All issues

Search results

Reproducing the training mixture: max_per_file ties the token budget to corpus sharding, not to the data

Status: Open.

#27 In sapientinc/HRM-Text;

· sdi1100041 opened on Aug 12, 2026

dist.broadcast called unconditionally in create_model_and_carry breaks single-GPU runs

Status: Open.

#26 In sapientinc/HRM-Text;

· dolceo opened on Jul 27, 2026

Verify evals on Papers with Code

Status: Open.

#25 In sapientinc/HRM-Text;

· NielsRogge opened on Jul 22, 2026

MMLU-Pro: invalid answers credited as if guessed correctly

Status: Open.

#24 In sapientinc/HRM-Text;

· Yefeng8899 opened on Jul 9, 2026

update_lr divides by zero when lr_warmup_steps == 0 and total_steps == 0

Status: Open.

#23 In sapientinc/HRM-Text;

· danielbennett888 opened on Jul 6, 2026

BoolQ ground-truth label appears inverted

Status: Open.

#22 In sapientinc/HRM-Text;

· HuggingNobody opened on Jul 4, 2026

No <eos> token in the training? Clarification: Cross-Sample Information Flow in PrefixLM Attention (Pass 2)

Status: Open.

#20 In sapientinc/HRM-Text;

· MugeTong opened on Jun 15, 2026

有没有可能在4090显卡上训练它

Status: Open.

#19 In sapientinc/HRM-Text;

· ak01user opened on Jun 14, 2026

Have anyone tried training a larger size of the model, and will we achieve even better perf?

Status: Open.

#18 In sapientinc/HRM-Text;

· Neo9061 opened on Jun 13, 2026

tokenizer.bos_token = cfg["boq"] may fail if token not in vocab

Status: Open.

#17 In sapientinc/HRM-Text;

· hty993 opened on Jun 13, 2026

Feature Request: Release checkpoints for HRM-Text 0.6B

Status: Open.

#16 In sapientinc/HRM-Text;

· DecoderW opened on Jun 9, 2026

Sparse MoE FFN support for HRM-Text and Better Performance

Status: Open.

#15 In sapientinc/HRM-Text;

· XiaoYee opened on Jun 9, 2026