Skip to content

[atom-vllm][DP/EP] enable DP/EP for atom-vllm#533

Draft
zejunchen-zejun wants to merge 24 commits intomainfrom
zejun/enable_dp_ep_for_atom_vllm_0409
Draft

[atom-vllm][DP/EP] enable DP/EP for atom-vllm#533
zejunchen-zejun wants to merge 24 commits intomainfrom
zejun/enable_dp_ep_for_atom_vllm_0409

Conversation

@zejunchen-zejun
Copy link
Copy Markdown
Collaborator

@zejunchen-zejun zejunchen-zejun commented Apr 9, 2026

Enable DP and EP feature for below models:
For mori memory model, use MORI_SHMEM_MODE=ISOLATION to set the allocation behavior

  • DeepSeek-FP8 DP8+EP8
Tasks Version Filter n-shot Metric Value Stderr
gsm8k 3 flexible-extract 3 exact_match 0.9409 ± 0.0065
strict-match 3 exact_match 0.9348 ± 0.0068
  • DeepSeek-FP8 TP8+EP8
Tasks Version Filter n-shot Metric Value Stderr
gsm8k 3 flexible-extract 3 exact_match 0.9424 ± 0.0064
strict-match 3 exact_match 0.9371 ± 0.0067
  • GPTOSS DP8
Tasks Version Filter n-shot Metric Value Stderr
gsm8k 3 flexible-extract 3 exact_match 0.4352 ± 0.0137
strict-match 3 exact_match 0.2441 ± 0.0118
  • GPTOSS DP8+EP8
Tasks Version Filter n-shot Metric Value Stderr
gsm8k 3 flexible-extract 3 exact_match 0.4056 ± 0.0135
strict-match 3 exact_match 0.2335 ± 0.0117
  • DeepSeek-MXFP4 DP8+EP8

  • Kimi-K2 DP8+EP8

  • Qwen3.5 FP8 DP8+EP8

Tasks Version Filter n-shot Metric Value Stderr
gsm8k 3 flexible-extract 3 exact_match 0.8635 ± 0.0095
strict-match 3 exact_match 0.8567 ± 0.0097

@zejunchen-zejun zejunchen-zejun force-pushed the zejun/enable_dp_ep_for_atom_vllm_0409 branch from f5239f0 to 1b0d687 Compare April 11, 2026 06:02
@zejunchen-zejun zejunchen-zejun changed the title [atom-vllm][DP/EP] enable DP/EP for atom-vllm path [atom-vllm][DP/EP] enable DP/EP for atom-vllm Apr 15, 2026
Signed-off-by: zejunchen-zejun <zejun.chen@amd.com>
Signed-off-by: zejunchen-zejun <zejun.chen@amd.com>
Signed-off-by: zejunchen-zejun <zejun.chen@amd.com>
Signed-off-by: zejunchen-zejun <zejun.chen@amd.com>
Signed-off-by: zejunchen-zejun <zejun.chen@amd.com>
Signed-off-by: zejunchen-zejun <zejun.chen@amd.com>
Signed-off-by: zejunchen-zejun <zejun.chen@amd.com>
Signed-off-by: zejunchen-zejun <zejun.chen@amd.com>
Signed-off-by: zejunchen-zejun <zejun.chen@amd.com>
Signed-off-by: zejunchen-zejun <zejun.chen@amd.com>
Signed-off-by: zejunchen-zejun <zejun.chen@amd.com>
Signed-off-by: zejunchen-zejun <zejun.chen@amd.com>
Signed-off-by: zejunchen-zejun <zejun.chen@amd.com>
Signed-off-by: zejunchen-zejun <zejun.chen@amd.com>
Signed-off-by: zejunchen-zejun <zejun.chen@amd.com>
Signed-off-by: zejunchen-zejun <zejun.chen@amd.com>
Signed-off-by: zejunchen-zejun <zejun.chen@amd.com>
Signed-off-by: zejunchen-zejun <zejun.chen@amd.com>
Signed-off-by: zejunchen-zejun <zejun.chen@amd.com>
Signed-off-by: zejunchen-zejun <zejun.chen@amd.com>
@zejunchen-zejun zejunchen-zejun force-pushed the zejun/enable_dp_ep_for_atom_vllm_0409 branch from 43ed162 to 3f975ca Compare April 17, 2026 03:51
Signed-off-by: zejunchen-zejun <zejun.chen@amd.com>
Signed-off-by: zejunchen-zejun <zejun.chen@amd.com>
Signed-off-by: zejunchen-zejun <zejun.chen@amd.com>
Signed-off-by: zejunchen-zejun <zejun.chen@amd.com>
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