fix: normalize torch version to >=2.1.0 across all dependency files#37
Open
spoturno wants to merge 1 commit into
Open
fix: normalize torch version to >=2.1.0 across all dependency files#37spoturno wants to merge 1 commit into
spoturno wants to merge 1 commit into
Conversation
pyproject.toml pinned torch==2.11.0 (an invalid version), requirements.txt used >=2.1.0, and training/requirements.txt used >=2.11.0. Aligned all three to >=2.1.0 for consistency.
amittell
added a commit
to amittell/OpenMythos
that referenced
this pull request
Jun 6, 2026
… + sft_lora (PR kyegomez#37) Three of the four GB10 cluster nodes have been sitting idle while 11 backfill runs (inference_benchmark + sft_lora) wait on Blackwell. The new Phase A auto-place feature from gpufarm PR kyegomez#37 substitutes resource_class: [auto] with the configured priority [blackwell, gb10, rtx5090, strix_halo] at load time so these jobs can fall back to GB10 when Blackwell is busy. inference_benchmark runs 5 min and was previously blackwell-only; sft_lora needs ~24 GB VRAM so any modern GPU fits. The other Blackwell-only jobs (consolidate_single_host, training_round_joint, etc.) keep their explicit declarations because they depend on Blackwell-specific behaviour or VRAM.
amittell
added a commit
to amittell/OpenMythos
that referenced
this pull request
Jun 6, 2026
The 11 queued backfill runs (inference_benchmark + sft_lora) declare resource_class: [auto] which substitutes to [blackwell, gb10, rtx5090, strix_halo] per PR kyegomez#37. With Blackwell occupied by the Qwen campaign, they should fall through to GB10. But cluster-4node's can_run list only includes training/eval_4node jobs -- single-node compute jobs couldn't dispatch there. Adds individual node resources gb10-spark, gb10-gx10, gb10-gx10-3 with can_run lists that include the standard single-node eval bundle. They overlap cluster-4node on hardware -- operator-mutex pattern, same as blackwell-host vs blackwell-gpu0/1. gx10-2 is deliberately omitted because it currently hosts the Qwen 35B-A3B-FP8 BCB endpoint (~92 GB unified mem); add it back once that completes.
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
1 participant
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.
Problem
The torch version requirement is inconsistent across the three dependency files:
pyproject.tomltorch = "2.11.0"(exact pin to a non-existent version)requirements.txttorch>=2.1.0training/requirements.txttorch>=2.11.0(also non-existent)torch 2.11.0does not exist on PyPI — the latest 2.x releases follow2.1.x,2.2.x,2.3.x, etc. This causes install failures for anyone usingpyproject.tomlor the training requirements.Fix
Aligned all three files to
torch>=2.1.0, matching the existingrequirements.txtconvention.