMP_DIR = "temp"
|
48 |
os.makedirs(TEMP_DIR, exist_ok=True)
|
49 |
|
50 |
+
# Use the standard environment variable name used by Spaces
|
51 |
+
HF_TOKEN = os.environ.get("HF_TOKEN") or os.environ.get("hugface")
|
52 |
+
if not HF_TOKEN:
|
53 |
+
raise ValueError("HF_TOKEN not found. Please set it in Hugging Face Space repository secrets.")
|
54 |
+
|
55 |
+
# Attempt login to HF hub (Spaces typically already provides token, but this keeps parity)
|
56 |
+
try:
|
57 |
+
login(token=HF_TOKEN)
|
58 |
+
print("Successfully logged into Hugging Face Hub!")
|
59 |
+
except Exception as e:
|
60 |
+
print("Warning: could not call huggingface_hub.login(). Proceeding — ensure your token is valid in the environment. Error:", e)
|
61 |
+
|
62 |
|
63 |
class WeeboAssistant:
|
64 |
def __init__(self):
|
|
|
76 |
self.torch_dtype = torch.bfloat16 if self.device == "cuda" else torch.float32
|
77 |
print(f"Using device: {self.device}")
|
78 |
|
79 |
+
# ---------------- STT ----------------
|
80 |
print(f"Loading STT model: {STT_MODEL_ID}")
|
81 |
self.stt_processor = AutoProcessor.from_pretrained(STT_MODEL_ID)
|
82 |
+
# Speech seq2seq model (e.g. Whisper-like)
|
83 |
self.stt_model = AutoModelForSpeechSeq2Seq.from_pretrained(
|
84 |
+
STT_MODEL_ID,
|
85 |
+
torch_dtype=self.torch_dtype,
|
86 |
+
low_cpu_mem_usage=True,
|
87 |
+
use_safetensors=True,
|
88 |
+
)
|
89 |
+
if self.device == "cuda":
|
90 |
+
try:
|
91 |
+
self.stt_model = self.stt_model.to("cuda")
|
92 |
+
except Exception:
|
93 |
+
pass
|
94 |
print("STT model loaded successfully.")
|
95 |
|
96 |
+
# ---------------- LLM ----------------
|
97 |
print(f"Loading LLM: {LLM_MODEL_ID}")
|
98 |
+
self.llm_tokenizer = AutoTokenizer.from_pretrained(LLM_MODEL_ID, use_fast=True)
|
99 |
+
|
100 |
+
# Attempt robust loading. If the repo lacks a model_type in config.json,
|
101 |
+
# try loading with trust_remote_code=True (this allows custom model code in repo).
|
102 |
+
try:
|
103 |
+
config = AutoConfig.from_pretrained(LLM_MODEL_ID)
|
104 |
+
# If config loaded but missing model_type, continue to try direct load
|
105 |
+
if not getattr(config, "model_type", None):
|
106 |
+
raise ValueError("config missing model_type - forcing trusted load")
|
107 |
+
|
108 |
+
# Try to load into a causal LM class (works for many standard model types)
|
109 |
+
self.llm_model = AutoModelForCausalLM.from_pretrained(
|
110 |
+
LLM_MODEL_ID,
|
111 |
+
config=config,
|
112 |
+
torch_dtype=self.torch_dtype,
|
113 |
+
low_cpu_mem_usage=True,
|
114 |
+
)
|
115 |
+
except Exception as first_err:
|
116 |
+
print("Standard AutoConfig/AutoModel load failed or model_type missing. Trying trust_remote_code=True. Error:", first_err)
|
117 |
+
# Try using trust_remote_code which will import repo-specific model code if present
|
118 |
+
try:
|
119 |
+
config = AutoConfig.from_pretrained(LLM_MODEL_ID, trust_remote_code=True)
|
120 |
+
self.llm_model = AutoModelForCausalLM.from_pretrained(
|
121 |
+
LLM_MODEL_ID,
|
122 |
+
config=config,
|
123 |
+
torch_dtype=self.torch_dtype,
|
124 |
+
trust_remote_code=True,
|
125 |
+
low_cpu_mem_usage=True,
|
126 |
+
device_map="auto" if torch.cuda.is_available() else None,
|
127 |
+
)
|
128 |
+
except Exception as second_err:
|
129 |
+
# Final fallback: try to load without special configs — may still fail for custom repos
|
130 |
+
print("Fallback load also failed:", second_err)
|
131 |
+
raise RuntimeError(
|
132 |
+
"Unable to load LLM model. Check the model repo, ensure config.json contains a model_type or that trust_remote_code is allowed."
|
133 |
+
)
|
134 |
+
|
135 |
+
# If device_map wasn't used and model is on CPU, ensure model is moved to CPU
|
136 |
+
if self.device == "cpu":
|
137 |
+
try:
|
138 |
+
# Many Hugging Face helpers use device_map; if not used, move model
|
139 |
+
self.llm_model = self.llm_model.to("cpu")
|
140 |
+
except Exception:
|
141 |
+
pass
|
142 |
+
|
143 |
+
# For convenience, create a pipeline for non-streaming quick calls (device expects int or -1)
|
144 |
+
device_index = 0 if torch.cuda.is_available() else -1
|
145 |
+
try:
|
146 |
+
self.llm_pipeline = pipeline(
|
147 |
+
"text-generation",
|
148 |
+
model=self.llm_model,
|
149 |
+
tokenizer=self.llm_tokenizer,
|
150 |
+
device=device_index,
|
151 |
+
model_kwargs={"torch_dtype": self.torch_dtype},
|
152 |
+
)
|
153 |
+
except Exception:
|
154 |
+
# pipeline is optional; if it fails we still support the streaming flow via model.generate
|
155 |
+
self.llm_pipeline = None
|
156 |
+
|
157 |
+
print("LLM loaded successfully.")
|
158 |
|
159 |
+
# ---------------- TTS ----------------
|
160 |
print(f"Loading TTS model: {TTS_ONNX_MODEL_PATH}")
|
161 |
+
# ONNX runtime session; providers include CUDA if available
|
162 |
+
providers = ["CPUExecutionProvider"]
|
163 |
+
if torch.cuda.is_available():
|
164 |
+
providers = ["CUDAExecutionProvider", "CPUExecutionProvider"]
|
165 |
+
self.tts_session = onnxruntime.InferenceSession(TTS_ONNX_MODEL_PATH, providers=providers)
|
166 |
self.tts_tokenizer = AutoTokenizer.from_pretrained(TTS_TOKENIZER_ID)
|
167 |
print("TTS model and tokenizer loaded successfully.")
|
168 |
|
169 |
print("-" * 30)
|
170 |
print("All models initialized successfully! ✅")
|
171 |
|
172 |
+
# ---------------- Utility methods ----------------
|
173 |
def transcribe_audio(self, audio_tuple):
|
174 |
+
"""Take a Gradio audio tuple (sample_rate, np_audio) and return transcription string."""
|
175 |
if audio_tuple is None:
|
176 |
return ""
|
177 |
sample_rate, audio_data = audio_tuple
|
178 |
+
# Convert to mono
|
179 |
if audio_data.ndim > 1:
|
180 |
audio_data = audio_data.mean(axis=1)
|
181 |
+
# Normalize to float32
|
182 |
if audio_data.dtype != np.float32:
|
183 |
+
# handle common integer audio dtypes
|
184 |
+
if np.issubdtype(audio_data.dtype, np.integer):
|
185 |
+
max_val = np.iinfo(audio_data.dtype).max
|
186 |
+
audio_data = audio_data.astype(np.float32) / float(max_val)
|
187 |
+
else:
|
188 |
+
audio_data = audio_data.astype(np.float32)
|
189 |
+
# Resample if needed
|
190 |
if sample_rate != self.STT_SAMPLE_RATE:
|
191 |
audio_data = librosa.resample(y=audio_data, orig_sr=sample_rate, target_sr=self.STT_SAMPLE_RATE)
|
192 |
if len(audio_data) < 1000:
|
193 |
return "(Audio too short to transcribe)"
|
194 |
+
|
195 |
inputs = self.stt_processor(audio_data, sampling_rate=self.STT_SAMPLE_RATE, return_tensors="pt")
|
196 |
+
inputs = {k: v.to(next(self.stt_model.parameters()).device) for k, v in inputs.items()}
|
197 |
with torch.no_grad():
|
198 |
generated_ids = self.stt_model.generate(**inputs, max_new_tokens=128)
|
199 |
transcription = self.stt_processor.batch_decode(generated_ids, skip_special_tokens=True)[0]
|
200 |
return transcription.strip()
|
201 |
|
202 |
def generate_speech(self, text):
|
203 |
+
"""Synthesize speech using the ONNX TTS model and return a filepath to a WAV file."""
|
204 |
if not text:
|
205 |
return None
|
206 |
text = text.strip()
|
207 |
+
# Tokenize with numpy arrays for ONNX
|
208 |
inputs = self.tts_tokenizer(text, return_tensors="np")
|
209 |
+
input_name = self.tts_session.get_inputs()[0].name
|
210 |
+
ort_inputs = {input_name: inputs["input_ids"]}
|
211 |
audio_waveform = self.tts_session.run(None, ort_inputs)[0].flatten()
|
212 |
+
|
213 |
+
# ONNX model might produce float audio in range [-1,1] or int16 depending on model. We'll safe-guard.
|
214 |
+
# Normalize to int16 WAV
|
215 |
+
if np.issubdtype(audio_waveform.dtype, np.floating):
|
216 |
+
# Clip and convert
|
217 |
+
audio_clip = np.clip(audio_waveform, -1.0, 1.0)
|
218 |
+
audio_int16 = (audio_clip * 32767).astype(np.int16)
|
219 |
+
else:
|
220 |
+
audio_int16 = audio_waveform.astype(np.int16)
|
221 |
+
|
222 |
output_path = os.path.join(TEMP_DIR, f"{os.urandom(8).hex()}.wav")
|
223 |
+
write_wav(output_path, self.TTS_SAMPLE_RATE, audio_int16)
|
224 |
return output_path
|
225 |
|
226 |
def get_llm_response(self, chat_history):
|
227 |
+
"""Return a TextIteratorStreamer that yields generated text pieces as the model produces them.
|
228 |
+
|
229 |
+
This implementation uses self.llm_model.generate(...) with a TextIteratorStreamer and
|
230 |
+
runs generate in a background thread so the caller can iterate over streamer.
|
231 |
+
"""
|
232 |
+
# Build prompt from system + conversation. Adjust this template to match your LLM's preferred format.
|
233 |
+
prompt_lines = [self.SYSTEM_PROMPT.strip(), "\n"]
|
234 |
for user_msg, assistant_msg in chat_history:
|
235 |
+
if user_msg:
|
236 |
+
# tag user messages clearly so model understands dialogue turns
|
237 |
+
prompt_lines.append("User: " + user_msg)
|
238 |
if assistant_msg:
|
239 |
+
prompt_lines.append("Assistant: " + assistant_msg)
|
240 |
+
prompt_lines.append("Assistant: ")
|
241 |
+
prompt = "\n".join(prompt_lines)
|
242 |
+
|
243 |
+
# Tokenize and prepare inputs on the same device as the model
|
244 |
+
inputs = self.llm_tokenizer(prompt, return_tensors="pt")
|
245 |
+
try:
|
246 |
+
model_device = next(self.llm_model.parameters()).device
|
247 |
+
except StopIteration:
|
248 |
+
model_device = torch.device("cpu")
|
249 |
+
inputs = {k: v.to(model_device) for k, v in inputs.items()}
|
250 |
+
|
251 |
+
streamer = TextIteratorStreamer(self.llm_tokenizer, skip_prompt=True, skip_special_tokens=True)
|
252 |
+
|
253 |
generation_kwargs = dict(
|
254 |
+
input_ids=inputs["input_ids"],
|
255 |
+
attention_mask=inputs.get("attention_mask", None),
|
256 |
max_new_tokens=512,
|
|
|
257 |
do_sample=True,
|
258 |
temperature=0.6,
|
259 |
top_p=0.9,
|
260 |
+
streamer=streamer,
|
261 |
+
eos_token_id=getattr(self.llm_tokenizer, "eos_token_id", None),
|
262 |
)
|
263 |
+
|
264 |
+
# Launch generation in a thread so we can yield from the streamer in the main thread
|
265 |
+
gen_thread = threading.Thread(target=self.llm_model.generate, kwargs=generation_kwargs, daemon=True)
|
266 |
+
gen_thread.start()
|
267 |
+
|
268 |
return streamer
|
269 |
|
270 |
+
|
271 |
+
# -------------------- Create assistant instance --------------------
|
272 |
assistant = WeeboAssistant()
|
273 |
|
274 |
|
275 |
+
# -------------------- Gradio pipelines --------------------
|
276 |
+
|
277 |
def s2s_pipeline(audio_input, chat_history):
|
278 |
+
# `chat_history` is expected to be a list of (user_text, assistant_text) tuples
|
279 |
user_text = assistant.transcribe_audio(audio_input)
|
280 |
if not user_text or user_text.startswith("("):
|
281 |
chat_history.append((user_text or "(No valid speech detected)", None))
|
282 |
yield chat_history, None, "Please record your voice again."
|
283 |
return
|
284 |
+
|
285 |
chat_history.append((user_text, ""))
|
286 |
yield chat_history, None, "..."
|
287 |
+
|
288 |
response_stream = assistant.get_llm_response(chat_history)
|
289 |
llm_response_text = ""
|
290 |
for text_chunk in response_stream:
|
291 |
llm_response_text += text_chunk
|
292 |
+
# Update last turn in chat history
|
293 |
chat_history[-1] = (user_text, llm_response_text)
|
294 |
yield chat_history, None, llm_response_text
|
295 |
+
|
296 |
+
# Once finished, synthesize audio
|
297 |
final_audio_path = assistant.generate_speech(llm_response_text)
|
298 |
yield chat_history, final_audio_path, llm_response_text
|
299 |
|
|
|
301 |
def t2t_pipeline(text_input, chat_history):
|
302 |
chat_history.append((text_input, ""))
|
303 |
yield chat_history
|
304 |
+
|
305 |
response_stream = assistant.get_llm_response(chat_history)
|
306 |
llm_response_text = ""
|
307 |
for text_chunk in response_stream:
|
|
|
314 |
return gr.Textbox(value="")
|
315 |
|
316 |
|
317 |
+
# -------------------- Gradio UI --------------------
|
318 |
with gr.Blocks(theme=gr.themes.Soft(), title="Msaidizi wa Kiswahili") as demo:
|
319 |
gr.Markdown("# 🤖 Msaidizi wa Sauti wa Kiswahili (Swahili Voice Assistant)")
|
320 |
gr.Markdown("Ongea na msaidizi kwa Kiswahili. Toa sauti, andika maandishi, na upate majibu kwa sauti au maandishi.")
|
321 |
+
|
322 |
with gr.Tabs():
|
323 |
with gr.TabItem("🎙️ Sauti-kwa-Sauti (Speech-to-Speech)"):
|
324 |
with gr.Row():
|
|
|
329 |
s2s_chatbot = gr.Chatbot(label="Mazungumzo (Conversation)", bubble_full_width=False, height=400)
|
330 |
s2s_audio_out = gr.Audio(type="filepath", label="Jibu la Sauti (Audio Response)", autoplay=True)
|
331 |
s2s_text_out = gr.Textbox(label="Jibu la Maandishi (Text Response)", interactive=False)
|
332 |
+
|
333 |
with gr.TabItem("⌨️ Maandishi-kwa-Maandishi (Text-to-Text)"):
|
334 |
t2t_chatbot = gr.Chatbot(label="Mazungumzo (Conversation)", bubble_full_width=False, height=500)
|
335 |
with gr.Row():
|
|
|
353 |
fn=s2s_pipeline,
|
354 |
inputs=[s2s_audio_in, s2s_chatbot],
|
355 |
outputs=[s2s_chatbot, s2s_audio_out, s2s_text_out],
|
356 |
+
queue=True,
|
357 |
).then(
|
358 |
fn=lambda: gr.Audio(value=None),
|
359 |
inputs=None,
|
360 |
+
outputs=s2s_audio_in,
|
361 |
)
|
362 |
|
363 |
t2t_submit_btn.click(
|
364 |
fn=t2t_pipeline,
|
365 |
inputs=[t2t_text_in, t2t_chatbot],
|
366 |
outputs=[t2t_chatbot],
|
367 |
+
queue=True,
|
368 |
).then(
|
369 |
fn=clear_textbox,
|
370 |
inputs=None,
|
371 |
+
outputs=t2t_text_in,
|
372 |
)
|
373 |
+
|
374 |
t2t_text_in.submit(
|
375 |
fn=t2t_pipeline,
|
376 |
inputs=[t2t_text_in, t2t_chatbot],
|
377 |
outputs=[t2t_chatbot],
|
378 |
+
queue=True,
|
379 |
).then(
|
380 |
fn=clear_textbox,
|
381 |
inputs=None,
|
382 |
+
outputs=t2t_text_in,
|
383 |
)
|
384 |
|
385 |
tool_s2t_btn.click(
|
386 |
fn=assistant.transcribe_audio,
|
387 |
inputs=tool_s2t_audio_in,
|
388 |
outputs=tool_s2t_text_out,
|
389 |
+
queue=True,
|
390 |
)
|
391 |
+
|
392 |
tool_t2s_btn.click(
|
393 |
fn=assistant.generate_speech,
|
394 |
inputs=tool_t2s_text_in,
|
395 |
outputs=tool_t2s_audio_out,
|
396 |
+
queue=True,
|
397 |
)
|
398 |
|
399 |
+
|
400 |
+
demo.queue().launch(debug=True)
|