Repository navigation

Commits

Commits on Nov 10, 2026
- [fix] change template support for thinking mode fusion
  tomiock
  committed
  3c0606e
Commits on Nov 5, 2026
- Merge pull request #38 from VLR-CVC/feat/qwen35-moe-slim
  tomiock
  authored
  1e37fb0
- [fix] jup_35b_2node_check.sh: no WSD decay, so 50 steps match the 2600-step baseline LR
  
  tomiock
  and
  claude
  committed
  bb1a29a
- [fix] JUPITER scripts: torchtitan venv_214 on top of torch_main_214
  
  tomiock
  and
  claude
  committed
  dbabfd0
- [chore] JUPITER scripts: default env torch_main_214
  
  tomiock
  and
  claude
  committed
  0133fc9
Commits on Nov 4, 2026
- [chore] scripts/jup_35b_2node_check.sh: 35B-A3B two-node check on JUPITER
  
  tomiock
  and
  claude
  committed
  c495973
- [docs] README, USAGE, INSTALL: MoE / EP / MTP, MN5 env, stale pointers
  
  tomiock
  and
  claude
  committed
  4c4cc57
- [docs] SCALABILITY.md: Qwen3.5-2B and 35B-A3B at 64 nodes, 2B at 128K
  
  tomiock
  and
  claude
  committed
  3baf165
- [refactor] move step logging out of train_qwen.py into train/metrics.py
  
  tomiock
  and
  claude
  committed
  f86637b
- [refactor] clearer names; load balancing and image cap as config options
  
  tomiock
  and
  claude
  committed
  64fd2fa
- [chore] trim comments; drop unused scaling scripts, model configs and test images
  
  tomiock
  and
  claude
  committed
  d9eb1e5
- [chore] prune tests that are stale, duplicated or manual-only
  
  tomiock
  and
  claude
  committed
  8b75407
- [chore] remove unused scaling and ttbench scripts
  
  tomiock
  and
  claude
  committed
  f804532
- [chore] drop trace_dir/view_trace.sh
  
  tomiock
  and
  claude
  committed
  9a34cc4
- [chore] local 2B config trains on finevision_energon
  tomiock
  committed
  749beb8
- [feat] training configs: MN5 updates and a text-only 2B
  
  tomiock
  and
  claude
  committed
  9599833
- [feat] cluster launchers and env setup
  
  tomiock
  and
  claude
  committed
  c6fdec6
- [feat] checkpoint export: instruct finalization, Hub sharding
  tomiock
  committed
  1b85b57
- [feat] dataset preparation tools
  tomiock
  committed
  37f22d6
- [feat] Multi-Token Prediction for Qwen3.5
  tomiock
  committed
  c9777f4
- [feat] Qwen3.5 MoE with expert parallelism; training QoL; new data cookers
  tomiock
  committed
  ec22f71
Commits on Oct 2, 2026
- Merge pull request #35 from VLR-CVC/titan-migration
  tomiock
  authored
  d289993
- [feat] new model denifitions
  tomiock
  committed
  11efaf9
Commits on Sep 26, 2026
- [feat] utils clean up and dataset indexing
  tomiock
  committed
  c4b6135
- [fix] template missmatch on cooker for OV-Instruct
  tomiock
  committed
  5799c09
Commits on Sep 20, 2026
- Merge pull request #34 from VLR-CVC/titan-migration
  tomiock
  authored
  5cd89a7
- [fix/feat] docs and compile
  tomiock
  committed
  de51baa
- [feat] final port
  tomiock
  committed
  eb18a8d
Commits on Sep 18, 2026
- new logs
  tomiock
  committed
  c39d1a4
Commits on Sep 17, 2026
- [fix] the great purge
  tomiock
  committed
  e366460
- [feat] scaling configs
  tomiock
  committed
  79f987d
- [feat] torchtitan model definitions and parallelism
  tomiock
  committed
  096edde
Commits on Sep 14, 2026
- fix(mn5): let QWEN_SECTION_TIMING be overridden per job
  
  tomiock
  and
  claude
  committed
  c730ae9
- fix(mn5): WANDB_MODE=offline in the bench launcher
  
  tomiock
  and
  claude
  committed
  51600af
- feat(mn5): Qwen3-VL-2B config sweep
  
  tomiock
  and
  claude
  committed
  1183704

Branch selector

User selector

Datepicker

Commit history

Commits on Nov 10, 2026

Commits on Nov 5, 2026

Commits on Nov 4, 2026

Commits on Oct 2, 2026

Commits on Sep 26, 2026

Commits on Sep 20, 2026

Commits on Sep 18, 2026

Commits on Sep 17, 2026

Commits on Sep 14, 2026