From cd0ff26e7acf2c691a33d4c44276db6980bab24b Mon Sep 17 00:00:00 2001 From: Michael Goin Date: Tue, 19 May 2026 23:21:01 -0400 Subject: [PATCH] [CI] Add DSV4-Flash to gsm8k moe-refactor/config-b200.txt (#42111) Signed-off-by: mgoin --- requirements/common.txt | 1 + requirements/test/cuda.txt | 3 ++- requirements/test/rocm.txt | 2 ++ requirements/test/xpu.txt | 1 + .../moe-refactor/DeepSeek-V4-Flash-deepgemm-mega-moe.yaml | 5 +++++ tests/evals/gsm8k/configs/moe-refactor/config-b200.txt | 1 + 6 files changed, 12 insertions(+), 1 deletion(-) create mode 100644 tests/evals/gsm8k/configs/moe-refactor/DeepSeek-V4-Flash-deepgemm-mega-moe.yaml diff --git a/requirements/common.txt b/requirements/common.txt index 492ae935731..d37ef1f1fed 100644 --- a/requirements/common.txt +++ b/requirements/common.txt @@ -9,6 +9,7 @@ blake3 py-cpuinfo transformers >= 4.56.0, != 5.0.*, != 5.1.*, != 5.2.*, != 5.3.*, != 5.4.*, != 5.5.0 tokenizers >= 0.21.1 # Required for fast incremental detokenization. +safetensors >= 0.6.2 # MXFP4/MXFP6 dtype support (F8_E8M0, F4) added in 0.6.0: https://github.com/huggingface/safetensors/pull/611 protobuf >= 5.29.6, !=6.30.*, !=6.31.*, !=6.32.*, !=6.33.0.*, !=6.33.1.*, !=6.33.2.*, !=6.33.3.*, !=6.33.4.* # Required by LlamaTokenizer, gRPC. CVE-2026-0994 fastapi[standard] >= 0.115.0 # Required by FastAPI's form models in the OpenAI API server's audio transcriptions endpoint. aiohttp >= 3.13.3 diff --git a/requirements/test/cuda.txt b/requirements/test/cuda.txt index 290e539c9be..245a86f93be 100644 --- a/requirements/test/cuda.txt +++ b/requirements/test/cuda.txt @@ -858,8 +858,9 @@ s3transfer==0.10.3 # via boto3 sacrebleu==2.4.3 # via lm-eval -safetensors==0.4.5 +safetensors==0.7.0 # via + # -c requirements/common.txt # accelerate # open-clip-torch # peft diff --git a/requirements/test/rocm.txt b/requirements/test/rocm.txt index 4b801383d89..b15e00edf1d 100644 --- a/requirements/test/rocm.txt +++ b/requirements/test/rocm.txt @@ -1048,6 +1048,8 @@ sacrebleu==2.6.0 # via lm-eval safetensors==0.7.0 # via + # -c requirements/common.txt + # -r requirements/test/../common.txt # accelerate # open-clip-torch # peft diff --git a/requirements/test/xpu.txt b/requirements/test/xpu.txt index 80b0c148116..5581d0a079c 100644 --- a/requirements/test/xpu.txt +++ b/requirements/test/xpu.txt @@ -539,6 +539,7 @@ sacrebleu==2.6.0 # via lm-eval safetensors==0.7.0 # via + # -c requirements/common.txt # accelerate # timm # transformers diff --git a/tests/evals/gsm8k/configs/moe-refactor/DeepSeek-V4-Flash-deepgemm-mega-moe.yaml b/tests/evals/gsm8k/configs/moe-refactor/DeepSeek-V4-Flash-deepgemm-mega-moe.yaml new file mode 100644 index 00000000000..742d9e40b8a --- /dev/null +++ b/tests/evals/gsm8k/configs/moe-refactor/DeepSeek-V4-Flash-deepgemm-mega-moe.yaml @@ -0,0 +1,5 @@ +model_name: "deepseek-ai/DeepSeek-V4-Flash" +accuracy_threshold: 0.95 +num_questions: 1319 +num_fewshot: 5 +server_args: "--trust-remote-code --kv-cache-dtype fp8 --block-size 256 --enable-expert-parallel --tensor-parallel-size 2 --attention_config.use_fp4_indexer_cache=True --moe-backend deep_gemm_mega_moe --tokenizer-mode deepseek_v4 --tool-call-parser deepseek_v4 --enable-auto-tool-choice --reasoning-parser deepseek_v4 --speculative_config.method=mtp --speculative_config.num_speculative_tokens=2" diff --git a/tests/evals/gsm8k/configs/moe-refactor/config-b200.txt b/tests/evals/gsm8k/configs/moe-refactor/config-b200.txt index d8bb5aa28fc..25106cd34f2 100644 --- a/tests/evals/gsm8k/configs/moe-refactor/config-b200.txt +++ b/tests/evals/gsm8k/configs/moe-refactor/config-b200.txt @@ -16,3 +16,4 @@ Mixtral-8x7B-BF16-triton.yaml Nemotron-Nano-30B-Fp8-ModelOpt-fi-trtllm.yaml Nemotron-Nano-30B-NvFp4-ModelOpt-fi-cutlass.yaml Nemotron-Nano-30B-NvFp4-ModelOpt-vllm-cutlass.yaml +DeepSeek-V4-Flash-deepgemm-mega-moe.yaml