Update app.py
Browse files
app.py
CHANGED
@@ -20,6 +20,11 @@ from peft import LoraConfig, TaskType, get_peft_model, prepare_model_for_kbit_tr
|
|
20 |
|
21 |
from sentence_transformers import SentenceTransformer
|
22 |
|
|
|
|
|
|
|
|
|
|
|
23 |
TEXT_PIPELINE = None
|
24 |
COMPARISON_PIPELINE = None
|
25 |
NUM_EXAMPLES = 1000
|
@@ -27,47 +32,46 @@ NUM_EXAMPLES = 1000
|
|
27 |
@spaces.GPU(duration=300)
|
28 |
def finetune_small_subset():
|
29 |
"""
|
30 |
-
1) Loads
|
31 |
2) Adds LoRA adapters (trainable),
|
32 |
-
3)
|
33 |
-
4) Saves LoRA adapter to
|
34 |
-
5) Reloads LoRA
|
35 |
"""
|
36 |
-
|
37 |
-
ds = load_dataset(
|
38 |
-
"Magpie-Align/Magpie-Reasoning-V2-250K-CoT-Deepseek-R1-Llama-70B",
|
39 |
-
split="train"
|
40 |
-
)
|
41 |
-
|
42 |
-
unique_ids = list(set(ds["conversation_id"]))
|
43 |
-
single_id = unique_ids[0]
|
44 |
-
ds = ds.filter(lambda x: x["conversation_id"] == single_id)
|
45 |
-
|
46 |
ds = ds.select(range(min(NUM_EXAMPLES, len(ds))))
|
47 |
|
48 |
bnb_config = BitsAndBytesConfig(
|
49 |
load_in_4bit=True,
|
50 |
-
bnb_4bit_compute_dtype=torch.bfloat16,
|
51 |
bnb_4bit_use_double_quant=True,
|
52 |
bnb_4bit_quant_type="nf4",
|
53 |
)
|
54 |
|
55 |
-
|
56 |
-
|
|
|
57 |
subfolder="myr1",
|
58 |
-
trust_remote_code=True
|
59 |
)
|
|
|
|
|
|
|
|
|
60 |
tokenizer = AutoTokenizer.from_pretrained(
|
61 |
-
"wuhp/myr1",
|
62 |
subfolder="myr1",
|
63 |
trust_remote_code=True
|
64 |
)
|
65 |
|
|
|
|
|
66 |
base_model = AutoModelForCausalLM.from_pretrained(
|
67 |
"wuhp/myr1",
|
68 |
subfolder="myr1",
|
69 |
-
config=
|
70 |
-
quantization_config=bnb_config,
|
71 |
device_map="auto",
|
72 |
trust_remote_code=True
|
73 |
)
|
@@ -84,10 +88,11 @@ def finetune_small_subset():
|
|
84 |
)
|
85 |
lora_model = get_peft_model(base_model, lora_config)
|
86 |
|
|
|
87 |
def tokenize_fn(ex):
|
88 |
text = (
|
89 |
-
f"
|
90 |
-
f"
|
91 |
)
|
92 |
return tokenizer(text, truncation=True, max_length=512)
|
93 |
|
@@ -102,9 +107,9 @@ def finetune_small_subset():
|
|
102 |
per_device_train_batch_size=1,
|
103 |
gradient_accumulation_steps=2,
|
104 |
logging_steps=5,
|
105 |
-
save_steps=999999,
|
106 |
save_total_limit=1,
|
107 |
-
fp16=False,
|
108 |
)
|
109 |
|
110 |
trainer = Trainer(
|
@@ -115,13 +120,15 @@ def finetune_small_subset():
|
|
115 |
)
|
116 |
trainer.train()
|
117 |
|
|
|
118 |
trainer.model.save_pretrained("finetuned_myr1")
|
119 |
tokenizer.save_pretrained("finetuned_myr1")
|
120 |
|
|
|
121 |
base_model_2 = AutoModelForCausalLM.from_pretrained(
|
122 |
"wuhp/myr1",
|
123 |
subfolder="myr1",
|
124 |
-
config=
|
125 |
quantization_config=bnb_config,
|
126 |
device_map="auto",
|
127 |
trust_remote_code=True
|
@@ -140,8 +147,8 @@ def finetune_small_subset():
|
|
140 |
|
141 |
def ensure_pipeline():
|
142 |
"""
|
143 |
-
If we haven't
|
144 |
-
load the base model in 4-bit
|
145 |
"""
|
146 |
global TEXT_PIPELINE
|
147 |
if TEXT_PIPELINE is None:
|
@@ -151,12 +158,14 @@ def ensure_pipeline():
|
|
151 |
bnb_4bit_use_double_quant=True,
|
152 |
bnb_4bit_quant_type="nf4",
|
153 |
)
|
154 |
-
|
|
|
|
|
155 |
tokenizer = AutoTokenizer.from_pretrained("wuhp/myr1", subfolder="myr1", trust_remote_code=True)
|
156 |
base_model = AutoModelForCausalLM.from_pretrained(
|
157 |
"wuhp/myr1",
|
158 |
subfolder="myr1",
|
159 |
-
config=
|
160 |
quantization_config=bnb_config,
|
161 |
device_map="auto",
|
162 |
trust_remote_code=True
|
@@ -166,7 +175,7 @@ def ensure_pipeline():
|
|
166 |
|
167 |
def ensure_comparison_pipeline():
|
168 |
"""
|
169 |
-
Load
|
170 |
"""
|
171 |
global COMPARISON_PIPELINE
|
172 |
if COMPARISON_PIPELINE is None:
|
@@ -183,7 +192,7 @@ def ensure_comparison_pipeline():
|
|
183 |
@spaces.GPU(duration=120)
|
184 |
def predict(prompt, temperature, top_p, min_new_tokens, max_new_tokens):
|
185 |
"""
|
186 |
-
|
187 |
"""
|
188 |
pipe = ensure_pipeline()
|
189 |
out = pipe(
|
@@ -199,7 +208,7 @@ def predict(prompt, temperature, top_p, min_new_tokens, max_new_tokens):
|
|
199 |
@spaces.GPU(duration=120)
|
200 |
def compare_models(prompt, temperature, top_p, min_new_tokens, max_new_tokens):
|
201 |
"""
|
202 |
-
Compare
|
203 |
"""
|
204 |
local_pipe = ensure_pipeline()
|
205 |
comp_pipe = ensure_comparison_pipeline()
|
@@ -224,75 +233,51 @@ def compare_models(prompt, temperature, top_p, min_new_tokens, max_new_tokens):
|
|
224 |
|
225 |
class ConversationRetriever:
|
226 |
"""
|
227 |
-
A simple in-memory
|
228 |
-
Each chunk is embedded
|
229 |
-
we embed the query, do similarity search, and retrieve top-k relevant chunks.
|
230 |
"""
|
231 |
-
|
232 |
def __init__(self, model_name="sentence-transformers/all-MiniLM-L6-v2", embed_dim=384):
|
233 |
-
"""
|
234 |
-
model_name: embedding model for messages
|
235 |
-
embed_dim: dimension of the embeddings from that model
|
236 |
-
"""
|
237 |
self.embed_model = SentenceTransformer(model_name)
|
238 |
self.embed_dim = embed_dim
|
239 |
-
|
240 |
self.index = faiss.IndexFlatL2(embed_dim)
|
241 |
-
self.texts = []
|
242 |
-
self.vectors = []
|
243 |
-
self.ids = []
|
244 |
-
|
245 |
self.id_counter = 0
|
246 |
|
247 |
def add_text(self, text):
|
248 |
-
"""
|
249 |
-
Add a new text chunk to the vector store.
|
250 |
-
Could chunk it up if desired, but here we treat the entire text as one chunk.
|
251 |
-
"""
|
252 |
if not text.strip():
|
253 |
return
|
254 |
-
|
255 |
emb = self.embed_model.encode([text], convert_to_numpy=True)
|
256 |
-
vec = emb[0].astype(np.float32)
|
257 |
self.index.add(vec.reshape(1, -1))
|
258 |
-
|
259 |
self.texts.append(text)
|
260 |
self.vectors.append(vec)
|
261 |
self.ids.append(self.id_counter)
|
262 |
-
|
263 |
self.id_counter += 1
|
264 |
|
265 |
def search(self, query, top_k=3):
|
266 |
-
"""
|
267 |
-
Given a query, embed it, do similarity search in FAISS, return top-k texts.
|
268 |
-
"""
|
269 |
q_emb = self.embed_model.encode([query], convert_to_numpy=True).astype(np.float32)
|
270 |
q_vec = q_emb[0].reshape(1, -1)
|
271 |
distances, indices = self.index.search(q_vec, top_k)
|
272 |
-
|
273 |
results = []
|
274 |
for dist, idx in zip(distances[0], indices[0]):
|
275 |
-
if idx < len(self.texts):
|
276 |
results.append((self.texts[idx], dist))
|
277 |
return results
|
278 |
|
279 |
-
retriever = ConversationRetriever()
|
280 |
|
281 |
def build_rag_prompt(user_query, retrieved_chunks):
|
282 |
"""
|
283 |
-
|
284 |
-
- The user's new query
|
285 |
-
- A "Relevant Context" section from retrieved chunks
|
286 |
-
- "Assistant:" to let the model continue
|
287 |
-
Feel free to customize the formatting as you like.
|
288 |
"""
|
289 |
context_str = ""
|
290 |
for i, (chunk, dist) in enumerate(retrieved_chunks):
|
291 |
-
context_str += f"Chunk #{i+1} (similarity
|
292 |
-
|
293 |
prompt = (
|
294 |
f"User's Query:\n{user_query}\n\n"
|
295 |
-
f"Relevant Context
|
296 |
"Assistant:"
|
297 |
)
|
298 |
return prompt
|
@@ -300,22 +285,13 @@ def build_rag_prompt(user_query, retrieved_chunks):
|
|
300 |
@spaces.GPU(duration=120)
|
301 |
def chat_rag(user_input, history, temperature, top_p, min_new_tokens, max_new_tokens):
|
302 |
"""
|
303 |
-
|
304 |
-
1) Add user input to FAISS
|
305 |
-
2) Retrieve top-k relevant older messages from FAISS
|
306 |
-
3) Build a prompt that includes the relevant chunks + user query
|
307 |
-
4) Generate a response from the pipeline
|
308 |
-
5) Add the assistant's response to FAISS as well
|
309 |
"""
|
310 |
pipe = ensure_pipeline()
|
311 |
-
|
312 |
retriever.add_text(f"User: {user_input}")
|
313 |
-
|
314 |
top_k = 3
|
315 |
results = retriever.search(user_input, top_k=top_k)
|
316 |
-
|
317 |
prompt = build_rag_prompt(user_input, results)
|
318 |
-
|
319 |
output = pipe(
|
320 |
prompt,
|
321 |
temperature=float(temperature),
|
@@ -331,16 +307,15 @@ def chat_rag(user_input, history, temperature, top_p, min_new_tokens, max_new_to
|
|
331 |
assistant_reply = output.strip()
|
332 |
|
333 |
retriever.add_text(f"Assistant: {assistant_reply}")
|
334 |
-
|
335 |
history.append([user_input, assistant_reply])
|
336 |
return history, history
|
337 |
|
|
|
338 |
with gr.Blocks() as demo:
|
339 |
-
gr.Markdown("# QLoRA Fine-tuning & RAG-based Chat Demo")
|
340 |
|
341 |
-
finetune_btn = gr.Button("Finetune 4-bit (QLoRA) on
|
342 |
status_box = gr.Textbox(label="Finetune Status")
|
343 |
-
|
344 |
finetune_btn.click(fn=finetune_small_subset, outputs=status_box)
|
345 |
|
346 |
gr.Markdown("## Direct Generation (No Retrieval)")
|
@@ -349,19 +324,18 @@ with gr.Blocks() as demo:
|
|
349 |
top_p = gr.Slider(0.0, 1.0, step=0.05, value=0.9, label="Top-p")
|
350 |
min_tokens = gr.Slider(1, 2500, value=50, step=10, label="Min New Tokens")
|
351 |
max_tokens = gr.Slider(1, 2500, value=200, step=50, label="Max New Tokens")
|
352 |
-
|
353 |
-
|
354 |
-
gen_btn = gr.Button("Generate with myr1")
|
355 |
gen_btn.click(
|
356 |
fn=predict,
|
357 |
inputs=[prompt_in, temperature, top_p, min_tokens, max_tokens],
|
358 |
outputs=output_box
|
359 |
)
|
360 |
|
361 |
-
gr.Markdown("## Compare
|
362 |
compare_btn = gr.Button("Compare")
|
363 |
-
out_local = gr.Textbox(label="
|
364 |
-
out_deepseek = gr.Textbox(label="DeepSeek Output", lines=6)
|
365 |
compare_btn.click(
|
366 |
fn=compare_models,
|
367 |
inputs=[prompt_in, temperature, top_p, min_tokens, max_tokens],
|
@@ -372,15 +346,13 @@ with gr.Blocks() as demo:
|
|
372 |
with gr.Row():
|
373 |
with gr.Column():
|
374 |
chatbot = gr.Chatbot(label="RAG Chat")
|
375 |
-
chat_state = gr.State([])
|
376 |
-
|
377 |
user_input = gr.Textbox(
|
378 |
show_label=False,
|
379 |
placeholder="Ask a question...",
|
380 |
lines=2
|
381 |
)
|
382 |
send_btn = gr.Button("Send")
|
383 |
-
|
384 |
user_input.submit(
|
385 |
fn=chat_rag,
|
386 |
inputs=[user_input, chat_state, temperature, top_p, min_tokens, max_tokens],
|
@@ -392,4 +364,4 @@ with gr.Blocks() as demo:
|
|
392 |
outputs=[chat_state, chatbot]
|
393 |
)
|
394 |
|
395 |
-
demo.launch()
|
|
|
20 |
|
21 |
from sentence_transformers import SentenceTransformer
|
22 |
|
23 |
+
# Import your custom configuration overrides.
|
24 |
+
# For example, your configuration_deepseek.py might export a dictionary called CONFIG_OVERRIDES.
|
25 |
+
import configuration_deepseek
|
26 |
+
|
27 |
+
# Global variables for pipelines and settings.
|
28 |
TEXT_PIPELINE = None
|
29 |
COMPARISON_PIPELINE = None
|
30 |
NUM_EXAMPLES = 1000
|
|
|
32 |
@spaces.GPU(duration=300)
|
33 |
def finetune_small_subset():
|
34 |
"""
|
35 |
+
1) Loads your custom model ("wuhp/myr1") in 4-bit quantization (QLoRA style),
|
36 |
2) Adds LoRA adapters (trainable),
|
37 |
+
3) Fine-tunes on a small subset of the ServiceNow-AI/R1-Distill-SFT dataset,
|
38 |
+
4) Saves the LoRA adapter to "finetuned_myr1",
|
39 |
+
5) Reloads the LoRA adapter for inference.
|
40 |
"""
|
41 |
+
# Load the new dataset.
|
42 |
+
ds = load_dataset("ServiceNow-AI/R1-Distill-SFT", split="train")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
43 |
ds = ds.select(range(min(NUM_EXAMPLES, len(ds))))
|
44 |
|
45 |
bnb_config = BitsAndBytesConfig(
|
46 |
load_in_4bit=True,
|
47 |
+
bnb_4bit_compute_dtype=torch.bfloat16,
|
48 |
bnb_4bit_use_double_quant=True,
|
49 |
bnb_4bit_quant_type="nf4",
|
50 |
)
|
51 |
|
52 |
+
# Load the base configuration from your model repository.
|
53 |
+
base_config = AutoConfig.from_pretrained(
|
54 |
+
"wuhp/myr1",
|
55 |
subfolder="myr1",
|
56 |
+
trust_remote_code=True,
|
57 |
)
|
58 |
+
# Apply your custom overrides (from configuration_deepseek.py).
|
59 |
+
for key, value in configuration_deepseek.CONFIG_OVERRIDES.items():
|
60 |
+
setattr(base_config, key, value)
|
61 |
+
|
62 |
tokenizer = AutoTokenizer.from_pretrained(
|
63 |
+
"wuhp/myr1",
|
64 |
subfolder="myr1",
|
65 |
trust_remote_code=True
|
66 |
)
|
67 |
|
68 |
+
# Load the model. With trust_remote_code=True, your custom model class (e.g. DeepseekV3ForCausalLM)
|
69 |
+
# will be loaded from the repository.
|
70 |
base_model = AutoModelForCausalLM.from_pretrained(
|
71 |
"wuhp/myr1",
|
72 |
subfolder="myr1",
|
73 |
+
config=base_config,
|
74 |
+
quantization_config=bnb_config,
|
75 |
device_map="auto",
|
76 |
trust_remote_code=True
|
77 |
)
|
|
|
88 |
)
|
89 |
lora_model = get_peft_model(base_model, lora_config)
|
90 |
|
91 |
+
# For this dataset, assume "problem" is the prompt and "solution" is the target.
|
92 |
def tokenize_fn(ex):
|
93 |
text = (
|
94 |
+
f"Problem: {ex['problem']}\n\n"
|
95 |
+
f"Solution: {ex['solution']}"
|
96 |
)
|
97 |
return tokenizer(text, truncation=True, max_length=512)
|
98 |
|
|
|
107 |
per_device_train_batch_size=1,
|
108 |
gradient_accumulation_steps=2,
|
109 |
logging_steps=5,
|
110 |
+
save_steps=999999, # High save interval
|
111 |
save_total_limit=1,
|
112 |
+
fp16=False, # Set to True if supported by your hardware
|
113 |
)
|
114 |
|
115 |
trainer = Trainer(
|
|
|
120 |
)
|
121 |
trainer.train()
|
122 |
|
123 |
+
# Save the LoRA adapter and tokenizer.
|
124 |
trainer.model.save_pretrained("finetuned_myr1")
|
125 |
tokenizer.save_pretrained("finetuned_myr1")
|
126 |
|
127 |
+
# Reload the base model and attach the LoRA adapter for inference.
|
128 |
base_model_2 = AutoModelForCausalLM.from_pretrained(
|
129 |
"wuhp/myr1",
|
130 |
subfolder="myr1",
|
131 |
+
config=base_config,
|
132 |
quantization_config=bnb_config,
|
133 |
device_map="auto",
|
134 |
trust_remote_code=True
|
|
|
147 |
|
148 |
def ensure_pipeline():
|
149 |
"""
|
150 |
+
If we haven't fine-tuned yet (i.e. TEXT_PIPELINE is None),
|
151 |
+
load the base model (without LoRA) in 4-bit mode.
|
152 |
"""
|
153 |
global TEXT_PIPELINE
|
154 |
if TEXT_PIPELINE is None:
|
|
|
158 |
bnb_4bit_use_double_quant=True,
|
159 |
bnb_4bit_quant_type="nf4",
|
160 |
)
|
161 |
+
base_config = AutoConfig.from_pretrained("wuhp/myr1", subfolder="myr1", trust_remote_code=True)
|
162 |
+
for key, value in configuration_deepseek.CONFIG_OVERRIDES.items():
|
163 |
+
setattr(base_config, key, value)
|
164 |
tokenizer = AutoTokenizer.from_pretrained("wuhp/myr1", subfolder="myr1", trust_remote_code=True)
|
165 |
base_model = AutoModelForCausalLM.from_pretrained(
|
166 |
"wuhp/myr1",
|
167 |
subfolder="myr1",
|
168 |
+
config=base_config,
|
169 |
quantization_config=bnb_config,
|
170 |
device_map="auto",
|
171 |
trust_remote_code=True
|
|
|
175 |
|
176 |
def ensure_comparison_pipeline():
|
177 |
"""
|
178 |
+
Load a reference DeepSeek model pipeline if not already loaded.
|
179 |
"""
|
180 |
global COMPARISON_PIPELINE
|
181 |
if COMPARISON_PIPELINE is None:
|
|
|
192 |
@spaces.GPU(duration=120)
|
193 |
def predict(prompt, temperature, top_p, min_new_tokens, max_new_tokens):
|
194 |
"""
|
195 |
+
Direct generation without retrieval.
|
196 |
"""
|
197 |
pipe = ensure_pipeline()
|
198 |
out = pipe(
|
|
|
208 |
@spaces.GPU(duration=120)
|
209 |
def compare_models(prompt, temperature, top_p, min_new_tokens, max_new_tokens):
|
210 |
"""
|
211 |
+
Compare outputs between your custom model and a reference DeepSeek model.
|
212 |
"""
|
213 |
local_pipe = ensure_pipeline()
|
214 |
comp_pipe = ensure_comparison_pipeline()
|
|
|
233 |
|
234 |
class ConversationRetriever:
|
235 |
"""
|
236 |
+
A simple in-memory FAISS-based retriever.
|
237 |
+
Each text chunk is embedded using SentenceTransformer.
|
|
|
238 |
"""
|
|
|
239 |
def __init__(self, model_name="sentence-transformers/all-MiniLM-L6-v2", embed_dim=384):
|
|
|
|
|
|
|
|
|
240 |
self.embed_model = SentenceTransformer(model_name)
|
241 |
self.embed_dim = embed_dim
|
|
|
242 |
self.index = faiss.IndexFlatL2(embed_dim)
|
243 |
+
self.texts = []
|
244 |
+
self.vectors = []
|
245 |
+
self.ids = []
|
|
|
246 |
self.id_counter = 0
|
247 |
|
248 |
def add_text(self, text):
|
|
|
|
|
|
|
|
|
249 |
if not text.strip():
|
250 |
return
|
|
|
251 |
emb = self.embed_model.encode([text], convert_to_numpy=True)
|
252 |
+
vec = emb[0].astype(np.float32)
|
253 |
self.index.add(vec.reshape(1, -1))
|
|
|
254 |
self.texts.append(text)
|
255 |
self.vectors.append(vec)
|
256 |
self.ids.append(self.id_counter)
|
|
|
257 |
self.id_counter += 1
|
258 |
|
259 |
def search(self, query, top_k=3):
|
|
|
|
|
|
|
260 |
q_emb = self.embed_model.encode([query], convert_to_numpy=True).astype(np.float32)
|
261 |
q_vec = q_emb[0].reshape(1, -1)
|
262 |
distances, indices = self.index.search(q_vec, top_k)
|
|
|
263 |
results = []
|
264 |
for dist, idx in zip(distances[0], indices[0]):
|
265 |
+
if idx < len(self.texts):
|
266 |
results.append((self.texts[idx], dist))
|
267 |
return results
|
268 |
|
269 |
+
retriever = ConversationRetriever()
|
270 |
|
271 |
def build_rag_prompt(user_query, retrieved_chunks):
|
272 |
"""
|
273 |
+
Build a prompt for retrieval-augmented generation.
|
|
|
|
|
|
|
|
|
274 |
"""
|
275 |
context_str = ""
|
276 |
for i, (chunk, dist) in enumerate(retrieved_chunks):
|
277 |
+
context_str += f"Chunk #{i+1} (similarity ~ {dist:.2f}):\n{chunk}\n\n"
|
|
|
278 |
prompt = (
|
279 |
f"User's Query:\n{user_query}\n\n"
|
280 |
+
f"Relevant Context:\n{context_str}"
|
281 |
"Assistant:"
|
282 |
)
|
283 |
return prompt
|
|
|
285 |
@spaces.GPU(duration=120)
|
286 |
def chat_rag(user_input, history, temperature, top_p, min_new_tokens, max_new_tokens):
|
287 |
"""
|
288 |
+
Chat function with retrieval augmentation.
|
|
|
|
|
|
|
|
|
|
|
289 |
"""
|
290 |
pipe = ensure_pipeline()
|
|
|
291 |
retriever.add_text(f"User: {user_input}")
|
|
|
292 |
top_k = 3
|
293 |
results = retriever.search(user_input, top_k=top_k)
|
|
|
294 |
prompt = build_rag_prompt(user_input, results)
|
|
|
295 |
output = pipe(
|
296 |
prompt,
|
297 |
temperature=float(temperature),
|
|
|
307 |
assistant_reply = output.strip()
|
308 |
|
309 |
retriever.add_text(f"Assistant: {assistant_reply}")
|
|
|
310 |
history.append([user_input, assistant_reply])
|
311 |
return history, history
|
312 |
|
313 |
+
# Build the Gradio interface.
|
314 |
with gr.Blocks() as demo:
|
315 |
+
gr.Markdown("# QLoRA Fine-tuning & RAG-based Chat Demo using Custom DeepSeekV3 Model")
|
316 |
|
317 |
+
finetune_btn = gr.Button("Finetune 4-bit (QLoRA) on ServiceNow-AI/R1-Distill-SFT subset (up to 5 min)")
|
318 |
status_box = gr.Textbox(label="Finetune Status")
|
|
|
319 |
finetune_btn.click(fn=finetune_small_subset, outputs=status_box)
|
320 |
|
321 |
gr.Markdown("## Direct Generation (No Retrieval)")
|
|
|
324 |
top_p = gr.Slider(0.0, 1.0, step=0.05, value=0.9, label="Top-p")
|
325 |
min_tokens = gr.Slider(1, 2500, value=50, step=10, label="Min New Tokens")
|
326 |
max_tokens = gr.Slider(1, 2500, value=200, step=50, label="Max New Tokens")
|
327 |
+
output_box = gr.Textbox(label="DeepSeekV3 Output", lines=8)
|
328 |
+
gen_btn = gr.Button("Generate with DeepSeekV3")
|
|
|
329 |
gen_btn.click(
|
330 |
fn=predict,
|
331 |
inputs=[prompt_in, temperature, top_p, min_tokens, max_tokens],
|
332 |
outputs=output_box
|
333 |
)
|
334 |
|
335 |
+
gr.Markdown("## Compare DeepSeekV3 vs Reference DeepSeek")
|
336 |
compare_btn = gr.Button("Compare")
|
337 |
+
out_local = gr.Textbox(label="DeepSeekV3 Output", lines=6)
|
338 |
+
out_deepseek = gr.Textbox(label="Reference DeepSeek Output", lines=6)
|
339 |
compare_btn.click(
|
340 |
fn=compare_models,
|
341 |
inputs=[prompt_in, temperature, top_p, min_tokens, max_tokens],
|
|
|
346 |
with gr.Row():
|
347 |
with gr.Column():
|
348 |
chatbot = gr.Chatbot(label="RAG Chat")
|
349 |
+
chat_state = gr.State([])
|
|
|
350 |
user_input = gr.Textbox(
|
351 |
show_label=False,
|
352 |
placeholder="Ask a question...",
|
353 |
lines=2
|
354 |
)
|
355 |
send_btn = gr.Button("Send")
|
|
|
356 |
user_input.submit(
|
357 |
fn=chat_rag,
|
358 |
inputs=[user_input, chat_state, temperature, top_p, min_tokens, max_tokens],
|
|
|
364 |
outputs=[chat_state, chatbot]
|
365 |
)
|
366 |
|
367 |
+
demo.launch()
|