Datasets:

Modalities:
Tabular
Text
Formats:
parquet
DOI:
Libraries:
Datasets
Dask
License:
anton-l HF staff commited on
Commit
8f233cf
1 Parent(s): 690f4f6

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +1 -1
README.md CHANGED
@@ -210,7 +210,7 @@ Analysis revealed that FineWeb's C4 filter removes pages containing '{' characte
210
  ### Phase 3: Refined quality assessment
211
  The expanded corpus underwent a more fine-grained quality evaluation:
212
 
213
- Once again, we used LLama-3.1-70B-Instruct to score a sample of newly extracted pages on a 5-point scale (full prompt available in [here](prompt.txt)):
214
  We finetuned a new [classifier](https://huggingface.co/HuggingFaceTB/finemath-classifier) on these annotations and scored the entire corpus.
215
  After leaving only pages with a score of 3 or higher, and deduplicating the samples using simple single-band MinHash-LSH, we obtained FineMath-3+ with 34B tokens.
216
 
 
210
  ### Phase 3: Refined quality assessment
211
  The expanded corpus underwent a more fine-grained quality evaluation:
212
 
213
+ Once again, we used LLama-3.1-70B-Instruct to score a sample of newly extracted pages on a 5-point scale (full prompt available in [here](assets/prompt.txt)):
214
  We finetuned a new [classifier](https://huggingface.co/HuggingFaceTB/finemath-classifier) on these annotations and scored the entire corpus.
215
  After leaving only pages with a score of 3 or higher, and deduplicating the samples using simple single-band MinHash-LSH, we obtained FineMath-3+ with 34B tokens.
216