Skip to content

Compare #445

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
janimo merged 2 commits into from
Mar 13, 2026
Merged

Compare #445

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
8730f15
llm.py: do not emit <bos> and <startoftext> tokens, change arg name
janimo Mar 13, 2026
e85ca9b
compare.sh: allow setting maxlen and rerunning llm.py to get referenc…
janimo Mar 13, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
23 changes: 21 additions & 2 deletions catgrad-llm/scripts/compare/compare.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -27,7 +27,26 @@ OUTPUT_DIR="$DIR/outputs"
mkdir -p "$OUTPUT_DIR"
rm -rf "$OUTPUT_DIR"/*

echo "Generating outputs for ${#MODELS[@]} models..."
MAXLEN="${CATGRAD_COMPARE_MAXLEN:-40}"

echo "Generating outputs of ${MAXLEN} tokens for ${#MODELS[@]} models..."

if [[ "${CATGRAD_COMPARE_HF_RUN:-}" ]]; then
REFERENCE_DIR=$DIR/expected/$MAXLEN
mkdir -p $REFERENCE_DIR

for model in "${MODELS[@]}"; do
# Replace slashes with dashes for the filename
filename="${model//\//-}"

echo "Running HF Transformers for $model -> $REFERENCE_DIR/$filename"

uv run catgrad-llm/scripts/llm.py -m "$model" -p 'Category theory is' -s $MAXLEN -r > "$REFERENCE_DIR/$filename" 2>/dev/null &
done

wait
fi


for model in "${MODELS[@]}"; do
# Replace slashes with dashes for the filename
Expand All @@ -37,7 +56,7 @@ for model in "${MODELS[@]}"; do

TYPECHECK="-t"

./target/release/examples/llama -m "$model" -p 'Category theory is' -s 40 --raw -k $TYPECHECK > "$OUTPUT_DIR/$filename" 2>/dev/null &
./target/release/examples/llama -m "$model" -p 'Category theory is' -s $MAXLEN --raw -k $TYPECHECK > "$OUTPUT_DIR/$filename" 2>/dev/null &

[[ -z "${GITHUB_ACTIONS:-}" ]] || wait
done
Expand Down
6 changes: 3 additions & 3 deletions catgrad-llm/scripts/llm.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -23,7 +23,7 @@
parser.add_argument("--revision", type=str, default="main")
parser.add_argument("-p", "--prompt", type=str, default="Category theory is")
parser.add_argument("-s", "--seq-len", type=int, default=10)
parser.add_argument("-r", "--raw-prompt", action="store_true")
parser.add_argument("-r", "--raw", action="store_true")
parser.add_argument("-t", "--thinking", action="store_true")
parser.add_argument("--cache", dest="use_cache", action=argparse.BooleanOptionalAction, default=True)
parser.add_argument("-d", "--dtype", type=str, default="float32")
Expand All @@ -42,7 +42,7 @@
print(f"Loaded model {args.model}, dtype:{model.dtype} on device {model.device}", file=sys.stderr)
prompt = args.prompt

if not args.raw_prompt and tokenizer.chat_template is not None:
if not args.raw and tokenizer.chat_template is not None:
chat = [
{"role": "user", "content": prompt},
]
Expand All @@ -60,6 +60,6 @@

inputs = tokenizer(prompt, return_token_type_ids=False, return_tensors="pt")
logits = model.generate(**inputs, max_new_tokens=args.seq_len, do_sample=False, use_cache=args.use_cache)
output = tokenizer.decode(logits[0])
output = tokenizer.decode(logits[0], skip_special_tokens=True)

print(output)
Loading