Skip to content

feat: support qwen3-omni talker and code2wav.#1070

Open
ethan686 wants to merge 6 commits intojd-opensource:mainfrom
ethan686:omni_code2wav
Open

feat: support qwen3-omni talker and code2wav.#1070
ethan686 wants to merge 6 commits intojd-opensource:mainfrom
ethan686:omni_code2wav

Conversation

@ethan686
Copy link
Copy Markdown
Contributor

No description provided.

Copy link
Copy Markdown
Contributor

@gemini-code-assist gemini-code-assist bot left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Code Review

The code review identified a critical issue in NpuQwen3MoeDecoderLayerImpl::forward where the function continues to return tensor_placeholder_ even after a failed decode layer execution, indicated by a non-zero status code. To prevent the propagation of potentially corrupted data, the reviewer suggested that the function should either return early or throw an exception when st != 0.

@@ -0,0 +1,362 @@
#pragma once
Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

add copyright

@@ -0,0 +1,284 @@
/* Copyright 2025 The xLLM Authors. All Rights Reserved.
Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

2026


void Qwen3MoeDecoderLoader::verify_loaded_weights(
const std::string& prefix) const {
VLOG(50) << "start verify";
Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

why add this line?

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Addressed.

at_weight_tensors_[IN_MLP_GATEUP_WEIGHT_EXPERT] =
at_npu::native::npu_format_cast(mlp_gateup_weight, 2).contiguous();
VLOG(50) << "gmm weight gateup shape"
<< at_weight_tensors_[IN_MLP_GATEUP_WEIGHT_EXPERT].sizes();
Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

ditto

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Addressed.


{"mlp_layer_scale.scale", IN_MLP_LAYER_SCALE_SCALE}};

static const std::map<int, int> WEIGHT_SHARD = {{IN_QKV_WEIGHT_Q, 0},
Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

why not use std::unordered_map?

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants