From 1adca39cd8560cfddd6a6875d444123bf064e53e Mon Sep 17 00:00:00 2001 From: ViperEkura <3081035982@qq.com> Date: Sat, 4 Jul 2026 08:35:45 +0800 Subject: [PATCH] fix: handle long sequences and optimize IFD computation --- .gitignore | 6 ++++-- scripts/eval/evaluate_ifd.py | 8 ++++++-- 2 files changed, 10 insertions(+), 4 deletions(-) diff --git a/.gitignore b/.gitignore index 9224a8c..bfa95b4 100644 --- a/.gitignore +++ b/.gitignore @@ -5,8 +5,10 @@ !*/ # Allow specific file types and root files -!*.py -!*.sh +!astrai/**/*.py +!scripts/**/*.py +!scripts/**/*.sh +!tests/**/*.py # Allow GitHub files !/.github/** diff --git a/scripts/eval/evaluate_ifd.py b/scripts/eval/evaluate_ifd.py index 433048e..5bb5843 100644 --- a/scripts/eval/evaluate_ifd.py +++ b/scripts/eval/evaluate_ifd.py @@ -89,8 +89,12 @@ def _compute_ifd_raw(model, tokenizer, instruction, response, device, max_len) - qa_ids = instr_ids + resp_ids with torch.inference_mode(): - logits_qa = model(torch.tensor([qa_ids], device=device, dtype=torch.long))["logits"][0] - logits_resp = model(torch.tensor([resp_ids], device=device, dtype=torch.long))["logits"][0] + logits_qa = model(torch.tensor([qa_ids], device=device, dtype=torch.long))[ + "logits" + ][0] + logits_resp = model(torch.tensor([resp_ids], device=device, dtype=torch.long))[ + "logits" + ][0] resp_logits = logits_qa[instr_len - 1 : -1] resp_targets = logits_resp.new_tensor(resp_ids, dtype=torch.long)