[CI] Clean DeepSeek V4 tests and installation scripts (#26609)

test/registered/cp/test_deepseek_v4_flash_fp4_b200_cp.py

@@ -21,7 +21,7 @@ from sglang.test.test_utils import (
     try_cached_model,
 )
 
-register_cuda_ci(est_time=235, stage="extra-b", runner_config="4-gpu-b200")
+register_cuda_ci(est_time=235, stage="extra-b", runner_config="deepep-4-gpu-b200")
 
 MODEL = "deepseek-ai/DeepSeek-V4-Flash"
 SERVER_LAUNCH_TIMEOUT = 3600

test/registered/disaggregation/test_disaggregation_dsv4.py

@@ -11,7 +11,7 @@ from sglang.test.test_utils import (
     try_cached_model,
 )
 
-register_cuda_ci(est_time=250, stage="base-c", runner_config="dsv4-8-gpu-h200")
+register_cuda_ci(est_time=250, stage="base-c", runner_config="deepep-8-gpu-h200")
 
 DSV4_FLASH_MODEL = "sgl-project/DeepSeek-V4-Flash-FP8"
 

test/registered/models_e2e/test_deepseek_v4_flash_fp4_b200.py

@@ -4,7 +4,7 @@ Launches TP=4 with flashinfer_mxfp4 MoE runner + EAGLE speculative decoding.
 Runs 12 ServerSanity probes (correctness, streaming, concurrency, determinism)
 plus a GSM8K accuracy gate.
 
-Registry: base-c-test-dsv4-4-gpu-b200 (per-commit, 4x B200)
+Registry: base-c-test-deepep-4-gpu-b200 (per-commit, 4x B200)
 """
 
 import unittest
@@ -20,7 +20,7 @@ from sglang.test.test_utils import (
     try_cached_model,
 )
 
-register_cuda_ci(est_time=465, stage="base-c", runner_config="dsv4-4-gpu-b200")
+register_cuda_ci(est_time=465, stage="base-c", runner_config="deepep-4-gpu-b200")
 
 MODEL = "deepseek-ai/DeepSeek-V4-Flash"
 SERVER_LAUNCH_TIMEOUT = 3600

test/registered/models_e2e/test_deepseek_v4_flash_fp4_h200.py

@@ -4,7 +4,7 @@ Launches TP=4 with Marlin FP4 MoE runner + EAGLE speculative decoding.
 Runs 12 ServerSanity probes (correctness, streaming, concurrency, determinism)
 plus a GSM8K accuracy gate.
 
-Registry: base-c-test-dsv4-8-gpu-h200 (per-commit, 8x H200 — only 4 used by TP=4)
+Registry: base-c-test-deepep-8-gpu-h200 (per-commit, 8x H200 — only 4 used by TP=4)
 """
 
 import unittest
@@ -20,7 +20,7 @@ from sglang.test.test_utils import (
     try_cached_model,
 )
 
-register_cuda_ci(est_time=370, stage="base-c", runner_config="dsv4-8-gpu-h200")
+register_cuda_ci(est_time=370, stage="base-c", runner_config="deepep-8-gpu-h200")
 
 
 def _flashinfer_has_sm90_cutlass_mxfp4() -> bool:

test/registered/models_e2e/test_deepseek_v4_flash_fp4_megamoe_b200.py

@@ -4,7 +4,7 @@ Launches TP=4 with flashinfer_mxfp4 MoE runner + EAGLE speculative decoding.
 Runs 12 ServerSanity probes (correctness, streaming, concurrency, determinism)
 plus a GSM8K accuracy gate.
 
-Registry: base-c-test-dsv4-4-gpu-b200 (per-commit, 4x B200)
+Registry: extra-b-test-deepep-4-gpu-b200 (label-gated, 4x B200)
 """
 
 import unittest
@@ -20,7 +20,7 @@ from sglang.test.test_utils import (
     try_cached_model,
 )
 
-register_cuda_ci(est_time=900, stage="base-c", runner_config="dsv4-4-gpu-b200")
+register_cuda_ci(est_time=900, stage="extra-b", runner_config="deepep-4-gpu-b200")
 
 MODEL = "deepseek-ai/DeepSeek-V4-Flash"
 SERVER_LAUNCH_TIMEOUT = 3600

test/registered/models_e2e/test_deepseek_v4_flash_fp8_h200.py

@@ -5,7 +5,7 @@ with FP4 experts disabled via SGLANG_DSV4_FP4_EXPERTS=0.
 Runs 12 ServerSanity probes (correctness, streaming, concurrency, determinism)
 plus a GSM8K accuracy gate.
 
-Registry: base-c-test-dsv4-8-gpu-h200 (per-commit, 8x H200 — only 4 used by TP=4)
+Registry: extra-b-test-deepep-8-gpu-h200 (label-gated, 8x H200 — only 4 used by TP=4)
 """
 
 import unittest
@@ -21,7 +21,7 @@ from sglang.test.test_utils import (
     try_cached_model,
 )
 
-register_cuda_ci(est_time=280, stage="base-c", runner_config="dsv4-8-gpu-h200")
+register_cuda_ci(est_time=280, stage="extra-b", runner_config="deepep-8-gpu-h200")
 
 MODEL_FP8 = "sgl-project/DeepSeek-V4-Flash-FP8"
 SERVER_LAUNCH_TIMEOUT = 3600

test/run_suite.py

@@ -59,8 +59,8 @@ PER_COMMIT_SUITES = {
         "base-c-test-8-gpu-h200",
         "base-c-test-8-gpu-b200",
         "base-c-test-deepep-4-gpu-h100",
-        "base-c-test-dsv4-4-gpu-b200",
-        "base-c-test-dsv4-8-gpu-h200",
+        "base-c-test-deepep-4-gpu-b200",
+        "base-c-test-deepep-8-gpu-h200",
         # extra-a / extra-b: label-gated PR opt-in suites in pr-test-extra.yml
         # (tests still tagged per-commit but skipped on default PR runs).
         "extra-a-test-1-gpu-small",
@@ -69,6 +69,7 @@ PER_COMMIT_SUITES = {
         "extra-b-test-4-gpu-h100",
         "extra-b-test-4-gpu-b200",
         "extra-b-test-8-gpu-h200",
+        "extra-b-test-deepep-4-gpu-b200",
         "extra-b-test-deepep-8-gpu-h200",
     ],
     HWBackend.NPU: [