allenai / allenai/OLMo-core

Missing all OLMo-3-1025-7B Stage 1 (pretraining) Olmo-core checkpoints

Aberta
#729 2 comentários 1 reação 0 responsáveis Ver no GitHub
Linguagem predominante
Python
Estrelas
1.5k
Forks
315
Merge médio
1d 9h
PRs com merge (30d)
11

Descrição

All 1415 stage 1 rows in src/scripts/official/OLMo3/OLMo-3-1025-7B.csv
(https://github.com/allenai/OLMo-core/blob/main/src/scripts/official/OLMo3/OLMo-3-1025-7B.csv) 404 while while every stage-2 (48) and stage-3 (12) checkpoint resolves fine. Tested with the script below

```bash
#!/usr/bin/env bash
# Shows that the stage1 (pretraining) Olmo-core checkpoints listed in
# OLMo-3-1025-7B.csv return 404, while stage2/stage3 return 200.
# Samples a few checkpoint dirs per stage and HEADs config.json in each.
CSV="https://github.com/allenai/OLMo-core/raw/refs/heads/main/src/scripts/official/OLMo3/OLMo-3-1025-7B.csv"
rows=$(curl -fsSL "$CSV")

for stage in stage1 stage2 stage3; do
echo "== $stage =="
# pick 6 dirs evenly spaced across this stage (endpoints included)
echo "$rows" | grep -oE "https://[^,]+/$stage/step[0-9]+/" \
| awk '{a[NR]=$0} END{n=NR; k=(n<6?n:6); for(i=0;i

Guia de contribuição

Abrir o guia de contribuição

Direção de pesquisa

Start with src/scripts/official/OLMo3/OLMo-3-1025-7B.csv and run the curl-based reproduction from the issue to compare stage1, stage2, and stage3 checkpoint URLs. Trace why the 1,415 stage1 entries return 404 while the other stages resolve. Done means the stage1 checkpoint URLs in the CSV resolve consistently with the stage2 and stage3 entries.

Escrita pelo modelo de indexação a partir do texto da issue.

Avaliação

Stack de tecnologia
python
Domínio
machine-learning
Tipo de issue
Bug
Dificuldade
3/5
Tempo estimado
1-2 dias
Status de atividade
Ativa
Clareza
Razoavelmente clara
Facilidade para iniciantes
66/100

Receba novas issues na sua caixa de entrada

Um resumo curto de issues do GitHub para quem está começando.