[ROCm][CI] Expand MI355 mirrors and route MIG-sized jobs to DPX (#59137)
Signed-off-by: Andreas Karatzas <andreas.karatzas@protonmail.com>
This commit is contained in:
@@ -0,0 +1,18 @@
|
||||
model_name: "nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-NVFP4"
|
||||
accuracy_threshold: 0.93
|
||||
num_questions: 1319
|
||||
num_fewshot: 5
|
||||
max_concurrency: 16
|
||||
rocm_request_timeout_seconds: 1800
|
||||
min_acceptance_length: 2.0
|
||||
startup_max_wait_seconds: 1200
|
||||
server_args: >-
|
||||
--enforce-eager
|
||||
--max-model-len 4096
|
||||
--max-num-seqs 16
|
||||
--max-num-batched-tokens 32768
|
||||
--gpu-memory-utilization 0.5
|
||||
--tensor-parallel-size 2
|
||||
--enable-expert-parallel
|
||||
--mamba-backend triton
|
||||
--speculative-config '{"method":"mtp","num_speculative_tokens":5}'
|
||||
@@ -0,0 +1,17 @@
|
||||
model_name: "nm-testing/Qwen3-Next-80B-A3B-Instruct-NVFP4"
|
||||
accuracy_threshold: 0.75
|
||||
num_questions: 1319
|
||||
num_fewshot: 5
|
||||
max_concurrency: 16
|
||||
rocm_request_timeout_seconds: 1800
|
||||
server_args: >-
|
||||
--enforce-eager
|
||||
--max-model-len 4096
|
||||
--max-num-seqs 16
|
||||
--max-num-batched-tokens 32768
|
||||
--gpu-memory-utilization 0.5
|
||||
--tensor-parallel-size 2
|
||||
--enable-expert-parallel
|
||||
--all2all-backend allgather_reducescatter
|
||||
--attention-backend TRITON_ATTN
|
||||
--no-async-scheduling
|
||||
@@ -0,0 +1,19 @@
|
||||
model_name: "nvidia/Qwen3-Next-80B-A3B-Instruct-NVFP4"
|
||||
accuracy_threshold: 0.85
|
||||
num_questions: 1319
|
||||
num_fewshot: 5
|
||||
max_concurrency: 16
|
||||
rocm_request_timeout_seconds: 1800
|
||||
min_acceptance_length: 1.5
|
||||
server_args: >-
|
||||
--enforce-eager
|
||||
--max-model-len 4096
|
||||
--max-num-seqs 16
|
||||
--max-num-batched-tokens 32768
|
||||
--gpu-memory-utilization 0.5
|
||||
--tensor-parallel-size 2
|
||||
--enable-expert-parallel
|
||||
--all2all-backend allgather_reducescatter
|
||||
--attention-backend TRITON_ATTN
|
||||
--no-async-scheduling
|
||||
--speculative-config '{"method":"qwen3_next_mtp","num_speculative_tokens":1}'
|
||||
@@ -0,0 +1,5 @@
|
||||
# Exact B200 compressed-tensors checkpoint; it does not contain MTP weights.
|
||||
Qwen3-Next-80B-A3B-NVFP4-EP2-CT-MI355.yaml
|
||||
# MTP-capable ModelOpt checkpoint exercises NVFP4 EP2 with MTP1.
|
||||
Qwen3-Next-80B-A3B-NVFP4-EP2-MTP-MI355.yaml
|
||||
Nemotron-3-Super-120B-A12B-NVFP4-MI355.yaml
|
||||
@@ -1,2 +1,5 @@
|
||||
Qwen3-0.6B-FP8.yaml
|
||||
Qwen3-1.7B-MXFP4.yaml
|
||||
Qwen3-30B-A3B-NVFP4.yaml
|
||||
Laguna-XS.2-NVFP4.yaml
|
||||
Laguna-XS-2.1-INT4.yaml
|
||||
|
||||
@@ -2,3 +2,4 @@ Qwen3.5-35B-A3B-DEP2.yaml
|
||||
Qwen3.5-35B-A3B-FP8-DEP2.yaml
|
||||
Qwen3.5-35B-A3B-MXFP4-AITER-TP2.yaml
|
||||
Qwen3.5-35B-A3B-MXFP4-EMU-TP2.yaml
|
||||
Qwen3.5-397B-A17B-NVFP4-DEP2.yaml
|
||||
|
||||
Reference in New Issue
Block a user