oberbics commited on
Commit
a606c0a
·
verified ·
1 Parent(s): 473abc3

Update app.py

Browse files
Files changed (1) hide show
  1. app.py +5 -4
app.py CHANGED
@@ -10,7 +10,7 @@ import spaces
10
 
11
 
12
  # Model configuration
13
- MODEL_ID = "oberbics/llama-3.1-newspaper-arguments-grpo-stable-v2"
14
 
15
 
16
  SYSTEM_PROMPT = """You are an expert at analyzing historical texts and you hate to summarize
@@ -26,7 +26,6 @@ OUTPUT FORMAT - EXACTLY these 4 XML tags and NOTHING else:
26
  <explanation>The prefect explicitly argues that existing help is "insufficient" and makes a direct claim that "extraordinary measures" are needed, presenting a clear premise-conclusion structure about the inadequacy of current response.</explanation>
27
  <human_verification_needed>False</human_verification_needed>
28
 
29
-
30
  EXAMPLE WITHOUT ARGUMENT:
31
  <argument>NA</argument>
32
  <claim>NA</claim>
@@ -34,6 +33,9 @@ EXAMPLE WITHOUT ARGUMENT:
34
  <human_verification_needed>FALSE</human_verification_needed>
35
 
36
  RULES:
 
 
 
37
  - NO SUMMARY; ONLY EXACT EXTRACTOM FROM THE TEXT; don't extract anything that is not in the text. Only extract word by word
38
  - ONLY output these 4 XML tags
39
  - Extract only original text without changes or use NA when you did not find an argument
@@ -41,12 +43,11 @@ RULES:
41
  - The CLAIM should say what the (implicite) argument implies, what the main conclusion is
42
  - Give attention to implicit argumetns
43
  - In cases of uncertainty or ambiguity, say human_verification_needed TRUE
44
- - If no argument exists, use NA for all fields except <human_verification_needed>FALSE or TRUE</human_verification_needed>
45
  - More than one argument possible for one aticle, one unit has one clear clame and all the xml structures
46
 
47
  VERIFICATION: BEfore you print the results, double check claims and explanations of the argument. When the claim is just a translation or the explanation states that this is not an argument, don't print it"""
48
 
49
-
50
  print("Loading tokenizer...")
51
  tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)
52
  tokenizer.pad_token = tokenizer.eos_token
 
10
 
11
 
12
  # Model configuration
13
+ MODEL_ID = "oberbics/llama-3.1-argument_mining-v2"
14
 
15
 
16
  SYSTEM_PROMPT = """You are an expert at analyzing historical texts and you hate to summarize
 
26
  <explanation>The prefect explicitly argues that existing help is "insufficient" and makes a direct claim that "extraordinary measures" are needed, presenting a clear premise-conclusion structure about the inadequacy of current response.</explanation>
27
  <human_verification_needed>False</human_verification_needed>
28
 
 
29
  EXAMPLE WITHOUT ARGUMENT:
30
  <argument>NA</argument>
31
  <claim>NA</claim>
 
33
  <human_verification_needed>FALSE</human_verification_needed>
34
 
35
  RULES:
36
+ - NEVER REPEAT ARGUMENTS, NEVER PRINT ARGUMENTS DOUBLE
37
+ - TRUE FOR VERIFICATION WHEN UNCERTAIN
38
+ - Only output arguments that appear verbatim (or nearly verbatim) in the text
39
  - NO SUMMARY; ONLY EXACT EXTRACTOM FROM THE TEXT; don't extract anything that is not in the text. Only extract word by word
40
  - ONLY output these 4 XML tags
41
  - Extract only original text without changes or use NA when you did not find an argument
 
43
  - The CLAIM should say what the (implicite) argument implies, what the main conclusion is
44
  - Give attention to implicit argumetns
45
  - In cases of uncertainty or ambiguity, say human_verification_needed TRUE
46
+ - If no argument exists, use NA for ALL fields without explanation except <human_verification_needed>FALSE or TRUE</human_verification_needed>
47
  - More than one argument possible for one aticle, one unit has one clear clame and all the xml structures
48
 
49
  VERIFICATION: BEfore you print the results, double check claims and explanations of the argument. When the claim is just a translation or the explanation states that this is not an argument, don't print it"""
50
 
 
51
  print("Loading tokenizer...")
52
  tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)
53
  tokenizer.pad_token = tokenizer.eos_token