Resuming CM dev meetings (#1277)

ck/ck/kernel.py

@@ -28,7 +28,7 @@
 
 
 # We use 3 digits for the main (released) version and 4th digit for development revision
-__version__ = "2.6.3"
+__version__ = "2.6.4"
 # Do not use characters (to detect outdated version)!
 
 # Import packages that are global for the whole kernel
@@ -6745,9 +6745,8 @@ def short_help(i):
     # .replace('   ','')+'\n'
     h += '\n'+cfg['help_web'].replace('\n', '').strip()+'\n'
 
-    h += 'CK Google group:      https://bit.ly/ck-google-group\n'
-    h += 'CK Slack channel:     https://cKnowledge.org/join-slack\n'
-    h += 'Stable CK components: https://cknow.io'
+    h += 'CK white paper:       https://royalsocietypublishing.org/doi/10.1098/rsta.2020.0211\n'
+    h += 'CK ACM TechTalk:      https://learning.acm.org/techtalks/reproducibility\n'
 
     if o == 'con':
         out(h)
@@ -12385,11 +12384,11 @@ def access(i):
         o = i.get('out', '')
 
         # Print message that this framework was discontinued
-        if o == 'con':
-            out('')
-            out('WARNING: this framework was discontinued in favor of the new CK2 framework aka CM being developed by the open taskforce on automation and reproducibility at MLCommons:')
-            out('         https://bit.ly/mlperf-edu-wg')
-            out('')
+#        if o == 'con':
+#            out('')
+#            out('WARNING: this framework was discontinued in favor of the new CK2 framework aka CM being developed by the open taskforce on automation and reproducibility at MLCommons:')
+#            out('         https://bit.ly/mlperf-edu-wg')
+#            out('')
 
         # If profile
         cp = i.get('ck_profile', '')

dev/common-paper/README.md

@@ -0,0 +1 @@
+TBD

dev/meetings/20240731.md

@@ -0,0 +1,118 @@
+# Topic
+
+Syncing about the next steps for CM, CM4MLOps, CM4MLPerf, CM4ABTF, etc.
+
+# People
+
+* Grigori Fursin
+* Arjun Suresh
+
+# Discussion
+
+## Need proper attribution
+
+Need to agree on the common text in the CM documentation and CM, CM4MLOps and CM4MLPerf GitHub
+using these examples:
+
+* https://github.com/spack/spack?tab=readme-ov-file#authors
+* https://cython.org (see Financial contributions section)
+
+* Author/creator
+* Core developers
+* Contributors (from the community, MLCommons and the Automation and Reproducibility TaskForce):
+  See https://github.com/mlcommons/ck/blob/master/CONTRIBUTING.md .
+* Sponsorship & financial Contributions
+
+Should add to main GitHub and docs.mlcommons.org ...
+
+## Remove/reduce dependencies on non-MLCommons GitHub repositories
+
+At this moment, various non-MLCommons GATEOverflow GitHub repositories are used 
+in the official MLPerf workflows by default - that creates many possible legal issues
+for CM and MLPerf users.
+
+We should either move all such repositories to MLCommons 
+or, if it's not easily possible, create another neutral GitHub ID 
+such as mlcommons-aux with a clear governances and agreement
+with MLCommons to keep all dependencies in the MLCommons space.
+
+## Improve cm4mlops package
+
+Current cm4mlops package hides extra installation of various system dependencies 
+and CM repositories while using non-default branches and is difficult to debug if something goes wrong.
+
+A most standard way is to install cmind package and have a function to bootstrap cm4mlops
+with a proper control over the flow, CM repositories and branches that can be changed via flags.
+
+For example:
+```bash
+pip install cmind
+cm bootstrap cm4mlperf
+cm bootstrap cm4mlops --branch=mlperf-inference
+...
+```
+
+That will perform the same functions as cm4mlops package but will be easy to debug and will have easy to trace errors
+that can be used in GitHub actions or other CI
+
+To be brainstormed further ...
+
+## Coordinate further developments
+
+* Document the roadmap and responsibilities for Q3-Q4 2024
+* Regular dev meetings (once a week or every two weeks)?
+* Resume Discord channel discussions or mailing list (to be able to track discussions)?
+
+## CM4MLPerf inference v4.1 & v5.0 automation
+
+* Add CM for as many v4.1 submissions as possible to make it easier for everyone to reproduce results shortly after publication of results.
+* Sync on the plans for inf v5.0 with MLCommons
+
+## CM4ABTF automation
+
+* Sync on the next steps during next meetings
+
+## Collaboration with Croissant
+
+* Sync on the next steps during next meetings
+
+## Testing infrastructure for CM4MLOps and CM4MLPerf
+
+* GitHub actions are not enough to test all dependencies and their versions for diverse hardware for CM-MLPerf workflows.
+  Brainstorm infrastructure for continuous testing (Grigori started prototyping some infrastructure).
+
+## Optimize MLPerf inference reference implementations
+
+* We need to add known optimizations to the MLPerf inference implementations
+
+## Support MLPerf training
+
+* We should start prototyping the unified CM interface and automation for MLPerf training and wrap existing MLCube tasks
+
+## Prepare tutorials
+
+* Sync on the tutorial about CM internals and scripts
+* Sync on the tutorial for SCC'24
+
+## Common paper
+
+* Start preparing a common paper about CM on GitHub 
+
+## Collect feedback from companies
+
+* There were various discussions with MLCommons companies about using CM for reproducibility.
+  We need to collect and aggregate all the feedback in one place.
+
+## Next generation of CM
+
+Grigori started testing some ideas and prototyping the next generation of CM, CM4MLOps and CM4MLPerf 
+bsaed on 3 years of using CM to modularize and automate MLPerf and will share notes in the future dev meetings.
+
+## Sync with MLCommons
+
+* Prepare official CM page - should we do it with the MLPerf in v4.1 release?
+* Prepare Press-release about CM with MLPerf inf v4.1 release?
+* Where to host CM developments and discussions within MLCommons?
+  * Infra WG?
+  * Create a new *official* taskforce or WG on automation and reproducibility?
+

dev/reproducibility/README.md

@@ -0,0 +1 @@
+TBD