@tduncan / Nvidia-Optimization-Stack-VLLM Mirror · read-only / commits / f4fd390f5d
Fork 0

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 16, 19:30 UTC

[Bugfix] Lower gemma's unloaded_params exception to warning (#7002)

Michael Goin committed Aug 1, 2024 at 15:01 UTC f4fd390f5de585fd94877158bea4e1b2d1920df3
3 files changed +12 -9
vllm/model_executor/models/gemma.py
+3 -3
@@ -404,6 +404,6 @@ class GemmaForCausalLM(nn.Module, SupportsLoRA):
404 loaded_params.add(name)
405 unloaded_params = params_dict.keys() - loaded_params
406 if unloaded_params:
407 - raise RuntimeError(
408 - "Some weights are not initialized from checkpoints: "
409 - f"{unloaded_params}")
407 + logger.warning(
408 + "Some weights are not initialized from checkpoints: %s",
409 + unloaded_params)
vllm/model_executor/models/gemma2.py
+6 -3
@@ -23,6 +23,7 @@ from transformers import Gemma2Config
23 from vllm.attention import Attention, AttentionMetadata
24 from vllm.config import CacheConfig, LoRAConfig
25 from vllm.distributed import get_tensor_model_parallel_world_size
26 +from vllm.logger import init_logger
27 from vllm.model_executor.layers.activation import GeluAndMul
28 from vllm.model_executor.layers.layernorm import GemmaRMSNorm
29 from vllm.model_executor.layers.linear import (MergedColumnParallelLinear,
@@ -41,6 +42,8 @@ from vllm.sequence import IntermediateTensors, SamplerOutput
42
43 from .interfaces import SupportsLoRA
44
45 +logger = init_logger(__name__)
46 +
47
48 class Gemma2MLP(nn.Module):
49
@@ -390,6 +393,6 @@ class Gemma2ForCausalLM(nn.Module, SupportsLoRA):
393
394 unloaded_params = params_dict.keys() - loaded_params
395 if unloaded_params:
393 - raise RuntimeError(
394 - "Some weights are not initialized from checkpoints: "
395 - f"{unloaded_params}")
396 + logger.warning(
397 + "Some weights are not initialized from checkpoints: %s",
398 + unloaded_params)
vllm/model_executor/models/paligemma.py
+3 -3
@@ -342,6 +342,6 @@ class PaliGemmaForConditionalGeneration(nn.Module, SupportsVision):
342
343 unloaded_params = params_dict.keys() - loaded_params
344 if unloaded_params:
345 - raise RuntimeError(
346 - "Some weights are not initialized from checkpoints: "
347 - f"{unloaded_params}")
345 + logger.warning(
346 + "Some weights are not initialized from checkpoints: %s",
347 + unloaded_params)