throwaway: conc-64 gsm8k eval for DEP8+MTP3 to reproduce dispatch token corruption

Narrow dsr1-fp4-mi355x-sglang-disagg-8k1k-mtp search-space to a single
DEP8+MTP3 conc-64 entry. With max(CONC_LIST)=64, the server computes
SGLANG_MORI_NUM_MAX_DISPATCH_TOKENS_PER_RANK=32, which is below the 256
threshold that selects the correct All2All kernel. Expected: ~0% gsm8k
(silent corruption from the low-latency All2All variant).

Not for merge — throwaway validation of the dispatch token bug.

Author: Oseltamivir

.github/configs/amd-master.yaml

@@ -1986,123 +1986,10 @@ dsr1-fp4-mi355x-sglang-disagg-8k1k-mtp:
     - isl: 8192
       osl: 1024
       search-space:
-      # MTP configurations
-      # 1P1D pure TP8
-      - spec-decoding: "mtp"
-        conc-list: [ 1, 2, 4, 8 ]
-        prefill:
-          num-worker: 1
-          tp: 8
-          ep: 1
-          dp-attn: false
-          additional-settings:
-          - "PREFILL_NODES=1"
-        decode:
-          num-worker: 1
-          tp: 8
-          ep: 1
-          dp-attn: false
-          additional-settings:
-          - "DECODE_NODES=1"
-          - "DECODE_MTP_SIZE=3"
-
-      # 1P2D TP8
-      - spec-decoding: "mtp"
-        conc-list: [ 2, 4, 8, 16, 32 ]
-        prefill:
-          num-worker: 1
-          tp: 8
-          ep: 1
-          dp-attn: false
-          additional-settings:
-          - "PREFILL_NODES=1"
-        decode:
-          num-worker: 2
-          tp: 8
-          ep: 1
-          dp-attn: false
-          additional-settings:
-          - "DECODE_NODES=2"
-          - "DECODE_MTP_SIZE=3"
-
-      # 1P2D TP8
-      - spec-decoding: "mtp"
-        conc-list: [ 32, 64 ]
-        prefill:
-          num-worker: 1
-          tp: 8
-          ep: 1
-          dp-attn: false
-          additional-settings:
-          - "PREFILL_NODES=1"
-        decode:
-          num-worker: 2
-          tp: 8
-          ep: 1
-          dp-attn: false
-          additional-settings:
-          - "DECODE_NODES=2"
-          - "DECODE_MTP_SIZE=3"
-
-      # 1*DEP8 + 1*DEP8
-      - spec-decoding: "mtp"
-        conc-list: [ 640, 512 ]
-        prefill:
-          num-worker: 1
-          tp: 8
-          ep: 8
-          dp-attn: true
-          additional-settings:
-          - "PREFILL_NODES=1"
-        decode:
-          num-worker: 1
-          tp: 8
-          ep: 8
-          dp-attn: true
-          additional-settings:
-          - "DECODE_NODES=1"
-          - "DECODE_MTP_SIZE=3"
-
-      # 1*DEP8 + 1*DEP8
-      - spec-decoding: "mtp"
-        conc-list: [ 256 ]
-        prefill:
-          num-worker: 1
-          tp: 8
-          ep: 8
-          dp-attn: true
-          additional-settings:
-          - "PREFILL_NODES=1"
-        decode:
-          num-worker: 1
-          tp: 8
-          ep: 8
-          dp-attn: true
-          additional-settings:
-          - "DECODE_NODES=1"
-          - "DECODE_MTP_SIZE=3"
-
-
-      # 1*DEP8 + 1*DEP8
-      - spec-decoding: "mtp"
-        conc-list: [ 128 ]
-        prefill:
-          num-worker: 1
-          tp: 8
-          ep: 8
-          dp-attn: true
-          additional-settings:
-          - "PREFILL_NODES=1"
-        decode:
-          num-worker: 1
-          tp: 8
-          ep: 8
-          dp-attn: true
-          additional-settings:
-          - "DECODE_NODES=1"
-          - "DECODE_MTP_SIZE=3"
-
-      # 1*DEP8 + 1*DEP8
+      # THROWAWAY (not for merge): conc-64 only DEP8+MTP3 to reproduce
+      # SGLANG_MORI_NUM_MAX_DISPATCH_TOKENS_PER_RANK < 256 corruption.
+      # max(CONC_LIST)=64 → dispatch_tokens=64/8*4=32 → broken All2All kernel.
+      # 1*DEP8 + 1*DEP8, MTP3
       - spec-decoding: "mtp"
         conc-list: [ 64 ]
         prefill:
@@ -2121,25 +2008,6 @@ dsr1-fp4-mi355x-sglang-disagg-8k1k-mtp:
           - "DECODE_NODES=1"
           - "DECODE_MTP_SIZE=3"
 
-      # 2*DEP8 + 1*DEP8
-      - spec-decoding: "mtp"
-        conc-list: [ 1024, 2048, 4096 ]
-        prefill:
-          num-worker: 2
-          tp: 8
-          ep: 8
-          dp-attn: true
-          additional-settings:
-          - "PREFILL_NODES=2"
-        decode:
-          num-worker: 1
-          tp: 8
-          ep: 8
-          dp-attn: true
-          additional-settings:
-          - "DECODE_NODES=1"
-          - "DECODE_MTP_SIZE=1"
-
 
 # DSv4-Pro FP4 on MI355X via SGLang. Uses a rocm720 mi35x image built off the
 # amd/deepseek_v4 branch in sgl-project/sglang; the SHA is encoded in the

perf-changelog.yaml

@@ -3395,3 +3395,10 @@
   description:
     - "Add DeepSeek-V4-Pro FP4 MI355X ATOM MTP3 benchmark; image rocm/atom:rocm7.2.4_ubuntu24.04_py3.12_pytorch_release_2.10.0_atom0.1.3"
   pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/1627
+
+- config-keys:
+    - dsr1-fp4-mi355x-sglang-disagg-8k1k-mtp
+  description:
+    - "Throwaway: conc-64-only gsm8k eval for DEP8+MTP3 to reproduce SGLANG_MORI_NUM_MAX_DISPATCH_TOKENS_PER_RANK < 256 corruption (dispatch=32 triggers broken All2All kernel, expect ~0% gsm8k). Not for merge."
+  pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/PENDING
+  evals-only: true