models: BEVFusion onboarding (stage graph, sparse plugin export, INT8/FP8) - #22
Draft
vividf wants to merge 1 commit into
Draft
models: BEVFusion onboarding (stage graph, sparse plugin export, INT8/FP8)#22vividf wants to merge 1 commit into
vividf wants to merge 1 commit into
Conversation
This was referenced Sep 5, 2026
…/FP8) Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
vividf
force-pushed
the
pr12-bevfusion-onboarding
branch
from
September 7, 2026 05:23
eacb904 to
8420501
Compare
vividf
force-pushed
the
pr11-spconv-trt-plugins
branch
from
September 7, 2026 05:23
e50ff65 to
7800a6f
Compare
vividf
changed the base branch from
pr11-spconv-trt-plugins
to
pr11b-bevfusion-export-fusion
September 7, 2026 05:23
This was referenced Sep 7, 2026
Draft
Draft
quant: declarative quantization (plan/placement, modelopt engine, PTQ/QAT, self-describing ckpt)
#17
Draft
Draft
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Stack(base
pr-stack-base=7e35d97):#11 → #12 → #13 → #14 → #15 → #16 → #17 → #18 → #19 → #20(CenterPoint 全功能點)→ #21 → #25 → #22 → #23 → #24(= 開發 branch HEAD,逐位元一致)。合併由下而上,每合一層把下一層 base retarget。這個 PR 是什麼
BEVFusion onboarding:stage graph(dense 圖 + sparse plugin 圖)、量化宣告、experiment configs。與 #20 同 pattern 的第二次應用,增量 = sparse stage。
實測數字
INT8 的兩個敏感 block 保 FP16(sensitivity 掃描記錄在 config 註解);FP8 是「linear 走 FP8 不走 INT8」證據鏈之一(FFN FP8 ±0)。
逐檔導覽
main_modules/bevfusion/stages.py:sparse GraphStage(plugin 圖:onnx_transforms=(#21 的 fusion pass,)、ORT 用torch_fallback_backends)+ dense GraphStage;sparse-stage 的 INT8 TODO(需要量化 libspconv ABI,阻擋條件寫明)。main_modules/bevfusion/quantization.py:量化宣告 + 「attention 投影在校準期抓不到」陷阱的正本 docstring(訓練態 packed qkv Parameter、export 態才誕生的 q/k/v/out_proj、out_proj 的 fast-path 繞過——walker 在框架層拒換它)。main_modules/bevfusion/model.py:stage-graph 版模型類。_int8/_int8_qat/_fp8experiments(_fp8關 onnx backend——ORT 載不了 trt-domain FP8 op;tolerance 帶 observed)。test_bevfusion_stages。檔案清單(點開)
Review checklist
_fp8config 關 onnx backend 的註解是否講清楚原因。🤖 Generated with Claude Code