Update app.py

app.py

@@ -23,7 +23,7 @@ base_model = AutoModelForCausalLM.from_pretrained(
     token=access_token,
     trust_remote_code=True, 
     device_map="auto",
-    load_in_8bit=True,
+    torch_dtype=torch.float16,
     offload_folder="offload/"
 )
 model = PeftModel.from_pretrained(
@@ -48,6 +48,7 @@ SYSTEM_PROMPT = "You are a seasoned stock market analyst. Your task is to list t
 
 
 def print_gpu_utilization():
+    
     nvmlInit()
     handle = nvmlDeviceGetHandleByIndex(0)
     info = nvmlDeviceGetMemoryInfo(handle)
@@ -55,31 +56,37 @@ def print_gpu_utilization():
 
 
 def get_curday():
+    
     return date.today().strftime("%Y-%m-%d")
 
 
 def n_weeks_before(date_string, n):
+    
     date = datetime.strptime(date_string, "%Y-%m-%d") - timedelta(days=7*n)
+
     return date.strftime("%Y-%m-%d")
 
 
 def get_stock_data(stock_symbol, steps):
+
     stock_data = yf.download(stock_symbol, steps[0], steps[-1])
     if len(stock_data) == 0:
         raise gr.Error(f"Failed to download stock price data for symbol {stock_symbol} from yfinance!")
     
+#     print(stock_data)
+    
     dates, prices = [], []
-    available_dates = stock_data.index.astype(str).tolist()
+    available_dates = stock_data.index.format()
     
     for date in steps[:-1]:
         for i in range(len(stock_data)):
             if available_dates[i] >= date:
-                prices.append(stock_data['Close'].iloc[i])
+                prices.append(stock_data['Close'][i])
                 dates.append(datetime.strptime(available_dates[i], "%Y-%m-%d"))
                 break
 
     dates.append(datetime.strptime(available_dates[-1], "%Y-%m-%d"))
-    prices.append(stock_data['Close'].iloc[-1])
+    prices.append(stock_data['Close'][-1])
     
     return pd.DataFrame({
         "Start Date": dates[:-1], "End Date": dates[1:],
@@ -88,12 +95,14 @@ def get_stock_data(stock_symbol, steps):
 
 
 def get_news(symbol, data):
+    
     news_list = []
     
-    for _, row in data.iterrows():
+    for end_date, row in data.iterrows():
         start_date = row['Start Date'].strftime('%Y-%m-%d')
         end_date = row['End Date'].strftime('%Y-%m-%d')
-        time.sleep(1)  # control qpm
+#         print(symbol, ': ', start_date, ' - ', end_date)
+        time.sleep(1) # control qpm
         weekly_news = finnhub_client.company_news(symbol, _from=start_date, to=end_date)
         if len(weekly_news) == 0:
             raise gr.Error(f"No company news found for symbol {symbol} from finnhub!")
@@ -105,7 +114,7 @@ def get_news(symbol, data):
             } for n in weekly_news
         ]
         weekly_news.sort(key=lambda x: x['date'])
-        news_list.append(weekly_news)
+        news_list.append(json.dumps(weekly_news))
     
     data['News'] = news_list
     
@@ -113,6 +122,7 @@ def get_news(symbol, data):
 
 
 def get_company_prompt(symbol):
+
     profile = finnhub_client.company_profile2(symbol=symbol)
     if not profile:
         raise gr.Error(f"Failed to find company profile for symbol {symbol} from finnhub!")
@@ -126,14 +136,15 @@ def get_company_prompt(symbol):
 
 
 def get_prompt_by_row(symbol, row):
+
     start_date = row['Start Date'] if isinstance(row['Start Date'], str) else row['Start Date'].strftime('%Y-%m-%d')
     end_date = row['End Date'] if isinstance(row['End Date'], str) else row['End Date'].strftime('%Y-%m-%d')
     term = 'increased' if row['End Price'] > row['Start Price'] else 'decreased'
     head = "From {} to {}, {}'s stock price {} from {:.2f} to {:.2f}. Company news during this period are listed below:\n\n".format(
         start_date, end_date, symbol, term, row['Start Price'], row['End Price'])
     
-    news = row["News"] if isinstance(row["News"], list) else json.loads(row["News"])
-    news_formatted = ["[Headline]: {}\n[Summary]: {}\n".format(
+    news = json.loads(row["News"])
+    news = ["[Headline]: {}\n[Summary]: {}\n".format(
         n['headline'], n['summary']) for n in news if n['date'][:8] <= end_date.replace('-', '') and \
         not n['summary'].startswith("Looking for stock market analysis and research with proves results?")]
 
@@ -144,21 +155,20 @@ def get_prompt_by_row(symbol, row):
     else:
         basics = "[Basic Financials]:\n\nNo basic financial reported."
     
-    return head, news_formatted, basics
+    return head, news, basics
 
 
 def sample_news(news, k=5):
+    
     return [news[i] for i in sorted(random.sample(range(len(news)), k))]
-
-
+    
 def latest_news(news, k=5):
-    if not isinstance(news, list) or not all(isinstance(item, dict) for item in news):
-        raise ValueError("News must be a list of dictionaries.")
     sorted_news = sorted(news, key=lambda x: x['date'], reverse=True)
     return sorted_news[:k]
 
 
 def get_current_basics(symbol, curday):
+
     basic_financials = finnhub_client.company_basic_financials(symbol, 'all')
     if not basic_financials['series']:
         raise gr.Error(f"Failed to find basic financials for symbol {symbol} from finnhub!")
@@ -183,6 +193,7 @@ def get_current_basics(symbol, curday):
     
 
 def get_all_prompts_online(symbol, data, curday, with_basics=True):
+
     company_prompt = get_company_prompt(symbol)
 
     prev_rows = []
@@ -194,7 +205,10 @@ def get_all_prompts_online(symbol, data, curday, with_basics=True):
     prompt = ""
     for i in range(-len(prev_rows), 0):
         prompt += "\n" + prev_rows[i][0]
-        latest_news_items = latest_news(prev_rows[i][1], min(5, len(prev_rows[i][1])))
+        latest_news_items = latest_news(
+            prev_rows[i][1],
+            min(5, len(prev_rows[i][1]))
+        )
         if latest_news_items:
             prompt += "\n".join(latest_news_items)
         else:
@@ -216,7 +230,9 @@ def get_all_prompts_online(symbol, data, curday, with_basics=True):
     return info, prompt
 
 
+
 def construct_prompt(ticker, curday, n_weeks, use_basics):
+
     try:
         steps = [n_weeks_before(curday, n) for n in range(n_weeks + 1)][::-1]
     except Exception:
@@ -225,26 +241,31 @@ def construct_prompt(ticker, curday, n_weeks, use_basics):
     data = get_stock_data(ticker, steps)
     data = get_news(ticker, data)
     data['Basics'] = [json.dumps({})] * len(data)
+    # print(data)
     
     info, prompt = get_all_prompts_online(ticker, data, curday, use_basics)
     
     prompt = B_INST + B_SYS + SYSTEM_PROMPT + E_SYS + prompt + E_INST
+    # print(prompt)
     
     return info, prompt
 
 
 def predict(ticker, date, n_weeks, use_basics):
+
     print_gpu_utilization()
 
     info, prompt = construct_prompt(ticker, date, n_weeks, use_basics)
       
-    inputs = tokenizer(prompt, return_tensors='pt', padding=False)
+    inputs = tokenizer(
+        prompt, return_tensors='pt', padding=False
+    )
     inputs = {key: value.to(model.device) for key, value in inputs.items()}
 
     print("Inputs loaded onto devices.")
         
     res = model.generate(
-        **inputs, max_length=4096, do_sample=False,
+        **inputs, max_length=4096, do_sample=True,
         eos_token_id=tokenizer.eos_token_id,
         use_cache=True, streamer=streamer
     )
@@ -291,9 +312,14 @@ demo = gr.Interface(
             label="Response"
         )
     ],
-    title="Pro Capital",
-    description="""Implementation**
+    title="FinGPT-Forecaster",
+    description="""FinGPT-Forecaster takes random market news and optional basic financials related to the specified company from the past few weeks as input and responds with the company's **positive developments** and **potential concerns**. Then it gives out a **prediction** of stock price movement for the coming week and its **analysis** summary.
+This model is finetuned on Llama2-7b-chat-hf with LoRA on the past year's DOW30 market data. Inference in this demo uses fp16 and **welcomes any ticker symbol**.
+Company profile & Market news & Basic financials & Stock prices are retrieved using **yfinance & finnhub**.
+This is just a demo showing what this model is capable of. Results inferred from randomly chosen news can be strongly biased.
+For more detailed and customized implementation, refer to our FinGPT project: <https://github.com/AI4Finance-Foundation/FinGPT>
+**Disclaimer: Nothing herein is financial advice, and NOT a recommendation to trade real money. Please use common sense and always first consult a professional before trading or investing.**
 """
 )
 
-demo.launch()
+demo.launch()