afrizalha commited on
Commit
91a8143
·
verified ·
1 Parent(s): 1940b48

Update app.py

Browse files
Files changed (1) hide show
  1. app.py +11 -10
app.py CHANGED
@@ -10,29 +10,30 @@ tinier = AutoModelForCausalLM.from_pretrained("afrizalha/Sasando-1-7M", token=hf
10
 
11
  desc = """Sasando-1 is a tiny, highly experimental text generator built using the Phi-3 architecture. It comes with two variations of microscopic sizes: 7M and 25M parameters. It is trained on a tightly-controlled Indo4B dataset filtered to only have 18000 unique words. The method is inspired by Microsoft's TinyStories paper which demonstrates that a tiny language model can produce fluent text when trained on tightly-controlled dataset.\n\nTry prompting with two simple words, and let the model continue. Fun examples provided below."""
12
 
13
- def generate(starting_text, choice, num_runs,temp,top_p):
14
- if choice == '7M':
15
  model = tinier
16
- elif choice == '25M':
17
  model = tiny
18
  elif choice == 'Info':
19
- return desc
20
-
 
21
  results = []
22
  for i in range(5):
23
  inputs = tokenizer([starting_text], return_tensors="pt").to(model.device)
24
  outputs = model.generate(
25
  inputs=inputs.input_ids,
26
- max_new_tokens=32-len(inputs.input_ids),
27
- do_sample=True,
28
  temperature=temp,
29
  top_p=top_p
30
  )
31
- outputs = tokenizer.batch_decode(outputs,skip_special_tokens=True)[0]
32
  outputs = outputs[:outputs.find(".")]
33
  results.append(outputs)
34
- yield "\n\n".join(results)
35
-
36
  with gr.Blocks(theme=gr.themes.Soft()) as app:
37
  starting_text = gr.Textbox(label="Starting text", value="cinta adalah")
38
  choice = gr.Radio(["7M", "25M", "Info"], label="Select model", info="Built with the Phi-3 architecture")
 
10
 
11
  desc = """Sasando-1 is a tiny, highly experimental text generator built using the Phi-3 architecture. It comes with two variations of microscopic sizes: 7M and 25M parameters. It is trained on a tightly-controlled Indo4B dataset filtered to only have 18000 unique words. The method is inspired by Microsoft's TinyStories paper which demonstrates that a tiny language model can produce fluent text when trained on tightly-controlled dataset.\n\nTry prompting with two simple words, and let the model continue. Fun examples provided below."""
12
 
13
+ def generate(starting_text, choice, temp, top_p):
14
+ if choice == '7M':
15
  model = tinier
16
+ elif choice == '25M':
17
  model = tiny
18
  elif choice == 'Info':
19
+ yield desc
20
+ return
21
+
22
  results = []
23
  for i in range(5):
24
  inputs = tokenizer([starting_text], return_tensors="pt").to(model.device)
25
  outputs = model.generate(
26
  inputs=inputs.input_ids,
27
+ max_new_tokens=32-len(inputs.input_ids[0]),
28
+ do_sample=True,
29
  temperature=temp,
30
  top_p=top_p
31
  )
32
+ outputs = tokenizer.batch_decode(outputs, skip_special_tokens=True)[0]
33
  outputs = outputs[:outputs.find(".")]
34
  results.append(outputs)
35
+ yield "\n\n".join(results)
36
+
37
  with gr.Blocks(theme=gr.themes.Soft()) as app:
38
  starting_text = gr.Textbox(label="Starting text", value="cinta adalah")
39
  choice = gr.Radio(["7M", "25M", "Info"], label="Select model", info="Built with the Phi-3 architecture")