A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 11, 15:00 UTC
[Bugfix][Model] Fix Afmoe rope_parameters issue (#30505)
Signed-off-by: mgoin <mgoin64@gmail.com> Signed-off-by: Michael Goin <mgoin64@gmail.com> Co-authored-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
Michael Goin committed
Dec 11, 2025 at 21:53 UTC
9f2fc16a6903f8988515ce2560d3ef0850809c42
2 files changed
+2
-5
tests/models/registry.py
+1
-4
@@ -173,10 +173,7 @@ class _HfExamplesInfo:
173
174
_TEXT_GENERATION_EXAMPLE_MODELS = {
175
# [Decoder-only]
176
- "AfmoeForCausalLM": _HfExamplesInfo(
177
- "arcee-ai/Trinity-Nano",
178
- is_available_online=False,
179
- ),
176
+ "AfmoeForCausalLM": _HfExamplesInfo("arcee-ai/Trinity-Nano-Preview"),
177
"ApertusForCausalLM": _HfExamplesInfo("swiss-ai/Apertus-8B-Instruct-2509"),
178
"AquilaModel": _HfExamplesInfo("BAAI/AquilaChat-7B", trust_remote_code=True),
179
"AquilaForCausalLM": _HfExamplesInfo("BAAI/AquilaChat2-7B", trust_remote_code=True),
vllm/model_executor/models/afmoe.py
+1
-1
@@ -242,7 +242,7 @@ class AfmoeAttention(nn.Module):
242
self.rotary_emb = get_rope(
243
self.head_dim,
244
max_position=max_position_embeddings,
245
- rope_parameters=config["rope_parameters"],
245
+ rope_parameters=config.rope_parameters,
246
is_neox_style=True,
247
)
248
else: