summaryrefslogtreecommitdiff
path: root/llmc/cmd/data/fineweb.go
diff options
context:
space:
mode:
authorJonas Knobloch <jonas.knobloch@t-online.de>2026-04-22 15:36:58 +0200
committerJonas Knobloch <jonas.knobloch@t-online.de>2026-04-23 20:17:53 +0200
commitbabd70eb9d4e4b028c0eb8c32e6bed65dafa0481 (patch)
tree75635ad8e0ef6541b7b5d6926f529830b32f715b /llmc/cmd/data/fineweb.go
parent54b11da0d2f853395f270379044674363b1a124a (diff)
Add llmc module
Diffstat (limited to 'llmc/cmd/data/fineweb.go')
-rw-r--r--llmc/cmd/data/fineweb.go43
1 files changed, 43 insertions, 0 deletions
diff --git a/llmc/cmd/data/fineweb.go b/llmc/cmd/data/fineweb.go
new file mode 100644
index 0000000..cabfa74
--- /dev/null
+++ b/llmc/cmd/data/fineweb.go
@@ -0,0 +1,43 @@
+package main
+
+import (
+ "log"
+
+ "go.jknobloc.com/x/dataset"
+ "go.jknobloc.com/x/llm"
+ "go.jknobloc.com/x/llmc"
+ "go.jknobloc.com/x/tokenizer/bpe"
+)
+
+// classic + 10B: fineweb10B; sample-10BT
+// classic + 100B: fineweb100B; sample-100BT
+// edu + 10B: edu_fineweb10B; sample-10BT
+// edu + 100B: edu_fineweb100B; sample-100BT
+
+func fineWeb() {
+ var reader dataset.Reader
+
+ if r, err := dataset.NewParquetReader("artifacts/data/raw/fineweb-edu/sample-100BT/train"); err != nil {
+ log.Fatal(err)
+ } else {
+ reader = r
+ }
+
+ var tokenizer llm.Tokenizer
+
+ if t, err := bpe.NewTokenizerFromFiles("gpt2/models/base/vocab.json", "gpt2/models/base/merges.txt"); err != nil {
+ log.Fatal(err)
+ } else {
+ tokenizer = t
+ }
+
+ docs := llmc.TokenizeAll(reader, tokenizer, 50256)
+
+ if err := llmc.WriteShards("artifacts/data/llmc/edu_fineweb100B", "edu_fineweb", 100_000_000, docs); err != nil {
+ log.Fatal(err)
+ }
+
+ if err := reader.Err(); err != nil {
+ log.Fatal(err)
+ }
+}