scripts/job_debug_he.sh

#!/bin/bash
# Why is the HumanEval response empty? Dump raw generated ids for ONE example.
set -uo pipefail
pip install -q "transformers==4.46.2" "huggingface_hub<1.0" "datasets<4" "accelerate" 2>&1 | tail -1
python -c "
from huggingface_hub import snapshot_download
snapshot_download('ashishk1331/ccd-repro-code', repo_type='dataset', local_dir='/work')"
cd /work

python - <<'EOF'
import sys, torch
sys.path.insert(0, "scripts")
from transformers import AutoModel, AutoTokenizer
from datasets import load_dataset
import ccd_decode
from run_eval import humaneval_prompt

M = "Dream-org/Dream-v0-Instruct-7B"
tok = AutoTokenizer.from_pretrained(M, trust_remote_code=True)
model = AutoModel.from_pretrained(M, torch_dtype=torch.bfloat16,
                                  trust_remote_code=True).to("cuda").eval()
doc = load_dataset("openai/openai_humaneval", split="test")[0]
p = humaneval_prompt(tok, doc)
enc = tok(p, return_tensors="pt")
print("prompt tokens:", enc.input_ids.shape)
print("prompt tail ids:", enc.input_ids[0, -8:].tolist())
print("prompt tail dec:", repr(tok.decode(enc.input_ids[0, -8:].tolist())))

# short generation so this is quick
x, st = ccd_decode.generate(
    model, enc.input_ids.to("cuda"), attention_mask=enc.attention_mask.to("cuda"),
    max_new_tokens=256, steps=256, temperature=0.1, top_p=0.9,
    mask_token_id=model.config.mask_token_id, method="baseline")
gen = x[0, enc.input_ids.shape[1]:]
ids = gen.tolist()
print("\nfirst 24 generated ids:", ids[:24])
print("unique ids (count):", len(set(ids)))
from collections import Counter
print("most common ids:", Counter(ids).most_common(6))
print("\nRAW decode (no split), first 400 chars:")
print(repr(tok.decode(ids)[:400]))
print("\nDecode skip_special_tokens=True, first 400:")
print(repr(tok.decode(ids, skip_special_tokens=True)[:400]))
print("\neos_token:", repr(tok.eos_token), tok.eos_token_id)
print("split-on-eos result (first 200):", repr(tok.decode(ids).split(tok.eos_token)[0][:200]))
EOF