pytorch - ✅(Solved) Fix PyTorch 2.10-2.11 Linux x86 Binary size increase investigation [1 pull requests, 3 comments, 3 participants]

pytorch2026-03-10 18:24:33

ON THIS PAGE

Recommended Tools

×6

Utilities matched from this issue’s tags and category — try them while you read without losing context.

GitHub issue graph ai analysis

Paste a GitHub issue URL. We fetch that issue, discover linked issues from bodies/comments/timeline, collect linked pull requests, and produce a structured English report.

The report is written in English Markdown for sharing and archival.

GitHub issue URL

Helpful · Quick feedback

GitHub stats

pytorch/pytorch#177050•Fetched 2026-04-08 00:22:31

View on GitHub

Comments

Participants

Timeline

Reactions

Author

Participants

Assignees

Timeline (top)

mentioned ×8subscribed ×8labeled ×4commented ×3

Fix Action

Fixed

Fixed by PR: Add claude skill to analyze wheel binary size increease (https://github.com/pytorch/test-infra/pull/7827)

PR fix notes

PR #7827: Add claude skill to analyze wheel binary size increease

Repository: pytorch/test-infra
Author: atalman
State: closed | merged: True
Link: https://github.com/pytorch/test-infra/pull/7827

Description (problem / solution / changelog)

This skill helps to track down wheel size . The output is similar to the following issue: https://github.com/pytorch/pytorch/issues/177050

Changed files

.claude/skills/wheel-size-analyzer/SKILL.md (added, +280/-0)

Code Example

┌────────┬─────────┬───────────────────────┐
  │  Date  │  Jump   │ Cumulative from Oct 1 │
  ├────────┼─────────┼───────────────────────┤
  │ Oct 11 │ +4.0 MB │ 859.4 MB              │
  ├────────┼─────────┼───────────────────────┤
  │ Oct 18 │ +2.3 MB │ 861.7 MB              │
  ├────────┼─────────┼───────────────────────┤
  │ Oct 29 │ +1.6 MB │ 863.3 MB              │
  ├────────┼─────────┼───────────────────────┤
  │ Oct 30 │ +3.8 MB │ 867.0 MB              │
  └────────┴─────────┴───────────────────────┘

---

October 2025 — Heavy Growth (+11.7 MB)

---

4 significant jumps: Oct 11 (+4.0), Oct 18 (+2.3), Oct 29 (+1.6), Oct 30 (+3.8)

  ---
  November 2025 — Volatile, Net Flat (+1.2 MB)

---

Key events: Spike +4.7 MB (Nov 4–5) then rollback -4.1 MB (Nov 12). Temporary
  cuda12_9 rename (Nov 15–17).

  December 2025 — Stable (+1.8 MB)

---

No days with >1 MB jump. Very stable month. Largest single-day: Dec 3 (+0.9 MB).

  ---



PyTorch Wheel Size Report: Jan–Mar 2026                                            
                                                                               
  manywheel-py3_12-cuda12_8 (x86)                                                    
                                                                                     
  ---                                                                                
  Quarter Overview

---

The wheel shrank by nearly 97 MB from Jan 1 to Mar 10 — the opposite of Q4 2025's
  growth trend.

  ---
  January 2026 — Massive Reduction (-90.6 MB)

---

Key events:
  - Jan 6 (-29.7 MB): Major dependency removal or build config change
  - Jan 17 (-67.1 MB): Largest single-day change — likely a significant
  library/dependency removal
  - Jan 25 (+4.9 MB): Only notable increase in the month

  ---
  February 2026 — Continued Shrinkage (-3.0 MB)

---

No days with >1 MB increase. Both significant changes were decreases. 50% of
  workflow runs had "failure" status but artifacts were still produced.

  ---
  March 2026 (1st–10th) — Stable, Slight Drop (-3.1 MB)

---

No days with >1 MB increase. One step-down on Mar 5.

  ---
  Full Timeline: Q4 2025 → Q1 2026

  Oct 2025:  855

RAW_BUFFERClick to expand / collapse

This issue is to document the

PyTorch Wheel Size Report: Q4 2025 (Oct–Dec) manywheel-py3_12-cuda12_8 (x86)

TLDR: Almost all of the Q4 growth (11.7 of 14.8 MB) happened in October, concentrated in just 4 days:

  ┌────────┬─────────┬───────────────────────┐
  │  Date  │  Jump   │ Cumulative from Oct 1 │
  ├────────┼─────────┼───────────────────────┤
  │ Oct 11 │ +4.0 MB │ 859.4 MB              │
  ├────────┼─────────┼───────────────────────┤
  │ Oct 18 │ +2.3 MB │ 861.7 MB              │
  ├────────┼─────────┼───────────────────────┤
  │ Oct 29 │ +1.6 MB │ 863.3 MB              │
  ├────────┼─────────┼───────────────────────┤
  │ Oct 30 │ +3.8 MB │ 867.0 MB              │
  └────────┴─────────┴───────────────────────┘

November saw a temporary spike (Nov 4–5, +4.7 MB) that was rolled back on Nov 12. December was essentially flat. The wheel ended Q4 at 870.1 MB, up from 855.3 MB at the start of October.

  │  Month   │ Start (MB) │ End (MB) │    Change     │ Significant Jumps (>1 MB) │   
  ├──────────┼────────────┼──────────┼───────────────┼───────────────────────────┤
  │ October  │ 855.3      │ 867.0    │ +11.7 (+1.4%) │ 4 days                    │
  ├──────────┼────────────┼──────────┼───────────────┼───────────────────────────┤
  │ November │ 867.1      │ 868.3    │ +1.2 (+0.1%)  │ 2 days up, 2 days down    │
  ├──────────┼────────────┼──────────┼───────────────┼───────────────────────────┤
  │ December │ 868.3      │ 870.1    │ +1.8 (+0.2%)  │ 0 days                    │
  ├──────────┼────────────┼──────────┼───────────────┼───────────────────────────┤
  │ Total Q4 │ 855.3      │ 870.1    │ +14.8 (+1.7%) │                           │
  └──────────┴────────────┴──────────┴───────────────┴───────────────────────────┘

October 2025 — Heavy Growth (+11.7 MB)

  ┌───────────┬─────────────┬───────┐
  │   Date    │  Size (MB)  │ Delta │
  ├───────────┼─────────────┼───────┤
  │ Oct 01    │       855.3 │     — │
  ├───────────┼─────────────┼───────┤
  │ Oct 02–10 │ 855.1–855.3 │ ~flat │
  ├───────────┼─────────────┼───────┤
  │ Oct 11    │       859.4 │  +4.0 │
  ├───────────┼─────────────┼───────┤
  │ Oct 12–17 │ 859.4–859.5 │ ~flat │
  ├───────────┼─────────────┼───────┤
  │ Oct 18    │       861.7 │  +2.3 │
  ├───────────┼─────────────┼───────┤
  │ Oct 19–28 │ 861.4–861.7 │ ~flat │
  ├───────────┼─────────────┼───────┤
  │ Oct 29    │       863.3 │  +1.6 │
  ├───────────┼─────────────┼───────┤
  │ Oct 30    │       867.0 │  +3.8 │
  ├───────────┼─────────────┼───────┤
  │ Oct 31    │       867.0 │  -0.0 │
  └───────────┴─────────────┴───────┘

4 significant jumps: Oct 11 (+4.0), Oct 18 (+2.3), Oct 29 (+1.6), Oct 30 (+3.8)

November 2025 — Volatile, Net Flat (+1.2 MB)

┌─────────┬─────────────┬─────────┬────────────────────────────────────────────┐
│  Date   │  Size (MB)  │  Delta  │                   Notes                    │
├─────────┼─────────────┼─────────┼────────────────────────────────────────────┤
│ Nov     │       867.1 │   ~flat │                                            │
│ 01–03   │             │         │                                            │
├─────────┼─────────────┼─────────┼────────────────────────────────────────────┤
│ Nov 04  │       868.3 │    +1.2 │                                            │
├─────────┼─────────────┼─────────┼────────────────────────────────────────────┤
│ Nov 05  │       871.8 │    +3.5 │ Largest single-day jump                    │
├─────────┼─────────────┼─────────┼────────────────────────────────────────────┤
│ Nov     │ 871.9–872.0 │   ~flat │ Peak: 872.0                                │
│ 06–11   │             │         │                                            │
├─────────┼─────────────┼─────────┼────────────────────────────────────────────┤
│ Nov 12  │       867.9 │    -4.1 │ Rollback                                   │
├─────────┼─────────────┼─────────┼────────────────────────────────────────────┤
│ Nov     │ 868.4–868.8 │ gradual │                                            │
│ 13–14   │             │         │                                            │
├─────────┼─────────────┼─────────┼────────────────────────────────────────────┤
│ Nov     │      ~1,211 │     N/A │ Artifact renamed to cuda12_9 (CUDA version │
│ 15–17   │             │         │  change, not code growth)                  │
├─────────┼─────────────┼─────────┼────────────────────────────────────────────┤
│ Nov     │       869.1 │       — │ cuda12_8 returns                           │
│ 18–19   │             │         │                                            │
├─────────┼─────────────┼─────────┼────────────────────────────────────────────┤
│ Nov 20  │       868.1 │    -1.0 │ Drop                                       │
├─────────┼─────────────┼─────────┼────────────────────────────────────────────┤
│ Nov     │       868.3 │   ~flat │                                            │
│ 21–24   │             │         │                                            │
├─────────┼─────────────┼─────────┼────────────────────────────────────────────┤
│ Nov     │           — │       — │ No runs (Thanksgiving week)                │
│ 25–30   │             │         │                                            │
└─────────┴─────────────┴─────────┴────────────────────────────────────────────┘

Key events: Spike +4.7 MB (Nov 4–5) then rollback -4.1 MB (Nov 12). Temporary cuda12_9 rename (Nov 15–17).

December 2025 — Stable (+1.8 MB)

│   Date    │  Size (MB)  │          Delta          │
├───────────┼─────────────┼─────────────────────────┤
│ Dec 01–02 │       868.3 │                   ~flat │
├───────────┼─────────────┼─────────────────────────┤
│ Dec 03    │       869.3 │                    +0.9 │
├───────────┼─────────────┼─────────────────────────┤
│ Dec 04–08 │ 869.3–869.5 │                   ~flat │
├───────────┼─────────────┼─────────────────────────┤
│ Dec 09    │           — │ Missing (build failure) │
├───────────┼─────────────┼─────────────────────────┤
│ Dec 10–12 │       870.2 │         +0.7 (vs Dec 8) │
├───────────┼─────────────┼─────────────────────────┤
│ Dec 13    │       869.8 │                    -0.4 │
├───────────┼─────────────┼─────────────────────────┤
│ Dec 14–24 │ 869.7–869.9 │                   ~flat │
├───────────┼─────────────┼─────────────────────────┤
│ Dec 25    │       870.1 │                    +0.4 │
├───────────┼─────────────┼─────────────────────────┤
│ Dec 26–31 │ 870.1–870.2 │                   ~flat │
└───────────┴─────────────┴─────────────────────────┘

No days with >1 MB jump. Very stable month. Largest single-day: Dec 3 (+0.9 MB).

PyTorch Wheel Size Report: Jan–Mar 2026

manywheel-py3_12-cuda12_8 (x86)

Quarter Overview

│   Month    │ Start (MB) │ End (MB) │     Change     │      Trend      │
├────────────┼────────────┼──────────┼────────────────┼─────────────────┤
│ January    │ 870.2      │ 779.6    │ -90.6 (-10.4%) │ Two large drops │
├────────────┼────────────┼──────────┼────────────────┼─────────────────┤
│ February   │ 779.6      │ 776.6    │ -3.0 (-0.4%)   │ Two small drops │
├────────────┼────────────┼──────────┼────────────────┼─────────────────┤
│ March 1–10 │ 776.6      │ 773.5    │ -3.1 (-0.4%)   │ One step down   │
├────────────┼────────────┼──────────┼────────────────┼─────────────────┤
│ Total      │ 870.2      │ 773.5    │ -96.7 (-11.1%) │                 │
└────────────┴────────────┴──────────┴────────────────┴─────────────────┘

The wheel shrank by nearly 97 MB from Jan 1 to Mar 10 — the opposite of Q4 2025's growth trend.

January 2026 — Massive Reduction (-90.6 MB)

│   Date    │  Size (MB)  │    Delta     │          Notes          │
├───────────┼─────────────┼──────────────┼─────────────────────────┤
│ Jan 01–05 │       870.2 │        ~flat │ Baseline from Dec 2025  │
├───────────┼─────────────┼──────────────┼─────────────────────────┤
│ Jan 06    │       840.5 │        -29.7 │ Large drop              │
├───────────┼─────────────┼──────────────┼─────────────────────────┤
│ Jan 07–16 │ 840.6–841.4 │        ~flat │                         │
├───────────┼─────────────┼──────────────┼─────────────────────────┤
│ Jan 17    │       774.4 │        -67.1 │ Largest single-day drop │
├───────────┼─────────────┼──────────────┼─────────────────────────┤
│ Jan 18–24 │ 774.5–775.3 │        ~flat │                         │
├───────────┼─────────────┼──────────────┼─────────────────────────┤
│ Jan 25    │       780.2 │         +4.9 │ Only >1 MB increase     │
├───────────┼─────────────┼──────────────┼─────────────────────────┤
│ Jan 26–31 │ 779.6–780.2 │ slight drift │                         │
└───────────┴─────────────┴──────────────┴─────────────────────────┘

Key events:

Jan 6 (-29.7 MB): Major dependency removal or build config change
Jan 17 (-67.1 MB): Largest single-day change — likely a significant library/dependency removal
Jan 25 (+4.9 MB): Only notable increase in the month

February 2026 — Continued Shrinkage (-3.0 MB)

  ┌───────────┬─────────────┬───────┬───────────┐
  │   Date    │  Size (MB)  │ Delta │   Notes   │
  ├───────────┼─────────────┼───────┼───────────┤
  │ Feb 01–05 │ 779.4–779.6 │ ~flat │           │
  ├───────────┼─────────────┼───────┼───────────┤
  │ Feb 06    │       777.6 │  -1.8 │ Step down │
  ├───────────┼─────────────┼───────┼───────────┤
  │ Feb 07–18 │ 777.6–778.2 │ ~flat │           │
  ├───────────┼─────────────┼───────┼───────────┤
  │ Feb 19    │       776.3 │  -1.6 │ Step down │
  ├───────────┼─────────────┼───────┼───────────┤
  │ Feb 20–28 │ 776.3–776.6 │ ~flat │           │
  └───────────┴─────────────┴───────┴───────────┘

No days with >1 MB increase. Both significant changes were decreases. 50% of workflow runs had "failure" status but artifacts were still produced.

March 2026 (1st–10th) — Stable, Slight Drop (-3.1 MB)

  ┌───────────┬─────────────┬───────┐
  │   Date    │  Size (MB)  │ Delta │
  ├───────────┼─────────────┼───────┤
  │ Mar 01–04 │ 776.6–776.7 │ ~flat │
  ├───────────┼─────────────┼───────┤
  │ Mar 05    │       774.1 │  -2.6 │
  ├───────────┼─────────────┼───────┤
  │ Mar 06–07 │       774.1 │ ~flat │
  ├───────────┼─────────────┼───────┤
  │ Mar 08    │       773.5 │  -0.6 │
  ├───────────┼─────────────┼───────┤
  │ Mar 09–10 │       773.5 │ ~flat │
  └───────────┴─────────────┴───────┘

No days with >1 MB increase. One step-down on Mar 5.

Full Timeline: Q4 2025 → Q1 2026

Oct 2025: 855

(+12)
Nov 2025:  867 ████████████████████████████████████████████████████████████████████
 868 MB    (+1)
Dec 2025:  868
█████████████████████████████████████████████████████████████████████ 870 MB   (+2)
Jan 2026:  780 ██████████████████████████████████████████████
  870 MB  (-91)
Feb 2026:  777 ████████████████████████████████████████████ 780 MB
          (-3)
Mar 2026:  774 ██████████████████████████████████████████ 777 MB
          (-3)

The wheel peaked at ~870 MB in early January 2026, then dropped sharply to ~774 MB by March 10 — a 96.7 MB (11.1%) reduction driven primarily by two large drops on Jan 6 (-29.7 MB) and Jan 17 (-67.1 MB).

cc @seemethere @malfet @tinglvv @nWEIdia

extent analysis

Problem Summary

The PyTorch wheel size report shows a significant reduction in size from Q4 2025 to Q1 2026, with a peak of ~870 MB in early January 2026 and a drop to ~774 MB by March 10.

Root Cause Analysis

The root cause of the issue is likely due to changes in dependencies or build configurations that led to the removal of significant libraries or dependencies, resulting in a reduction in wheel size.

Fix Plan

To fix the issue, we need to identify and address the changes that led to the reduction in wheel size. Here are the steps to follow:

Step 1: Review Build Configurations

Review the build configurations for Q4 2025 and Q1 2026 to identify any changes that may have led to the reduction in wheel size.

Step 2: Identify Removed Dependencies

Identify the dependencies that were removed between Q4 2025 and Q1 2026. This can be done by reviewing the build logs and dependency lists.

Step 3: Re-add Removed Dependencies

Re-add the removed dependencies to the build configurations to see if it resolves the issue.

Step 4: Test and Verify

Test the wheel with the re-added dependencies and verify that the size has increased to the expected level.

Step 5: Refine Build Configurations

Refine the build configurations to ensure that the dependencies are properly included and the wheel size is optimized.

Example Code

import setuptools

# Review build configurations
with open('setup.cfg', 'r') as f:
    config = f.read()

# Identify removed dependencies
removed_dependencies = []
for line in config.splitlines():
    if 'dependency' in line and 'removed' in line:
        removed_dependencies.append(line.split('=')[1].strip())

# Re-add removed dependencies
for dependency in removed_dependencies:
    config += f'\ndependency = {dependency}'

# Refine

Vote matrix · Quick signals

Works

Did the solution work? Tap to confirm.

Easy Fix

Was it a quick fix?

Time Saver

Did it save you time?

Blocking

Was it severely blocking?

Common Issue

Are others likely hitting this too?

Flaky / Intermittent

Is it intermittent?

Verified / Reproducible

Can you reproduce it reliably?

#api #ssr #installation #tensor shape #autograd error #docker error #permission error #memory optimization #batch processing #GPU compatibility

Still need to ship something?

×6

Another batch ranked right after the header list — different links, same matching logic.

Data

Security

Network

Code

UI/UX

Text

System

Multimedia

Protocol

API

Engineering

pytorch - ✅(Solved) Fix PyTorch 2.10-2.11 Linux x86 Binary size increase investigation [1 pull requests, 3 comments, 3 participants]

Recommended Tools

GitHub issue graph ai analysis

Fix Action

Fixed

PR fix notes

PR #7827: Add claude skill to analyze wheel binary size increease

Description (problem / solution / changelog)

Changed files

Code Example

extent analysis

Problem Summary

Root Cause Analysis

Fix Plan

Step 1: Review Build Configurations

Step 2: Identify Removed Dependencies

Step 3: Re-add Removed Dependencies

Step 4: Test and Verify

Step 5: Refine Build Configurations

Example Code

Still need to ship something?

TRENDING

pytorch - ✅(Solved) Fix PyTorch 2.10-2.11 Linux x86 Binary size increase investigation [1 pull requests, 3 comments, 3 participants]

Recommended Tools

GitHub issue graph ai analysis

Fix Action

Fixed

PR fix notes

PR #7827: Add claude skill to analyze wheel binary size increease

Description (problem / solution / changelog)

Changed files

Code Example

extent analysis

Problem Summary

Root Cause Analysis

Fix Plan

Step 1: Review Build Configurations

Step 2: Identify Removed Dependencies

Step 3: Re-add Removed Dependencies

Step 4: Test and Verify

Step 5: Refine Build Configurations

Example Code

Still need to ship something?

RELATED_DISCOVERY

TRENDING