Skip to content

Add DSv4.1 Flash DSpark golden AL curve / 新增黄金 AL 曲线 #2976

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
Oseltamivir wants to merge 1 commit into from
Closed

Add DSv4.1 Flash DSpark golden AL curve / 新增黄金 AL 曲线 #2976

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
3 changes: 3 additions & 0 deletions golden_al_distribution/README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -127,6 +127,7 @@ Before accepting an updated curve, reviewers should verify:
| --- | --- | --- | --- |
| DeepSeek V4 Pro | MTP | [`dsv4_mtp.yaml`](dsv4_mtp.yaml) | [27180633016](https://github.com/SemiAnalysisAI/InferenceX/actions/runs/27180633016) |
| DeepSeek V4 Pro 0813 | DSpark (probabilistic draft) | [`dsv4-pro-0813-dspark.yaml`](dsv4-pro-0813-dspark.yaml) | [31742838308](https://github.com/SemiAnalysisAI/InferenceX/actions/runs/31742838308) |
| DeepSeek V4.1 Flash | DSpark (probabilistic draft, block verify) | [`dsv41flash_dspark.yaml`](dsv41flash_dspark.yaml) | [34493175056](https://github.com/SemiAnalysisAI/InferenceX/actions/runs/34493175056) |
| Qwen3.5 397B-A17B | MTP | [`qwen3.5_mtp.yaml`](qwen3.5_mtp.yaml) | [27317114007](https://github.com/SemiAnalysisAI/InferenceX/actions/runs/27317114007) |
| Kimi K2.5 | EAGLE3 | [`kimik2.5_eagle3.yaml`](kimik2.5_eagle3.yaml) | [28122195822](https://github.com/SemiAnalysisAI/InferenceX/actions/runs/28122195822) |
| Kimi K3 | DSpark | [`kimik3_dspark.yaml`](kimik3_dspark.yaml) | [30304797750](https://github.com/SemiAnalysisAI/InferenceX/actions/runs/30304797750) |
Expand All @@ -151,3 +152,5 @@ Before accepting an updated curve, reviewers should verify:
- [InferenceX initial AL collector PR](https://github.com/SemiAnalysisAI/InferenceX/pull/1650)
- [InferenceX multi-model AL collectors PR](https://github.com/SemiAnalysisAI/InferenceX/pull/1706)
- [InferenceX multi-node synthetic-acceptance bring-up](https://github.com/SemiAnalysisAI/InferenceX/pull/1789)

DeepSeek V4.1 Flash includes measured draft lengths 1–5 for thinking off/on. The same image rejected lengths 6–8 before serving in [run 34494319147](https://github.com/SemiAnalysisAI/InferenceX/actions/runs/34494319147); no AL values are assigned to those lengths.
3 changes: 3 additions & 0 deletions golden_al_distribution/README_zh.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -127,6 +127,7 @@ gh workflow run speedbench-al.yml \
| --- | --- | --- | --- |
| DeepSeek V4 Pro | MTP | [`dsv4_mtp.yaml`](dsv4_mtp.yaml) | [27180633016](https://github.com/SemiAnalysisAI/InferenceX/actions/runs/27180633016) |
| DeepSeek V4 Pro 0813 | DSpark（概率采样草稿） | [`dsv4-pro-0813-dspark.yaml`](dsv4-pro-0813-dspark.yaml) | [31742838308](https://github.com/SemiAnalysisAI/InferenceX/actions/runs/31742838308) |
| DeepSeek V4.1 Flash | DSpark（概率采样草稿 + 块验证） | [`dsv41flash_dspark.yaml`](dsv41flash_dspark.yaml) | [34493175056](https://github.com/SemiAnalysisAI/InferenceX/actions/runs/34493175056) |
| Qwen3.5 397B-A17B | MTP | [`qwen3.5_mtp.yaml`](qwen3.5_mtp.yaml) | [27317114007](https://github.com/SemiAnalysisAI/InferenceX/actions/runs/27317114007) |
| Kimi K2.5 | EAGLE3 | [`kimik2.5_eagle3.yaml`](kimik2.5_eagle3.yaml) | [28122195822](https://github.com/SemiAnalysisAI/InferenceX/actions/runs/28122195822) |
| Kimi K3 | DSpark | [`kimik3_dspark.yaml`](kimik3_dspark.yaml) | [30304797750](https://github.com/SemiAnalysisAI/InferenceX/actions/runs/30304797750) |
Expand All @@ -151,3 +152,5 @@ gh workflow run speedbench-al.yml \
- [InferenceX 初始 AL 收集器 PR](https://github.com/SemiAnalysisAI/InferenceX/pull/1650)
- [InferenceX 多模型 AL 收集器 PR](https://github.com/SemiAnalysisAI/InferenceX/pull/1706)
- [InferenceX 多节点合成接受验证](https://github.com/SemiAnalysisAI/InferenceX/pull/1789)

DeepSeek V4.1 Flash 包含草稿长度 1–5 在 thinking 开关两种模式下的实测值。同一镜像在[运行 34494319147](https://github.com/SemiAnalysisAI/InferenceX/actions/runs/34494319147) 中于服务启动前拒绝长度 6–8，因此未为这些长度填写 AL 数值。
23 changes: 23 additions & 0 deletions golden_al_distribution/dsv41flash_dspark.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,23 @@
# Source GitHub Actions run: https://github.com/SemiAnalysisAI/InferenceX/actions/runs/34493175056
# Draft lengths 6-8 rejected by the image (no measured AL): https://github.com/SemiAnalysisAI/InferenceX/actions/runs/34494319147
# Acceptance Length (AL) measured with SPEED-Bench Qualitative coding.
# model: deepseek-ai/DeepSeek-V4.1-Flash | image: vllm/vllm-openai:deepseekv41-flash-0909 | TP: 4
# temperature: 1.0 | output_len: 4096
# thinking_on chat_template_kwargs: {"thinking":true,"reasoning_effort":"high"}
# thinking_off chat_template_kwargs: {"thinking":false}
# method: dspark | draft_sample_method: probabilistic | rejection_sample_method: block
# enable_adaptive_verification: false | engram cpu_offload: true
# key = num_speculative_tokens; AL includes the target verification token.
deepseek-v4.1-flash:
thinking_off:
1: 1.88
2: 2.62
3: 3.27
4: 3.73
5: 4.07
thinking_on:
1: 1.81
2: 2.43
3: 2.90
4: 3.26
5: 3.51