aboutsummaryrefslogtreecommitdiffstats
path: root/scripts/download.sh
diff options
context:
space:
mode:
Diffstat (limited to 'scripts/download.sh')
-rwxr-xr-xscripts/download.sh28
1 files changed, 28 insertions, 0 deletions
diff --git a/scripts/download.sh b/scripts/download.sh
new file mode 100755
index 0000000..91c8a00
--- /dev/null
+++ b/scripts/download.sh
@@ -0,0 +1,28 @@
+#!/usr/bin/env bash
+# Downloads the base model (SmolLM2-135M-Instruct) and the dolly-15k dataset
+# into the default locations llm-tune expects.
+set -euo pipefail
+cd "$(dirname "$0")/.."
+
+HF=https://huggingface.co
+MODEL=models/smollm2-135m-instruct
+DATA=data/dolly-15k
+
+fetch() { # url dest
+ [ -s "$2" ] && { echo "ok $2"; return; }
+ echo "get $2"
+ curl -fL --progress-bar -o "$2.part" "$1" && mv "$2.part" "$2"
+}
+
+mkdir -p "$MODEL" "$DATA"
+for f in config.json generation_config.json model.safetensors tokenizer.json tokenizer_config.json; do
+ fetch "$HF/HuggingFaceTB/SmolLM2-135M-Instruct/resolve/main/$f" "$MODEL/$f"
+done
+fetch "$HF/datasets/databricks/databricks-dolly-15k/resolve/main/databricks-dolly-15k.jsonl" \
+ "$DATA/databricks-dolly-15k.jsonl"
+
+sha256sum -c --quiet <<SUMS
+5af571cbf074e6d21a03528d2330792e532ca608f24ac70a143f6b369968ab8c $MODEL/model.safetensors
+2df9083338b4abd6bceb5635764dab5d833b393b55759dffb0959b6fcbf794ec $DATA/databricks-dolly-15k.jsonl
+SUMS
+echo "done"