Skip to content

Allow Mooncake 0.6#1436

Merged
yebai merged 1 commit into
mainfrom
support-mooncake-0.6
Jul 14, 2026
Merged

Allow Mooncake 0.6#1436
yebai merged 1 commit into
mainfrom
support-mooncake-0.6

Conversation

@yebai

@yebai yebai commented Jul 14, 2026

Copy link
Copy Markdown
Member

No description provided.

Mooncake 0.6 is a breaking release (forward-mode redesign); the reverse-mode
rule API used by DynamicPPLMooncakeExt (`@zero_derivative`, `DefaultCtx`) is
unchanged, so the extension works as-is. Widen the `Mooncake` compat bound to
add `0.6` (package and test env) and bump the patch version.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01SRMQzvS6gE7QGen3U7wURH
@github-actions

Copy link
Copy Markdown
Contributor

DynamicPPL.jl documentation for PR #1436 is available at:
https://TuringLang.github.io/DynamicPPL.jl/previews/PR1436/

@codecov

codecov Bot commented Jul 14, 2026

Copy link
Copy Markdown

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 82.20%. Comparing base (5989b33) to head (e0f3948).

Additional details and impacted files
@@            Coverage Diff             @@
##             main    #1436      +/-   ##
==========================================
+ Coverage   81.64%   82.20%   +0.56%     
==========================================
  Files          50       50              
  Lines        3579     3552      -27     
==========================================
- Hits         2922     2920       -2     
+ Misses        657      632      -25     

☔ View full report in Codecov by Harness.
📢 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.

@github-actions

Copy link
Copy Markdown
Contributor

Benchmarks @ e0f3948

Performance Ratio: gradient time divided by log-density time.

For very small models these ratios are noisy across runs and machines; raw primal and gradient timings are more reliable. The benchmarks are aimed at DynamicPPL developers and mainly catch obvious allocation or type-stability regressions. See benchmark notes for details.

===================================================================================================
                                               eval                       gradient                 
                                            ----------  -------------------------------------------
Model                        dim    linked      primal     FwdDiff    RvsDiff    Mooncake    Enzyme
---------------------------------------------------------------------------------------------------
Simple assume observe*         1     false     4.63 ns       12.45    1505.94       39.72     12.31
Simple assume observe*         1      true     4.63 ns       12.64    1664.87       39.81     12.20
Smorgasbord                  201     false     6.02 μs       71.42     137.82        6.94      9.62
Smorgasbord                  201      true     7.62 μs       81.04     140.39        6.32      6.86
Loop univariate 1k          1000     false     17.6 μs     1039.56     309.13        8.23      6.75
Loop univariate 1k          1000      true     19.1 μs     1469.79     285.53        7.75      6.27
Multivariate 1k             1000     false     22.2 μs      370.34      78.38        9.18      2.97
Multivariate 1k             1000      true     21.4 μs      278.62      59.05       10.65      2.87
Loop univariate 10k        10000     false    172.0 μs    12209.16     334.75        8.00      6.43
Loop univariate 10k        10000      true    198.0 μs    11612.78     292.06        7.34      5.97
Multivariate 10k           10000     false    209.0 μs     5845.93      85.55       10.91      2.23
Multivariate 10k           10000      true    197.0 μs     5057.13      86.51       11.64      2.30
Dynamic                       15     false     1.37 μs         err      46.73       15.31     11.25
Dynamic                       10      true     1.91 μs        1.99      55.99       12.58     18.59
Submodel*                      1     false     4.63 ns       12.49    1687.77       40.23     12.38
Submodel*                      1      true     4.63 ns       12.69    1868.32       40.08     12.36
LDA                           12      true     23.2 μs        0.50       1.96       33.53       err
===================================================================================================
Main @ 5989b33
===================================================================================================
                                               eval                       gradient                 
                                            ----------  -------------------------------------------
Model                        dim    linked      primal     FwdDiff    RvsDiff    Mooncake    Enzyme
---------------------------------------------------------------------------------------------------
Simple assume observe*         1     false     4.63 ns       13.67    1565.58       40.02     12.44
Simple assume observe*         1      true     4.63 ns       13.65    1707.31       40.30     12.46
Smorgasbord                  201     false     5.98 μs       72.99     139.67        6.98      9.63
Smorgasbord                  201      true     7.62 μs       75.45     148.27        6.44      6.97
Loop univariate 1k          1000     false     17.8 μs      968.92     283.97        8.10      6.48
Loop univariate 1k          1000      true     19.2 μs     1462.72     281.52        7.57      6.06
Multivariate 1k             1000     false     23.1 μs      403.47      75.78        8.73      2.97
Multivariate 1k             1000      true     31.8 μs      309.00      51.24        7.11      3.16
Loop univariate 10k        10000     false    174.0 μs    12856.86     327.53        8.43      7.09
Loop univariate 10k        10000      true    187.0 μs    12392.02     308.26        8.00      6.15
Multivariate 10k           10000     false    202.0 μs     5179.92      87.40       11.48      2.37
Multivariate 10k           10000      true    198.0 μs     5056.18      92.08       11.75      2.31
Dynamic                       15     false     1.41 μs         err      49.45       15.84     11.51
Dynamic                       10      true     1.94 μs        1.88      56.79       18.67     19.12
Submodel*                      1     false     4.63 ns       13.75    1699.44       40.23     12.21
Submodel*                      1      true     4.63 ns       12.57    1805.77       40.40     12.33
LDA                           12      true     23.0 μs        0.59       2.13       30.08       err
===================================================================================================
Environment
Julia Version 1.11.9
Commit 53a02c0720c (2026-02-06 00:27 UTC)
Build Info:
  Official https://julialang.org/ release
Platform Info:
  OS: Linux (x86_64-linux-gnu)
  CPU: 4 × AMD EPYC 7763 64-Core Processor
  WORD_SIZE: 64
  LLVM: libLLVM-16.0.6 (ORCJIT, znver3)
Threads: 1 default, 0 interactive, 1 GC (on 4 virtual cores)

@yebai yebai merged commit a49099c into main Jul 14, 2026
25 checks passed
@yebai yebai deleted the support-mooncake-0.6 branch July 14, 2026 22:21
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.

1 participant