Update app.py
Browse files
app.py
CHANGED
@@ -21,7 +21,7 @@ from transformers import (
|
|
21 |
|
22 |
NUM_EXAMPLES_FOR_FINETUNING = 50 # Constant for the number of examples to use for finetuning
|
23 |
TEXT_PIPELINE = None # Global to store the custom R1 text generation pipeline
|
24 |
-
COMPARISON_PIPELINE = None
|
25 |
|
26 |
|
27 |
def _load_model_and_tokenizer(model_name: str, subfolder: str = None, quantization_config: BitsAndBytesConfig = None, device_map: str = "auto", trust_remote_code: bool = True) -> tuple[AutoModelForCausalLM, AutoTokenizer]:
|
@@ -66,7 +66,6 @@ def finetune_small_subset() -> str:
|
|
66 |
Returns:
|
67 |
str: A message indicating finetuning completion.
|
68 |
"""
|
69 |
-
# Specify the configuration ("v0" or "v1") explicitly.
|
70 |
ds = load_dataset("ServiceNow-AI/R1-Distill-SFT", "v0", split="train")
|
71 |
ds = ds.select(range(min(NUM_EXAMPLES_FOR_FINETUNING, len(ds))))
|
72 |
|
@@ -76,8 +75,6 @@ def finetune_small_subset() -> str:
|
|
76 |
bnb_4bit_use_double_quant=True,
|
77 |
bnb_4bit_quant_type="nf4",
|
78 |
)
|
79 |
-
|
80 |
-
# Load the custom model configuration from the repository.
|
81 |
base_model, tokenizer = _load_model_and_tokenizer(
|
82 |
"wuhp/myr1", subfolder="myr1", quantization_config=bnb_config, device_map="auto"
|
83 |
)
|
@@ -112,8 +109,8 @@ def finetune_small_subset() -> str:
|
|
112 |
per_device_train_batch_size=1,
|
113 |
gradient_accumulation_steps=2,
|
114 |
logging_steps=5,
|
115 |
-
save_steps=999999,
|
116 |
-
save_total_limit=1,
|
117 |
fp16=False,
|
118 |
)
|
119 |
|
@@ -128,7 +125,7 @@ def finetune_small_subset() -> str:
|
|
128 |
trainer.model.save_pretrained("finetuned_myr1")
|
129 |
tokenizer.save_pretrained("finetuned_myr1")
|
130 |
|
131 |
-
base_model_2, tokenizer_2 = _load_model_and_tokenizer(
|
132 |
"wuhp/myr1", subfolder="myr1", quantization_config=bnb_config, device_map="auto"
|
133 |
)
|
134 |
base_model_2 = prepare_model_for_kbit_training(base_model_2)
|
@@ -139,7 +136,7 @@ def finetune_small_subset() -> str:
|
|
139 |
)
|
140 |
|
141 |
global TEXT_PIPELINE
|
142 |
-
TEXT_PIPELINE = pipeline("text-generation", model=lora_model_2, tokenizer=tokenizer_2)
|
143 |
|
144 |
return "Finetuning complete. Model loaded for inference."
|
145 |
|
@@ -205,18 +202,26 @@ def predict(
|
|
205 |
max_new_tokens (int): Maximum number of new tokens to generate.
|
206 |
|
207 |
Returns:
|
208 |
-
str: The generated text output.
|
209 |
"""
|
210 |
pipe = ensure_pipeline()
|
211 |
-
|
|
|
|
|
|
|
|
|
212 |
prompt,
|
213 |
temperature=float(temperature),
|
214 |
top_p=float(top_p),
|
215 |
min_new_tokens=int(min_new_tokens),
|
216 |
max_new_tokens=int(max_new_tokens),
|
217 |
do_sample=True
|
218 |
-
)
|
219 |
-
|
|
|
|
|
|
|
|
|
220 |
|
221 |
|
222 |
@spaces.GPU(duration=120)
|
@@ -238,28 +243,41 @@ def compare_models(
|
|
238 |
max_new_tokens (int): Maximum number of new tokens to generate.
|
239 |
|
240 |
Returns:
|
241 |
-
tuple[str, str]: A tuple containing the generated text from the custom R1 and official R1 models.
|
242 |
"""
|
243 |
local_pipe = ensure_pipeline()
|
244 |
comp_pipe = ensure_comparison_pipeline()
|
245 |
|
246 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
247 |
prompt,
|
248 |
temperature=float(temperature),
|
249 |
top_p=float(top_p),
|
250 |
min_new_tokens=int(min_new_tokens),
|
251 |
max_new_tokens=int(max_new_tokens),
|
252 |
do_sample=True
|
253 |
-
)
|
254 |
-
|
|
|
255 |
prompt,
|
256 |
temperature=float(temperature),
|
257 |
top_p=float(top_p),
|
258 |
min_new_tokens=int(min_new_tokens),
|
259 |
max_new_tokens=int(max_new_tokens),
|
260 |
do_sample=True
|
261 |
-
)
|
262 |
-
|
|
|
|
|
|
|
|
|
263 |
|
264 |
|
265 |
class ConversationRetriever:
|
@@ -335,15 +353,20 @@ def build_rag_prompt(user_query: str, retrieved_chunks: list[tuple[str, float]])
|
|
335 |
retrieved_chunks (list[tuple[str, float]]): List of retrieved text chunks and their distances.
|
336 |
|
337 |
Returns:
|
338 |
-
str: The formatted prompt string.
|
339 |
"""
|
340 |
context_str = ""
|
341 |
-
|
342 |
-
context_str +=
|
|
|
|
|
|
|
|
|
343 |
prompt = (
|
344 |
-
f"User
|
345 |
-
f"
|
346 |
-
"
|
|
|
347 |
)
|
348 |
return prompt
|
349 |
|
@@ -369,13 +392,18 @@ def chat_rag(
|
|
369 |
max_new_tokens (int): Maximum number of new tokens to generate.
|
370 |
|
371 |
Returns:
|
372 |
-
tuple[list[list[str]], list[list[str]]]: Updated chat history and chatbot display history.
|
373 |
"""
|
374 |
pipe = ensure_pipeline()
|
375 |
retriever.add_text(f"User: {user_input}")
|
376 |
top_k = 3
|
377 |
results = retriever.search(user_input, top_k=top_k)
|
378 |
prompt = build_rag_prompt(user_input, results)
|
|
|
|
|
|
|
|
|
|
|
379 |
output = pipe(
|
380 |
prompt,
|
381 |
temperature=float(temperature),
|
@@ -385,10 +413,14 @@ def chat_rag(
|
|
385 |
do_sample=True
|
386 |
)[0]["generated_text"]
|
387 |
|
388 |
-
|
389 |
-
|
|
|
|
|
|
|
|
|
390 |
else:
|
391 |
-
assistant_reply =
|
392 |
|
393 |
retriever.add_text(f"Assistant: {assistant_reply}")
|
394 |
history.append([user_input, assistant_reply])
|
@@ -398,46 +430,56 @@ def chat_rag(
|
|
398 |
# Build the Gradio interface.
|
399 |
with gr.Blocks() as demo:
|
400 |
gr.Markdown("# QLoRA Fine-tuning & RAG-based Chat Demo using Custom R1 Model")
|
|
|
401 |
|
402 |
-
|
403 |
-
|
|
|
|
|
404 |
finetune_btn.click(fn=finetune_small_subset, outputs=status_box)
|
|
|
405 |
|
406 |
-
gr.Markdown("## Direct Generation (No Retrieval)
|
407 |
-
|
408 |
-
|
409 |
-
|
|
|
410 |
min_tokens = gr.Slider(1, 2500, value=50, step=10, label="Min New Tokens")
|
411 |
max_tokens = gr.Slider(1, 2500, value=200, step=50, label="Max New Tokens")
|
412 |
-
output_box = gr.Textbox(label="Custom R1 Output", lines=8)
|
413 |
-
gen_btn = gr.Button("Generate
|
414 |
gen_btn.click(
|
415 |
fn=predict,
|
416 |
inputs=[prompt_in, temperature, top_p, min_tokens, max_tokens],
|
417 |
outputs=output_box
|
418 |
)
|
419 |
-
|
420 |
-
|
421 |
-
|
422 |
-
|
423 |
-
|
|
|
|
|
|
|
424 |
compare_btn.click(
|
425 |
fn=compare_models,
|
426 |
-
inputs=[
|
427 |
outputs=[out_custom, out_official]
|
428 |
)
|
|
|
429 |
|
430 |
-
gr.Markdown("## Chat with Retrieval-Augmented Memory")
|
|
|
431 |
with gr.Row():
|
432 |
with gr.Column():
|
433 |
-
chatbot = gr.Chatbot(label="RAG
|
434 |
chat_state = gr.State([])
|
435 |
user_input = gr.Textbox(
|
436 |
show_label=False,
|
437 |
-
placeholder="Ask a question...",
|
438 |
lines=2
|
439 |
)
|
440 |
-
send_btn = gr.Button("Send")
|
441 |
user_input.submit(
|
442 |
fn=chat_rag,
|
443 |
inputs=[user_input, chat_state, temperature, top_p, min_tokens, max_tokens],
|
@@ -448,5 +490,7 @@ with gr.Blocks() as demo:
|
|
448 |
inputs=[user_input, chat_state, temperature, top_p, min_tokens, max_tokens],
|
449 |
outputs=[chat_state, chatbot]
|
450 |
)
|
|
|
|
|
451 |
|
452 |
demo.launch()
|
|
|
21 |
|
22 |
NUM_EXAMPLES_FOR_FINETUNING = 50 # Constant for the number of examples to use for finetuning
|
23 |
TEXT_PIPELINE = None # Global to store the custom R1 text generation pipeline
|
24 |
+
COMPARISON_PIPELINE = None # Global to store the official R1 text generation pipeline
|
25 |
|
26 |
|
27 |
def _load_model_and_tokenizer(model_name: str, subfolder: str = None, quantization_config: BitsAndBytesConfig = None, device_map: str = "auto", trust_remote_code: bool = True) -> tuple[AutoModelForCausalLM, AutoTokenizer]:
|
|
|
66 |
Returns:
|
67 |
str: A message indicating finetuning completion.
|
68 |
"""
|
|
|
69 |
ds = load_dataset("ServiceNow-AI/R1-Distill-SFT", "v0", split="train")
|
70 |
ds = ds.select(range(min(NUM_EXAMPLES_FOR_FINETUNING, len(ds))))
|
71 |
|
|
|
75 |
bnb_4bit_use_double_quant=True,
|
76 |
bnb_4bit_quant_type="nf4",
|
77 |
)
|
|
|
|
|
78 |
base_model, tokenizer = _load_model_and_tokenizer(
|
79 |
"wuhp/myr1", subfolder="myr1", quantization_config=bnb_config, device_map="auto"
|
80 |
)
|
|
|
109 |
per_device_train_batch_size=1,
|
110 |
gradient_accumulation_steps=2,
|
111 |
logging_steps=5,
|
112 |
+
save_steps=999999,
|
113 |
+
save_total_limit=1,
|
114 |
fp16=False,
|
115 |
)
|
116 |
|
|
|
125 |
trainer.model.save_pretrained("finetuned_myr1")
|
126 |
tokenizer.save_pretrained("finetuned_myr1")
|
127 |
|
128 |
+
base_model_2, tokenizer_2 = _load_model_and_tokenizer(
|
129 |
"wuhp/myr1", subfolder="myr1", quantization_config=bnb_config, device_map="auto"
|
130 |
)
|
131 |
base_model_2 = prepare_model_for_kbit_training(base_model_2)
|
|
|
136 |
)
|
137 |
|
138 |
global TEXT_PIPELINE
|
139 |
+
TEXT_PIPELINE = pipeline("text-generation", model=lora_model_2, tokenizer=tokenizer_2)
|
140 |
|
141 |
return "Finetuning complete. Model loaded for inference."
|
142 |
|
|
|
202 |
max_new_tokens (int): Maximum number of new tokens to generate.
|
203 |
|
204 |
Returns:
|
205 |
+
str: The generated text output with "Thinking Process" and "Solution" sections.
|
206 |
"""
|
207 |
pipe = ensure_pipeline()
|
208 |
+
thinking_prefix = "**Thinking Process:**\n"
|
209 |
+
solution_prefix = "\n**Solution:**\n"
|
210 |
+
formatted_output = thinking_prefix
|
211 |
+
|
212 |
+
output = pipe(
|
213 |
prompt,
|
214 |
temperature=float(temperature),
|
215 |
top_p=float(top_p),
|
216 |
min_new_tokens=int(min_new_tokens),
|
217 |
max_new_tokens=int(max_new_tokens),
|
218 |
do_sample=True
|
219 |
+
)[0]["generated_text"]
|
220 |
+
|
221 |
+
formatted_output += output.strip() + solution_prefix
|
222 |
+
formatted_output += "Final Answer (This part is a placeholder and needs better extraction): ... "
|
223 |
+
|
224 |
+
return formatted_output
|
225 |
|
226 |
|
227 |
@spaces.GPU(duration=120)
|
|
|
243 |
max_new_tokens (int): Maximum number of new tokens to generate.
|
244 |
|
245 |
Returns:
|
246 |
+
tuple[str, str]: A tuple containing the formatted generated text from the custom R1 and official R1 models, each with "Thinking Process" and "Solution" sections.
|
247 |
"""
|
248 |
local_pipe = ensure_pipeline()
|
249 |
comp_pipe = ensure_comparison_pipeline()
|
250 |
|
251 |
+
def format_comparison_output(model_name, raw_output):
|
252 |
+
thinking_prefix = f"**{model_name} - Thinking Process:**\n"
|
253 |
+
solution_prefix = f"\n**{model_name} - Solution:**\n"
|
254 |
+
formatted_output = thinking_prefix
|
255 |
+
formatted_output += raw_output.strip() + solution_prefix
|
256 |
+
formatted_output += f"{model_name} Final Answer: ... "
|
257 |
+
return formatted_output
|
258 |
+
|
259 |
+
local_out_raw = local_pipe(
|
260 |
prompt,
|
261 |
temperature=float(temperature),
|
262 |
top_p=float(top_p),
|
263 |
min_new_tokens=int(min_new_tokens),
|
264 |
max_new_tokens=int(max_new_tokens),
|
265 |
do_sample=True
|
266 |
+
)[0]["generated_text"]
|
267 |
+
|
268 |
+
comp_out_raw = comp_pipe(
|
269 |
prompt,
|
270 |
temperature=float(temperature),
|
271 |
top_p=float(top_p),
|
272 |
min_new_tokens=int(min_new_tokens),
|
273 |
max_new_tokens=int(max_new_tokens),
|
274 |
do_sample=True
|
275 |
+
)[0]["generated_text"]
|
276 |
+
|
277 |
+
local_out_formatted = format_comparison_output("Custom R1", local_out_raw)
|
278 |
+
comp_out_formatted = format_comparison_output("Official R1", comp_out_raw)
|
279 |
+
|
280 |
+
return local_out_formatted, comp_out_formatted
|
281 |
|
282 |
|
283 |
class ConversationRetriever:
|
|
|
353 |
retrieved_chunks (list[tuple[str, float]]): List of retrieved text chunks and their distances.
|
354 |
|
355 |
Returns:
|
356 |
+
str: The formatted prompt string including instructions for step-by-step thinking and using context.
|
357 |
"""
|
358 |
context_str = ""
|
359 |
+
if retrieved_chunks:
|
360 |
+
context_str += "**Relevant Context:**\n"
|
361 |
+
for i, (chunk, dist) in enumerate(retrieved_chunks):
|
362 |
+
context_str += f"Chunk #{i+1} (similarity ~ {dist:.2f}):\n> {chunk}\n\n"
|
363 |
+
|
364 |
+
prompt_instruction = "Please provide a detailed answer, showing your thinking process step-by-step before stating the final answer. Use the provided context if relevant."
|
365 |
prompt = (
|
366 |
+
f"**User Query:**\n{user_query}\n\n"
|
367 |
+
f"{context_str}\n"
|
368 |
+
f"{prompt_instruction}\n\n"
|
369 |
+
"**Answer:**\n"
|
370 |
)
|
371 |
return prompt
|
372 |
|
|
|
392 |
max_new_tokens (int): Maximum number of new tokens to generate.
|
393 |
|
394 |
Returns:
|
395 |
+
tuple[list[list[str]], list[list[str]]]: Updated chat history and chatbot display history, with formatted assistant replies.
|
396 |
"""
|
397 |
pipe = ensure_pipeline()
|
398 |
retriever.add_text(f"User: {user_input}")
|
399 |
top_k = 3
|
400 |
results = retriever.search(user_input, top_k=top_k)
|
401 |
prompt = build_rag_prompt(user_input, results)
|
402 |
+
|
403 |
+
thinking_prefix = "**Thinking Process:**\n"
|
404 |
+
solution_prefix = "\n**Solution:**\n"
|
405 |
+
formatted_output = thinking_prefix
|
406 |
+
|
407 |
output = pipe(
|
408 |
prompt,
|
409 |
temperature=float(temperature),
|
|
|
413 |
do_sample=True
|
414 |
)[0]["generated_text"]
|
415 |
|
416 |
+
formatted_output += output.strip() + solution_prefix
|
417 |
+
formatted_output += "Final Answer (This part is a placeholder and needs better extraction): ... "
|
418 |
+
assistant_reply = formatted_output
|
419 |
+
|
420 |
+
if assistant_reply.startswith(prompt):
|
421 |
+
assistant_reply = assistant_reply[len(prompt):].strip()
|
422 |
else:
|
423 |
+
assistant_reply = assistant_reply.strip()
|
424 |
|
425 |
retriever.add_text(f"Assistant: {assistant_reply}")
|
426 |
history.append([user_input, assistant_reply])
|
|
|
430 |
# Build the Gradio interface.
|
431 |
with gr.Blocks() as demo:
|
432 |
gr.Markdown("# QLoRA Fine-tuning & RAG-based Chat Demo using Custom R1 Model")
|
433 |
+
gr.Markdown("---")
|
434 |
|
435 |
+
gr.Markdown("## ⚙️ Fine-tuning (Optional)")
|
436 |
+
gr.Markdown("This section allows you to fine-tune the custom R1 model on a small subset of the ServiceNow dataset. This step is optional but can potentially improve the model's performance on ServiceNow-related tasks. **Note:** This process may take up to 5 minutes.")
|
437 |
+
finetune_btn = gr.Button("🚀 Start Fine-tuning (QLoRA)")
|
438 |
+
status_box = gr.Textbox(label="Fine-tuning Status", interactive=False)
|
439 |
finetune_btn.click(fn=finetune_small_subset, outputs=status_box)
|
440 |
+
gr.Markdown("---")
|
441 |
|
442 |
+
gr.Markdown("## ✍️ Direct Generation (No Retrieval)")
|
443 |
+
gr.Markdown("Enter a prompt below to generate text directly using the custom R1 model. This is standard text generation without retrieval augmentation.")
|
444 |
+
prompt_in = gr.Textbox(lines=3, label="Input Prompt", placeholder="Enter your prompt here...")
|
445 |
+
temperature = gr.Slider(0.0, 1.5, step=0.1, value=0.7, label="Temperature (Creativity)")
|
446 |
+
top_p = gr.Slider(0.0, 1.0, step=0.05, value=0.9, label="Top-p (Sampling Nucleus)")
|
447 |
min_tokens = gr.Slider(1, 2500, value=50, step=10, label="Min New Tokens")
|
448 |
max_tokens = gr.Slider(1, 2500, value=200, step=50, label="Max New Tokens")
|
449 |
+
output_box = gr.Textbox(label="Custom R1 Output", lines=8, interactive=False)
|
450 |
+
gen_btn = gr.Button("✨ Generate Text")
|
451 |
gen_btn.click(
|
452 |
fn=predict,
|
453 |
inputs=[prompt_in, temperature, top_p, min_tokens, max_tokens],
|
454 |
outputs=output_box
|
455 |
)
|
456 |
+
gr.Markdown("---")
|
457 |
+
|
458 |
+
gr.Markdown("## 🆚 Compare Custom R1 vs Official R1")
|
459 |
+
gr.Markdown("Enter a prompt to compare the text generation of your fine-tuned custom R1 model with the official DeepSeek-R1-Distill-Llama-8B model.")
|
460 |
+
compare_prompt_in = gr.Textbox(lines=3, label="Comparison Prompt", placeholder="Enter prompt for comparison...")
|
461 |
+
compare_btn = gr.Button("⚖️ Compare Models")
|
462 |
+
out_custom = gr.Textbox(label="Custom R1 Output", lines=6, interactive=False)
|
463 |
+
out_official = gr.Textbox(label="Official R1 Output", lines=6, interactive=False)
|
464 |
compare_btn.click(
|
465 |
fn=compare_models,
|
466 |
+
inputs=[compare_prompt_in, temperature, top_p, min_tokens, max_tokens],
|
467 |
outputs=[out_custom, out_official]
|
468 |
)
|
469 |
+
gr.Markdown("---")
|
470 |
|
471 |
+
gr.Markdown("## 💬 Chat with Retrieval-Augmented Memory (RAG)")
|
472 |
+
gr.Markdown("Chat with the custom R1 model, enhanced with a retrieval-augmented memory. The model will retrieve relevant information based on your queries to provide more informed responses.")
|
473 |
with gr.Row():
|
474 |
with gr.Column():
|
475 |
+
chatbot = gr.Chatbot(label="RAG Chatbot")
|
476 |
chat_state = gr.State([])
|
477 |
user_input = gr.Textbox(
|
478 |
show_label=False,
|
479 |
+
placeholder="Ask a question to the RAG Chatbot...",
|
480 |
lines=2
|
481 |
)
|
482 |
+
send_btn = gr.Button("➡️ Send")
|
483 |
user_input.submit(
|
484 |
fn=chat_rag,
|
485 |
inputs=[user_input, chat_state, temperature, top_p, min_tokens, max_tokens],
|
|
|
490 |
inputs=[user_input, chat_state, temperature, top_p, min_tokens, max_tokens],
|
491 |
outputs=[chat_state, chatbot]
|
492 |
)
|
493 |
+
gr.Markdown("---")
|
494 |
+
|
495 |
|
496 |
demo.launch()
|