[Models]support GLM4.7 Flash

[chang-wenbin]

Motivation

support GLM4.7 Flash

💡 If this PR is a Cherry Pick, the PR title needs to follow the format by adding the [Cherry-Pick] label at the very beginning and appending the original PR ID at the end. For example, [Cherry-Pick][CI] Add check trigger and logic(#5191)

💡 如若此PR是Cherry Pick，PR标题需遵循格式，在最开始加上[Cherry-Pick]标签，以及最后面加上原PR ID，例如[Cherry-Pick][CI] Add check trigger and logic(#5191)

Modifications

support GLM4.7 Flash

Usage or Command

Accuracy Tests

Checklist

• Add at least a tag in the PR title.
  • Tag list: [[FDConfig],[APIServer],[Engine], [Scheduler], [PD Disaggregation], [Executor], [Graph Optimization], [Speculative Decoding], [RL], [Models], [Quantization], [Loader], [OP], [KVCache], [DataProcessor], [BugFix], [Docs], [CI], [Optimization], [Feature], [Benchmark], [Others], [XPU], [HPU], [GCU], [DCU], [Iluvatar], [Metax]]
  • You can add new tags based on the PR content, but the semantics must be clear.
• Format your code, run pre-commit before commit.
• Add unit tests. Please write the reason in this PR if no unit tests.
• Provide accuracy results.
• If the current PR is submitting to the release branch, make sure the PR has been submitted to the develop branch, then cherry-pick it to the release branch with the [Cherry-Pick] PR tag.

[paddle-bot]

Thanks for your contribution!

[codecov-commenter]

Codecov Report

❌ Patch coverage is 44.44444% with 20 lines in your changes missing coverage. Please review.
⚠️ Please upload report for BASE (develop@3651113). Learn more about missing BASE report.

Files with missing lines	Patch %	Lines
...executor/layers/attention/mla_attention_backend.py	11.76%	11 Missing and 4 partials ⚠️
fastdeploy/model_executor/models/deepseek_v3.py	72.22%	4 Missing and 1 partial ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             develop    #7139   +/-   ##
==========================================
  Coverage           ?   73.70%           
==========================================
  Files              ?      376           
  Lines              ?    52876           
  Branches           ?     8244           
==========================================
  Hits               ?    38973           
  Misses             ?    11193           
  Partials           ?     2710           

Flag	Coverage Δ
GPU	73.70% <44.44%> (?)

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[Copilot]

Pull request overview

该 PR 旨在为新增/兼容的模型形态（GLM4.7 Flash 相关）打通 DeepSeekV3/MLA 路径：将 position_ids 与 encoder mask 统一收敛到 ForwardMeta，并在 MLA backend 中增加对 head 数不足 64 的 padding 处理，同时补充新的架构注册类。

Changes:

• 将 DeepSeekV3 的 position_ids / mask_encoder_batch 从显式参数改为通过 ForwardMeta 传递，减少调用链参数数量。
• 在 MLAAttentionBackend decode 路径对 num_heads < 64（TP>1）时对 Q/out 做 padding/裁剪以适配内核约束。
• 新增 Glm4MoeLiteForCausalLM 架构注册，复用 DeepseekV3 实现。

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 5 comments.

File	Description
fastdeploy/model_executor/models/deepseek_v3.py	通过 ForwardMeta 传递 position_ids/mask，并注册 Glm4MoeLite 架构
fastdeploy/model_executor/layers/attention/mla_attention_backend.py	MLA decode 增加 head padding 支持，并调整 rope_scaling 判断逻辑
fastdeploy/model_executor/forward_meta.py	为 MLA/DSA 增加 position_ids 与 mask_encoder_batch 字段

Comments suppressed due to low confidence (1)

fastdeploy/model_executor/layers/attention/mla_attention_backend.py:295

• 这里同样使用了 getattr(self.rope_scaling, "factor", None) 来判断 rope scaling，但 rope_scaling 在配置中是 dict，会导致缩放逻辑永远不生效；同时分支内仍然读取 fd_config.model_config.rope_scaling，与上面缓存到 self.rope_scaling 不一致。建议：改为 dict key 判断（如 self.rope_scaling.get("factor")）并在分支内统一使用 self.rope_scaling 取值。

        self.rope_scaling = getattr(fd_config.model_config, "rope_scaling", None)
        if self.rope_scaling and getattr(self.rope_scaling, "factor", None):
            # if fd_config.model_config.rope_scaling:
            mscale_all_dim = fd_config.model_config.rope_scaling.get("mscale_all_dim", False)  # 1.0
            scaling_factor = fd_config.model_config.rope_scaling["factor"]  # 40
            mscale = yarn_get_mscale(scaling_factor, float(mscale_all_dim))
            self.attn_softmax_scale = self.attn_softmax_scale * mscale * mscale

[EmmonsCurse]

LGTM～ Skip coverage check as it mainly relies on tests with flashmla.