[ROCm][CI] Mirror the three TurboQuant evaluation groups on MI355 (#58282)

Signed-off-by: Andreas Karatzas <akaratza@amd.com>
Co-authored-by: OpenAI Codex <noreply@openai.com>
This commit is contained in:
Andreas Karatzas
2026-09-23 02:37:23 -05:00
committed by GitHub
co-authored by OpenAI Codex
parent 9127170b88
commit a9f07d0bc5
8 changed files with 183 additions and 0 deletions
@@ -0,0 +1,6 @@
model_name: "Qwen/Qwen3-4B"
accuracy_threshold: 0.78
num_questions: 1319
num_fewshot: 5
# Triton serves the unquantized boundary layers with TurboQuant-compatible layout.
server_args: "--kv-cache-dtype turboquant_k3v4_nc --max-model-len 4096 --attention-backend TRITON_ATTN"
@@ -0,0 +1,6 @@
model_name: "Qwen/Qwen3-4B"
accuracy_threshold: 0.80
num_questions: 1319
num_fewshot: 5
# Triton serves the unquantized boundary layers with TurboQuant-compatible layout.
server_args: "--kv-cache-dtype turboquant_k8v4 --max-model-len 4096 --attention-backend TRITON_ATTN"
@@ -0,0 +1,6 @@
model_name: "Qwen/Qwen3-4B"
accuracy_threshold: 0.80
num_questions: 1319
num_fewshot: 5
# Triton serves the unquantized boundary layers with TurboQuant-compatible layout.
server_args: "--kv-cache-dtype turboquant_4bit_nc --max-model-len 4096 --attention-backend TRITON_ATTN"
@@ -0,0 +1 @@
Qwen3-4B-TQ-k3v4nc-ROCm.yaml
@@ -0,0 +1 @@
Qwen3-4B-TQ-k8v4-ROCm.yaml
@@ -0,0 +1 @@
Qwen3-4B-TQ-t4nc-ROCm.yaml