Skip to content

Add AutoTP=2 smoke test config and script#1004

Open
delock wants to merge 1 commit into
masterfrom
gma/opsd-autotp
Open

Add AutoTP=2 smoke test config and script#1004
delock wants to merge 1 commit into
masterfrom
gma/opsd-autotp

Conversation

@delock

@delock delock commented Jul 21, 2026

Copy link
Copy Markdown
Contributor

Verified on 2x RTX 5090 with Qwen2.5-0.5B-Instruct:

  • Student and teacher both use AutoTP=2
  • HybridEngineRollout.generate() produces identical output on both ranks
  • OPSD training loop runs end-to-end with temperature=0 (greedy)
  • No DistributedSampler in pure-TP mode (all ranks see same data)
  • Teacher uses ZeRO-0 when AutoTP is enabled (AutoTP+ZeRO-3 not supported)

Requires DeepSpeed >= commit 53a2ac4 (AutoTP KV cache consistency fix).

Verified on 2x RTX 5090 with Qwen2.5-0.5B-Instruct:
- Student and teacher both use AutoTP=2
- HybridEngineRollout.generate() produces identical output on both ranks
- OPSD training loop runs end-to-end with temperature=0 (greedy)
- No DistributedSampler in pure-TP mode (all ranks see same data)
- Teacher uses ZeRO-0 when AutoTP is enabled (AutoTP+ZeRO-3 not supported)

Requires DeepSpeed >= commit 53a2ac4 (AutoTP KV cache consistency fix).

Signed-off-by: Guokai Ma <guokai.ma@intel.com>
Signed-off-by: Guokai Ma <guokai.ma@gmail.com>
@delock
delock requested a review from tjruwase as a code owner July 21, 2026 14:42
@PKUWZP
PKUWZP self-requested a review July 21, 2026 14:43
@delock

delock commented Jul 22, 2026

Copy link
Copy Markdown
Contributor Author

Note DeepSpeed AutoTP for now does not have column parallel in lm_head, we expect deepspeedai/DeepSpeed#8146 and its follow up (inplace softmax) would significantly reduce memory overhead in case of long sequence OPSD.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant