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 7, 00:15 UTC
Fix LFM2 MoE test for Transformers v5 (#36534)
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
Harry Mellor committed
Mar 10, 2026 at 05:29 UTC
195c9972037034355c5e85207f611aa09023cb66
1 file changed
+7
-7
tests/models/registry.py
+7
-7
@@ -351,7 +351,11 @@ _TEXT_GENERATION_EXAMPLE_MODELS = {
351
),
352
"Lfm2ForCausalLM": _HfExamplesInfo("LiquidAI/LFM2-1.2B"),
353
"Lfm2MoeForCausalLM": _HfExamplesInfo(
354
- "LiquidAI/LFM2-8B-A1B", min_transformers_version="4.58"
354
+ "LiquidAI/LFM2-8B-A1B",
355
+ min_transformers_version="5.0.0",
356
+ use_original_num_layers=True,
357
+ # Initialize at least one MoE layer
358
+ hf_overrides={"num_hidden_layers": 4},
359
),
360
"LlamaForCausalLM": _HfExamplesInfo(
361
"meta-llama/Llama-3.2-1B-Instruct",
@@ -511,9 +515,7 @@ _TEXT_GENERATION_EXAMPLE_MODELS = {
515
"stepfun-ai/Step-3.5-Flash",
516
use_original_num_layers=True,
517
# Initialize at least one MoE layer
514
- hf_overrides={
515
- "num_hidden_layers": 4,
516
- },
518
+ hf_overrides={"num_hidden_layers": 4},
519
),
520
"Step3TextForCausalLM": _HfExamplesInfo("stepfun-ai/step3", trust_remote_code=True),
521
"SolarForCausalLM": _HfExamplesInfo(
@@ -1233,9 +1235,7 @@ _SPECULATIVE_DECODING_EXAMPLE_MODELS = {
1235
speculative_model="stepfun-ai/Step-3.5-Flash",
1236
use_original_num_layers=True,
1237
# Initialize at least one MoE layer
1236
- hf_overrides={
1237
- "num_hidden_layers": 4,
1238
- },
1238
+ hf_overrides={"num_hidden_layers": 4},
1239
is_available_online=False,
1240
),
1241
}