[dryrun-only] d10 batch64 (OOM test)

Dryrun Job Failed

UUID:295deda6-cff5-4666-bd7d-067b87c8630b

Tao Linvia🐎 pentiumCreated 212 days agoUpdated 212 days ago

DRYRUN FAILED — OOM

batch64 (grad_accum=4) OOMs on A100 40GB.

torch.OutOfMemoryError: CUDA out of memory.
Tried to allocate 8.00 GiB. GPU has 39.49 GiB of which 697.44 MiB is free.

Conclusion: batch32 is the maximum for d10 on A100 40GB. No intermediate option (batch48 fails assert: non-integer grad_accum).

Funding

Funded Job

Done

Fundings at snapshot$0.00

Individual funding and transaction records are not included in the public archive.

Repository: https://github.com/mlpatron/mlpatron-nanochat
Version: experiment/baseline-search → 85b8c8c2d15af9a4165cf2a159a87f94c3bd434f
Image: ghcr.io/nblintao/mlpatron-nanochat:latest
→ ghcr.io/nblintao/mlpatron-nanochat@sha256:28c5544c206237e407241b0a72a8707e63e52750f487b751675b51ff1bce1520
Entrypoint: a100_preset
Resources: NVIDIA A100 40GB × 1·a2-highgpu-1g (12 vCPU / 85 GiB)
Parameters: depth=10·eval_every=300·max_seq_len=2048·tok_max_chars=1000000000·num_iterations=1500(dryrun: 20)·num_data_shards=14·device_batch_size=64